DAS23CoverCope 2/28/14 1:22 PM Page 1

VOLUME 23 THE COMPUTER MUSIC AND DIGITAL AUDIO SERIES

VOLUME 23 THE COMPUTER MUSIC AND DIGITAL AUDIO SERIES

Hidden Structure
Hidden Structure

Hidden Structure
Music Analysis
Using Computers Music Analysis
David Cope
Today’s computers provide music theorists with unprecedented opportunities to analyze
Using Computers
music more quickly and accurately than ever before. Where analysis once required several
weeks or even months to complete—often replete with human errors, computers now pro-
vide the means to accomplish these same analyses in a fraction of the time and with far

Music Analysis Using Computers

more accuracy. However, while such computer music analyses represent significant improve-
ments in the field, computational analyses using traditional approaches by themselves do
not constitute the true innovations in music theory that computers offer. In Hidden Structure:
Music Analysis Using Computers David Cope introduces a series of analytical processes
that—by virtue of their concept and design—can be better, and in some cases, only accom-
plished by computer programs, thereby presenting unique opportunities for music theorists to
understand more thoroughly the various kinds of music they study.
Following the introductory chapter that covers several important premises, Hidden Struc-
ture focuses on several unique approaches to music analysis offered by computer programs.
While these unique approaches do not represent an all-encompassing and integrated
global theory of music analysis, they do represent significantly more than a compilation of
loosely related computer program descriptions. For example, Chapter 5 on function in post-
tonal music, firmly depends on the scalar foundations presented in chapter 4. Likewise, chap-
ter 7 presents a multi-tiered approach to musical analysis that builds on the material found
in all of the preceding chapters. In short, Hidden Structure uniquely offers an integrated view
of computer music analysis for today’s musicians.

Of Related Interest
The Algorithmic Composer
By David Cope
DAS 16 ISBN 978-0-89579-454-3 (2000) xiii + 130 pp. (with CD-ROM) $49.95

Hyperimprovisation: Computer-Interactive Sound Improvisation

By Roger Dean
DAS 19 ISBN 978-0-89579-508-3 (2003) xxvi + 206 pp. (with CD-ROM) $49.95

New Digital Musical Instruments: Control and Interaction Beyond the Keyboard
By Eduardo Reck Miranda and Marcelo M. Wanderley
DAS 21 ISBN 978-0-89579-585-4 (2006) xxii + 286 pp. (with CD-ROM) $49.95

David Cope
I S B N 978-0-89579-640-0
A-R Editions, Inc. 90000
8551 Research Way, Suite 180
Middleton, WI 53562
800-736-0070
608-836-9000
http://www.areditions.com 9 780895 796400

Í David Cope
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page i

HIDDEN STRUCTURE
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page ii

THE COMPUTER MUSIC AND DIGITAL AUDIO SERIES

John Strawn, Founding Editor
James Zychowicz, Series Editor

Digital Audio Signal Processing Knowledge-Based Programming for

Edited by John Strawn Music Research
John W. Schaffer and Deron McGee
Composers and the Computer
Edited by Curtis Roads Fundamentals of Digital Audio
Alan P. Kefauver
Digital Audio Engineering
Edited by John Strawn The Digital Audio Music List: A Critical
Guide to Listening
Computer Applications in Music: Howard W. Ferstler
A Bibliography
Deta S. Davis The Algorithmic Composer
David Cope
The Compact Disc Handbook
Ken C. Pohlman The Audio Recording Handbook
Alan P. Kefauver
Computers and Musical Style
David Cope Cooking with Csound
Part I: Woodwind and Brass Recipes
MIDI: A Comprehensive Introduction Andrew Horner and Lydia Ayers
Joseph Rothstein
William Eldridge, Volume Editor Hyperimprovisation: Computer-
Interactive Sound Improvisation
Synthesizer Performance and Roger T. Dean
Real-Time Techniques
Jeff Pressing Introduction to Audio
Chris Meyer, Volume Editor Peter Utz

Music Processing New Digital Musical Instruments:

Edited by Goffredo Haus Control and Interaction Beyond
the Keyboard
Computer Applications in Music: Eduardo R. Miranda and
A Bibliography, Supplement I Marcelo M. Wanderley, with a
Deta S. Davis Foreword by Ross Kirk
Garrett Bowles, Volume Editor
Fundamentals of Digital Audio
General MIDI New Edition
Stanley Jungleib Alan P. Kefauver and David Patschke

Experiments in Musical Intelligence Hidden Structure: Music Analysis

David Cope Using Computers
David Cope
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page iii

Volume 23 • THE COMPUTER MUSIC AND DIGITAL AUDIO SERIES

HIDDEN STRUCTURE

Music Analysis Using Computers

David Cope

Í A-R Editions, Inc.

Middleton, Wisconsin
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page iv

Library of Congress Cataloging-in-Publication Data

Cope, David, 1941–

Hidden structure : music analysis using computers / David Cope.
p. cm. — (Computer music and digital audio series : v. 23)
Includes bibliographical references and index.
ISBN 978-0-89579-640-0
1. Music—Data processing. 2. Musical analysis—Data processing.
I. Title.

ML74.C69 2008
781.0285—dc22
2008019352
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page v

Pulchritudo est Splendor Ordinis.

Beauty is the splendor of order.
Saint Augustine (345–430)
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page vi
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page vii

Contents

List of Figures xi

Preface xxi
Description of CD-ROM xxvii
Chapter One Background 1
Principles and Definitions 1
A Brief History of Algorithmic Analysis 7
A Brief Survey of Computational Music Analysis 22
Musical Examples 39
Program Description 41
Conclusions 43

Chapter Two Lisp, Algorithmic Information Theory, 45

and Music
Lisp 45
Definitions 50
Musical Algorithmic Information Theory 62
Musical Examples 74
Program Description 90
Conclusions 97
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page viii

Chapter Three Register and Range in Set Analysis 99

Basics of Set Theory 99
Register 112
Ranges and Vectors 118
Comparisons 120
Musical Examples 127
Program Description 139
Conclusions 144

Chapter Four Computer Analysis of Scales in 145

Post-Tonal Music
Mathematical Sequences 145
Scales 148
Vector Classes and Metaclasses 153
Varèse’s Density 21.5 160
Schoenberg’s Six Little Piano Pieces, op. 19, no. 1 (1911) 169
Other Musical Examples 172
Program Description 182
Conclusions 188

Chapter Five Function and Structure in Post-Tonal 189

Music
Object-Oriented Programming 189
Definitions 197
The Acoustic Theory of Chord Roots 201
Musical Tension 205
Context and SPEAC 210
Function 215
Form and Structure 217
Musical Examples 225
Program Description 225
Conclusions 228
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page ix

Chapter Six Generative Models of Music 231

Modeling 231
Recombinancy 234
Probabilities 249
Rules and Markov Chains 252
Musical Examples 269
Program Description 269
Conclusions 274

Chapter Seven A Look to the Future 275

Principles 275
Mathematics 277
Artificial Intelligence 290
Muse 295
Musical Examples 305
The Future 310
Conclusions 316

Bibliography 321
Glossary 331
Index 337
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page x
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page xi

List of Figures

Chapter 1
Figure 1.1 From Aristoxenus, The Harmonics of Aristoxenus, trans.. Henry S.
Macran (New York: Oxford University Press, 1902), 249.
Figure 1.2 A woodcut of Boethius with an instrument designed for methodologi-
cally deriving tunings (an algorithm).
Figure 1.3 An example of Gregorian chant, Parce Domine, notated in neumes; the
second staff is a continuation of the first.
Figure 1.4 An example of medieval organum.
Figure 1.5 The Guidonian hand, in which a specific note is assigned to each part
of the hand. By pointing to a part of the hand, the conductor can indi-
cate to a group of singers which note to sing.
Figure 1.6 An example of three-voice counterpoint around the time of Tinctoris.
Figure 1.7 The division of the octave according to Zarlino.
Figure 1.8 An exercise by Beethoven for his student Archduke Rudolph based on
Fux’s Gradus ad Parnassum.
Figure 1.9 Two diagrams from Rameau’s Treatise on Harmony describing (a) the
major (perfect) triad and its inversions, and (b) the dominant (funda-
mental) seventh chord and its inversions.
Figure 1.10 First published after Mozart’s death in 1793 by J. J. Hummel in Berlin,
this musikalisches Würfelspiel instruction page shows the first two
phrases of number selections (chosen by throws of the dice) that were
keyed to measures of music. Also shown, an example page of music by
Haydn for this own musikalisches Würfelspiel.
Figure 1.11 A sample Schenker analysis of Bach’s organ prelude Wenn wir in höch-
sten Noten sein.
Figure 1.12 Charles Babbage’s first Difference Machine (1833), a forerunner of the
modern-day computer.
Figure 1.13 The author, Horizons for Orchestra, in graphic format. Time moves left
to right.

xi
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page xii

xii HIDDEN STRUCTURE

Figure 1.14 From Experiment 2 from Hiller and Isaacson, Illiac Suite, using the
Illiac computer at the University of Illinois, ca. 1956.
Figure 1.15 Flowcharts for MUSANA and its analysis module.
Figure 1.16 Ten commands found in Humdrum.
Figure 1.17 Bach Chorale 002 from the Music Theory Workbench, version 0.1
(http://pinhead.music.uiuc.edu/~hkt/mtw/pdf/), by Heinrich Taube
(accessed 24 July 2007).
Figure 1.18 An example of the OpenMusic visual programming environment of mu-
sic for composition and analysis.
Figure 1.19 The first few measures of Stravinsky, Three Pieces for String Quartet
(1914), 3rd movement.
Figure 1.20 Two screens from the Sets and Vectors program on the CD-ROM that
accompanies this book.
Chapter 2
Figure 2.1 The first 2000 digits of π.
Figure 2.2 Morse’s code for English letters and numbers.
Figure 2.3 An example of the kind of three-dimensional modeling of musical style
by Böker-Heil.
Figure 2.4 Cellular automata output generated by the code in the text.
Figure 2.5 Examples of retrograde (b), inversion (c), and retrograde inversion (d)
of a motive (a).
Figure 2.6 Two matrixes of the four-note motive of Figure 2.5. The first matrix
provides the motive and its eleven transpositions (down from right
to left) along with its retrograde and its eleven transpositions (read
right to left). The second matrix provides the inversion and its eleven
transpositions along with its retrograde inversion and its eleven
transpositions.
Figure 2.7 A simple downward scale beginning on different pitches, thus produc-
ing different arrangements of whole and half steps.
Figure 2.8 Brief monophonic samplings of music by (a) J. S. Bach, Suite no. 1 in G
Major for Violoncello Solo (1720), Minuet, mm. 1–8; (b) W. A. Mozart,
Symphony no. 40 in G Minor, K. 550 (1788), 1st movement, mm. 1–5;
(c) Ludwig van Beethoven, Symphony no. 5, op. 67 (1808), 1st move-
ment, mm. 1–5; (d), Johannes Brahms, Symphony no. 1, op. 68 (1876),
4th movement, mm. 30–38; (e) Anton Webern, Variations for Piano, op.
27 (1936), 2nd movement, mm. 1– 4; (f) Ernst Krenek, Suite for Violon-
cello, op. 84 (1939), 1st movement, mm. 1–4; (g) The author, Three
Pieces for Solo Clarinet(1965), 1st movement, mm. 1–3.
Figure 2.9 Bartók, Mikrokosmos, no. 81.
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page xiii

LIST OF FIGURES xiii

Figure 2.10 AIT structural analysis of the music presented in Figure 2.9.
Figure 2.11 A simple graph showing four parameters of a melody by Brahms (pre-
sented at the top of the figure).
Figure 2.12 Pitches reduced to intervals for pattern matching.
Figure 2.13 Bartók, Mikrokosmos, no. 81, from Figure 2.9, in a Dynamic Musical AIT
graph.
Figure 2.14 Multigraph output for the beginning of Stravinsky, Three Pieces for
String Quartet (1914), shown in Figure 1.19.
Figure 2.15 Two mazurkas in the style of Chopin for analytical comparison.
Figure 2.16 Two DMAIT graphs representing the Chopin mazurkas shown in Figure
2.15.
Figure 2.17 A prelude by Chopin comparable in length to the pieces in Figure 2.15.
Figure 2.18 A Dynamic Musical Algorithmic Information Theory graph of the Cho-
pin prelude shown in Figure 2.17.
Figure 2.19 Sample Multigraph output.

Chapter 3
Figure 3.1 Symbols representing several ways in which sets can be more formally
compared.
Figure 3.2 A Venn graphic of set theory logical deductions.
Figure 3.3 Root positions of triads have a smaller range between their outer notes.
Figure 3.4 A clock face arranged to allow clearer visualization of pitch-class
relationships.
Figure 3.5 The [1,6,t] pitch-class set does not resolve to the pitch-class set
[0,5,9] (9 distance when rotated to 0) because of the smaller range of
[6,t,1] (7 distance and equating to [0,4,7] when rotated to 0).
Figure 3.6 A straightforward example of the use of set theory to find similarities
in post-tonal music.
Figure 3.7 (a) [e,0,2,5,7,8] is [0,1,3,6,8,9] when rotated one number clockwise;
(b) [6,7,9,0,2,3] is [0,1,3,6,8,9] when rotated six numbers clockwise;
(c) [7,6,4,1,e,t] is [0,1,3,6,8,9] when rotated seven numbers counter-
clockwise; (d) [0,e,9,6,4,3] is [0,1,3,6,8,9] when rotated twelve num-
bers counterclockwise.
Figure 3.8 The opening 4 measures of Schoenberg, Six Little Piano Pieces, Op. 19,
no. 6.
Figure 3.9 The opening measures of Claude Debussy, La cathédrale engloutie
(The Sunken Cathedral), 1909.
Figure 3.10 A conversion chart relating decimal to duodecimal notation.
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page xiv

xiv HIDDEN STRUCTURE

Figure 3.11 A more understandable and musical way to think in terms of duodeci-
mal notation.
Figure 3.12 Conversion methods used for decimal to duodecimal numbers and
vice versa.
Figure 3.13 Subtraction, addition, multiplication, and division using arbitrary
numbers in duodecimal notation.
Figure 3.14 Two two-set progressions showing pitch classes (first), t-normal form
(second), and prime (third).
Figure 3.15 Same reduction as Figure 3.14 but with registers intact in the form of
duodecimal notation.
Figure 3.16 Range notation shown to the lower right.
Figure 3.17 Arnold Schoenberg, Suite for Piano, op. 25 (1923), Gavotte, mm. 1–6,
with groupings circled and numbered.
Figure 3.18 Computer analysis of Schoenberg‘s Gavotte by tetrachords and tri-
chords as shown circled and numbered in figure 3.17.
Figure 3.19 Webern, Variations (1936), mm. 1–7.
Figure 3.20 Webern analysis by dyads (a) and trichords by hand (b) as shown
circled—dyads by solid line and trichords by dotted line—in Figure
3.19.
Figure 3.21 Nine measures from Boulez, Structures (1952).
Figure 3.22 Boulez analysis by tetrachords by hand as shown circled in Figure
3.21.
Figure 3.23 Webern, Concerto for 9 Instruments, op. 24, 1st movement, mm. 1–7.
Figure 3.24 (a) Calculations of registral information in Webern, Concerto for 9 In-
struments, op. 24, 1st movement, mm. 1–3.
Figure 3.25 Webern, Concerto for 9 Instruments, op. 24, 1st movement, last 7
measures.
Figure 3.26 Calculations of registral information from Figure 3.25.
Figure 3.27 Boulez, Second Sonata for Piano (1948), 2nd movement, beginning.
Figure 3.28 Calculations of registral information in mm. 1–6 in Boulez, Second
Sonata for Piano (1948), 2nd movement.
Figure 3.29 The author, Triplum for flute and piano (1975), beginning.
Figure 3.30 Calculations of registral information in the author’s Triplum for flute
and piano (1975), mm. 1–7.
Figure 3.31 The author, Triplum for flute and piano (1975), mm. 98–104.
Figure 3.32 Another passage from the author’s Triplum for flute and piano (1975),
m. 43.
Figure 3.33 The 97 possible origin sets for [0,2,6].
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page xv

LIST OF FIGURES xv

Chapter 4
Figure 4.1 Beethoven, Piano Sonata in C Major, op. 53, beginning.
Figure 4.2 The numbers of iterations per pitch of the 298 pitches present in
Beethoven, Piano Sonata in C Major, op. 53, beginning (C, 48; C-sharp,
1; D, 26; E-flat, 11; E, 17; F, 57; F-sharp, 2; G, 52; A-flat, 16; A 13; B-flat,
31; B, 24).
Figure 4.3 A list of all 77 linear interval vector classes.
Figure 4.4 Four sets of scales related by symmetry (a, b, and c) and repeating
pattern (d).
Figure 4.5 Beethoven, Piano Sonata op. 7, mm. 97–105.
Figure 4.6 A series of possible scales, with each successive scale possibility lack-
ing one of the previous scale’s notes, the one with the shortest overall
duration. Each line here represents the scale’s transposed pitch-class
set beginning on 0, its linear interval vector, and the original pitch-
class set in ascending order.
Figure 4.7 Varèse, Density 21.5 for solo flute.
Figure 4.8 Computer scale analysis of the phrases of the three sections of Varèse,
Density 21.5 (see Figure 4.7), with author’s determination of best scale
possibility in boldface.
Figure 4.9 Computer analysis of Varèse, Density 21.5, by complete section with
author’s determination of best scale possibility in boldface and sec-
ondary possibilities marked by “*” and “†.”
Figure 4.10 Unordered but nontransposed scale pitch classes for each section in
Figure 4.7.
Figure 4.11 The first section of Varèse, Density 21.5, with non-scale tones circled.
Figure 4.12 Schoenberg, Six Little Piano Pieces, op. 19, no. 1 (1911).
Figure 4.13 Computer readouts from a three-phrase analysis of Schoenberg, Six Little
Piano Pieces op. 19, no. 1.
Figure 4.14 Computer analysis of the entirety of Schoenberg, op. 19, no. 1.
Figure 4.15 From the author’s Concerto for Cello and Orchestra (1994).
Figure 4.16 The author, Concerto for Cello and Orchestra (1994), computer analy-
sis by phrase.
Figure 4.17 Computer analysis of the excerpt from the author’s concerto in Figure
4.15, by section.
Figure 4.18 Computer analysis of the entirety of the excerpt from the author’s
concerto shown in Figure 4.15.
Figure 4.19 Selected measures of the solo cello part of the author’s concerto. The
scale used in the third movement has a related complement used
much as the dominant key might be used in a tonal concerto.
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page xvi

xvi HIDDEN STRUCTURE

Figure 4.20 A scale analysis of Stravinsky, Three Pieces for String Quartet (1914),
3rd movement (opening), as shown in Figure 1.19.
Figure 4.21 A pitch field that covers a span of two-plus octaves and contains two
exclusionary hexachord pitch-class sets.
Figure 4.22 A scale with a range of two-plus octaves for the example shown in Fig-
ure 4.21.
Figure 4.23 An example where the notes C-sharp, G, and B appear only in the up-
per register and the notes E-flat, F, and A appear only in the lower reg-
ister (Arnold Schoenberg, Das Buch der hängenden Gärten, II).
Figure 4.24 The scale differences between a one-octave representation and a two-
octave representation of the music in Figure 4.23.

Chapter 5
Figure 5.1 One possible overview of the structure of a work of music that is at
one and the same time a visual/musical representation and an OOP
representation.
Figure 5.2 Examples where (a) roots remain the same but tensions change and
(b) tensions remain fairly equal but roots change, demonstrating how
neither process alone can determine function in post-tonal music.
Figure 5.3 The overtone series from the fundamental C.
Figure 5.4 (a) Interval derivations from the overtone series along with their root
designations, and (b) interval root strengths shown in order of strength.
Figure 5.5 Six groupings to serve as examples of the root-identification process.
Figure 5.6 A list of intervals weighted according to their tension levels.
Figure 5.7 Figuring interval weights from the bass note by using the chart in Fig-
ure 5.6 produces 0.3 (M3 at 0.2 + P5 at 0.1), .5 (m3 at 0.225 + m6 at
0.275), and 0.8 (P4 at 0.55 + M6 at 0.25), respectively. The minor triad
and its inversions produce 0.325 (m3 at 0.225 + P5 at 0.1), .45 (M3 at
0.2 + M6 at 0.25), and 0.825 (P4 at 0.55 + m6 at 0.275).
Figure 5.8 The augmented triad at 0.475 (M3 at 0.2 + m6 at 0.275), the diminished
triad at 0..775 (m3 at 0.225 + A4 at 0.55), the dominant seventh chord
at 1.0 (M3 at 0.2 + P5 at 0.1 + m7 at 0.7), and the diminished seventh
chord at 1.125 (m3 at 0.225 + A4 at 0.65 + M6 at 0.25).
Figure 5.9 Four chords producing tensions of 1.2 (M3 at 0.2 + P5 at 0.1 + M7 at
0.9), 1.25 (m3 at 0.225 + P5 at 0.1 + m7 at 0.7), 0.55 (M3 at 0.2 + P5 at
0.1 + M6 at 0.25), and 2.0 (M3 at 0.2 + P5 at 0.1 + M7 at 0.9 + M2 at 0.8).
Figure 5.10 Chord tensions adding to 1.0 (M2 at 0.8 + M3 at 0..2), 1.225 (m2 at 1.0 +
m3 at 0.225), 1.8 (m2 at 1.0 + M2 at 0.8), and 1.475 (P4 at 0.55 + m7 at
0.7 + m3 at 0.225).
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page xvii

LIST OF FIGURES xvii

Figure 5.11 A simple lookup table of metric tensions for principal beats in twelve
different meters.
Figure 5.12 (a) J. S. Bach, Chorale no. 42, demonstrating the same chord in differ-
ing contexts; (b) Schoenberg, Six Pieces for Piano, op. 19, no. 6, open-
ing measures, demonstrating the same types of contextual differences.
Figure 5.13 SPEAC analysis of the music in Figure 5.12a and b.
Figure 5.14 An example of a unification as derived from Schoenberg, Three Piano
Pieces, op. 11, no. 1: (a) the unification, and (b) the first 8 measures of
the work from which the unification was drawn. Note the other unifi-
cations present here as well.
Figure 5.15 Two patterns, a target pattern and a potential matching pattern. If the
*intervals-off* controller were set to 2, the two patterns would
not match. However, if the *intervals-off* controller were set to
3, the two patterns would match.
Figure 5.16 A phrase of music by Mozart (a); a machine-composed replication (b);
the meta-pattern that binds them together (c).
Figure 5.17 Various levels (gradient map) of music revealed by applying filters.
Figure 5.18 The opening few beats of a graphical analysis of Stravinsky, Three
Pieces for String Quartet (1914), 3rd movement.
Figure 5.19 A root analysis of Stravinsky, Three Pieces for String Quartet (1914),
3rd movement.
Figure 5.20 SPEAC analysis (by chord change) of Stravinsky, Three Pieces for
String Quartet (1914), 3rd movement.
Figure 5.21 A structural graph of a section from the author’s Triplum for flute and
piano.

Chapter 6
Figure 6.1 (a) Mozart, Piano Sonata K. 279, 1st movement, mm. 1–3; (b) results of
the Melodic Predictor; (c) Mozart, Piano Sonata K. 533, 1st movement,
mm. 1–4; (d) results of the Melodic Predictor.
Figure 6.2 Recombination process leading to the completion of a new Bach-like
chorale phrase.
Figure 6.3 An example of differing types of internal patterns causing problems in
recombination: (a) and (b) originals, (c) a recombination without re-
gard to texture sensitivity.
Figure 6.4 An example of Experiments in Musical Intelligence output: the begin-
ning of the Rondo Capriccio for cello and orchestra arguably in the
style of Mozart.
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page xviii

xviii HIDDEN STRUCTURE

Figure 6.5 An example of allusion in (a) an Experiments in Musical Intelligence

replication; and (b) a Beethoven bagatelle Op. 119, no. 1 (1820), upon
which it is partially based.
Figure 6.6 A Bach chorale as example of a jigsaw puzzle.
Figure 6.7 The Bayes rule.
Figure 6.8 A state-transition matrix for a first-order Markov chain involving all
twelve notes of the chromatic scale.
Figure 6.9 The upper left-hand corner of a second-order Markov chain state tran-
sition matrix.
Figure 6.10 (a) Original, Bartók, Mikrokosmos, no. 71, transposed to begin on G;
(b–f) five computer extensions beginning five notes from the end of
the passage.
Figure 6.11 (a) Original, Bartók, Mikrokosmos, no. 77, transposed to begin on D;
(b–d) three computer extensions beginning five notes from the end of
the passage.
Figure 6.12 A simple two-voice counterpoint in which each voice moves stepwise
in various directions.
Figure 6.13 A post-tonal example, where the various two-note groupings do not
have clear functionalism.
Figure 6.14 Two possible nonidentical computer extensions (b and c) to the origi-
nal by Bartók (Mikrokosmos, no. 80) presented in (a).
Figure 6.15 Three different forms of analysis provided by Alice.
Figure 6.16 The rule (((3 3) 2 1) ((5 1) -2 2) ((0 0) 2 3) ((8 5) -1 4)) in music
notation.
Figure 6.17 Extending voice-leading rules.
Figure 6.18 Several student-created extensions (b–d) based on Bartók (a),
Mikrokosmos, no. 81.
Figure 6.19 The rule (((19 16) -3)) in music notation.

Chapter 7
Figure 7.1 The top ten levels of Pascal’s triangle.
Figure 7.2 Pascal’s triangle shown stacked to the left and modulo 2 with the ze-
ros identifying the Sierpinski gasket (a), and zeros removed (b) to
make the graphic more readable.
Figure 7.3 A simple 5 × 5 magic square in which all horizontal ranks and vertical
columns sum to 65.
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page xix

LIST OF FIGURES xix

Figure 7.4 A magic square containing intervals that equate (when added to-
gether) to 5 in top-to-bottom and left-to-right directions, along with
musical examples.
Figure 7.5 The author, Concerto for Cello and Orchestra, 2nd movement
(excerpt).
Figure 7.6 A magic cube with both incremental numbers and music intervals
(with directions).
Figure 7.7 A formal analysis of the first 89 measures of Bartók, Music for Strings,
Percussion and Celeste (Lendvai 1983, p. 74), based on the golden
mean and the Fibonacci series.
Figure 7.8 The first 14 measures of Varèse, Density 21.5, with much the same
pitch classes as the silver variation.
Figure 7.9 A simple model of a neural network.
Figure 7. 10 A simple example of an association network.
Figure 7.11 An example of Muse’s grouping process.
Figure 7.12 All of the possible permutations of a six-note list of nonrepeating num-
bers not including the original arrangement ([0,2,7] [1,5,8]) and its ret-
rograde ([1,5,8] [0,2,7]).
Figure 7.13 A view of the listener window after the completion of one analysis by
Muse.
Figure 7.14 Muse’s program includes a function called sleep.
Figure 7.15 The Eine Kleine Stück arguably in the style of Arnold Schoenberg by
Experiments in Musical Intelligence.
Figure 7.16 Schoenberg, Three Piano Pieces, op. 11, no. 1 (opening).
Figure 7.17 The first page of the score to Beethoven, Symphony no. 5, revision.
Figure 7.18 Dies irae, a medieval Latin sequence and the first words of the Re-
quiem Mass.
Figure 7.19 (a) A passage from the Mystic Circle of the Young Girls in Stravinsky,
The Rite of Spring ; (b) a passage from Stravinsky, Three Pieces for
String Quartet.
Figure 7.20 (a) A passage from the 3rd movement of Stravinsky, Three Pieces for
String Quartet; (b) a similar passage in the Symphonies of Wind Instru-
ments.
Figure 7.21 A post-tonal work by the author resembling Stravinsky’s style.
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page xx
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page xxi

Preface

Today’s computers provide music theorists with unprecedented opportunities to

analyze music more quickly and accurately than ever before. Whereas analysis once
required several weeks or even months to complete and was often replete with hu-
man errors, computers now provide the means to accomplish these same analyses
in a fraction of the time and with far more accuracy. However, although such com-
puter music analyses represent significant improvements in the field, computational
analyses using traditional approaches do not in themselves constitute the true inno-
vations in music theory that computers offer. In this book, I introduce a series of an-
alytical processes that—by virtue of their concept and design—can better, and in
some cases only, be accomplished by computer programs, presenting unique oppor-
tunities for music theorists to better understand the music they study.
As an example, when comparing computational and human analysis, we need to
distinguish between time differences of minutes or hours and of months or years.
Analyses that may take a human minutes, hours, or days may still be undertaken
by hand. Analyses that require a human months or even years may not even be
considered feasible. Without these latter analyses, however, our range of choices of
analytical processes becomes arbitrarily limited. With computers able to reduce
our months or years to fractions of seconds and accuracy to near perfect, music
analysts must rethink many theories that they may have previously dismissed as
impossible due to the magnitude of the undertaking. Computers can not only re-
duce our fatigue and increase our accuracy, but they can open whole new worlds
previously considered unachievable—if considered at all.
Many readers of a book like this might value an annotated bibliography of cur-
rently available software for tonal and post-tonal music analysis, but most would
find this kind of listing woefully out of date the very day it became available; the
field simply changes too rapidly to make this kind of offering practical. Instead, this
book—with the exception of the broad introductory first chapter—focuses on sev-
eral unique approaches to music analysis offered by computer programs. Although
these approaches do not represent an all-encompassing and integrated global
theory of music analysis, they do represent significantly more than a compilation of
loosely related computer program descriptions. Chapter 5, on function in post-
tonal music, for example, firmly depends on the scalar foundations presented in

xxi
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page xxii

xxii HIDDEN STRUCTURE

Chapter 4. Likewise, the logic of Chapter 7’s presentation of a multi-tiered approach

to musical analysis depends on the material of all of the preceding chapters. In
short, this book represents an integrated view of computer music analysis.
Hidden Structure also centers on post-tonal rather than tonal music for several
reasons. First, as we shall see in this chapter, tonal music analysis has a long and
distinguished history that details approaches to melody, harmony, counterpoint,
form, and structure. Although the advent of computer technology can certainly add
tools and allow the discovery of new concepts, the range of potential for truly new
approaches is, I feel, somewhat limited compared to the potential for analytical
possibilities in post-tonal music. Second, many diverse and useful programs for
tonal music analysis already exist. In fact, most early computer computational ex-
periments with music analysis involved tonal music. On the other hand, although
many computer pitch-class set programs exist, little beyond this easily created soft-
ware is currently available. Third, and possibly most important, post-tonal music
represents the lingua franca of today’s concert music, the music in which I am most
interested and to which I wish to devote my research time.
The ideas described in this book are ordered from simple to complex, and from
more traditional to possibly more innovative. Chapter 2’s focus on information
theory mirrors the work done by early music theorists using computers, while
Chapter 6’s concentration on computer modeling parallels more recent work in
computer composition using rule acquisition. Most of these ideas also share com-
mon perspectives. For example, they have the same post-tonal focus, use similar
artificial-intelligence approaches, and share the same computational foci. Thus, this
book represents an integrated view of music analysis rather than a potpourri of
interesting but unrelated analytical perspectives.
Although this book is meant for a general audience, with the only prerequisite
being the ability to read music, it is particularly intended for both musicians inter-
ested in the application of computers to the understanding of music and computer
scientists—both professionals and those amateurs who enjoy the sport of benign
hacking—wishing to use their expertise to better understand music. Given my sus-
picion that many of these individuals do not share a great deal of common ground,
I have attempted to educate each group in the territory of the other. I therefore
define the terms and concepts used very carefully. For example, music theorists
will require an understanding of information theory, programming, and so on. Like-
wise, non–music theorists will require an understanding of musical set theory,
chord roots, and so on. By doing this, however, I fear that both will consider many
of the chapters here uneven, at least in the sense that computer scientists, for ex-
ample, understand set theory, just not musical set theory, and vice versa for musi-
cians. Thus, readers may find themselves in shallow water for one section of a
chapter and then suddenly in deep water. I apologize for the lack of uniformity that
may result. Hopefully, the chapter subheadings will allow those already familiar
with one or more of the concepts described to skip to the next section relevant to
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page xxiii

PREFACE xxiii

their interests if they so desire. I believe that those who stick with this book, despite
its bumpy ride, will gain a new appreciation for the views of music offered here.
Chapter 1 begins with principles and definitions, followed by a survey of histori-
cal algorithmic analysis processes and a broad overview of the computer music
programs in use today. This overview does not just list the many different imple-
mentations of particular analytical processes, but describes representative exam-
ples of them. For example, pitch-class set analysis programs in Java can be found at
dozens of Internet sites at present, the variations between them limited primarily
to differences in graphical user interfaces. Therefore, the coverage here is relatively
limited. Other programs, such as the Humdrum Toolkit, that provide more univer-
sal algorithms are discussed in more detail. Chapter 2, on music and algorithmic
information theory, describes how a form of compression (replacing redundant ma-
terial with signifiers to conserve space) can reveal important statistics about post-
tonal music. It also proposes an analytical technique that incorporates a dynamic
integrative approach to aspects of pitch, rhythm, texture, and dynamics in an at-
tempt to understand the constantly changing foci of musical works.
Chapter 3, on set analysis of register and range, introduces a duodecimal nota-
tion for representing pitch-class sets, a notation that, by virtue of its having a base
(radix) of twelve, allows both register and pitch class to share one relatively simple
notation. Integrating these two interrelated aspects of musical pitch in a pitch-class
abstraction enables the comparison of equivalent prime-form sets with significant
registral differences, revealing contrasts as significant as those between prime-form
sets that share very similar registral arrangements. Chapter 4 presents a computer
analysis of musical scales and describes a method for generating all possible equal-
tempered scales and an approach to efficiently discovering and cataloging these
scales, particularly in post-tonal music. Because scales have scale degrees that in
tonal music signify musical functions, the concept of post-tonal scale analysis
has far-reaching implications. This is the subject of Chapter 5, which more fully
explores the implications of discovering function in post-tonal music, defining and
analyzing potential musical progressions, cadences, chromatic harmonies, hierar-
chies, and so on.
Chapter 6 focuses on generating rules from music itself, rather than imposing
user-prescribed rules. I end in Chapter 7 with a possibly brazen (but hopefully use-
ful) look to the future at some of the areas of musical analysis I feel will develop
over the next few decades thanks to the availability of new computational tools.
I have chosen commonly analyzed musical examples for this book so that read-
ers will be able to better compare the results of the analytical processes I describe
here with the results of other, more standard analytical techniques. I have also in-
terpolated examples of my own music here and there in order that readers may
more directly ascertain how the techniques described further prove valuable in the
creative process. I explicitly used the analytical processes described in these com-
positions, and these works represent the best cases I can prove of such usage.
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page xxiv

xxiv HIDDEN STRUCTURE

I have limited the scope of study for this book to classical post-tonal music.
Other genres of music might serve equally well. However, classical post-tonal music
provides a comprehensive range of music over a significant historical period. As
well, my own background consists almost exclusively of classical tonal and post-
tonal music, and hence I lack the expertise to intelligently discuss other genres of
music. Readers may offset this shortcoming by applying the techniques defined
and described here to whatever style of music they know best.
The various computer programs discussed in this book, along with MP3 versions
of all of the book’s musical examples, are available on the CD-ROM that accompa-
nies this book. The programs are written in Common Lisp, of which two versions
are available: (1) Macintosh platform, or (2) any platform that supports Common
Lisp. To ensure that these latter programs will perform in different environments
requires the omission of platform-dependent code such as all Musical Instrument
Digital Interface (MIDI) and graphical user interface (GUI) functions. Full documen-
tation and operating instructions are included with each program.
As time permits, I will provide code updates for new versions of the Macintosh
operating system on my Web site (arts.ucsc.edu/faculty/cope). However, if history
proves accurate, as soon as I write new platform-dependent programs, system
hardware or software changes will once again render the code almost immediately
obsolete. Furthermore, the software for the programs in this book, although very
helpful in demonstrating the principles of each chapter and in clarifying the ana-
lytical principles proposed, is not critical to the understanding of the material
presented.
Whenever possible in this book, I have included the thoughts of music theorists,
musicologists, mathematicians, computer scientists, and cognitive scientists whose
work complements or poses the greatest challenges to the ideas presented here.
My apologies to those whose work may seem relevant, but to which I have not re-
ferred here due to space limitations.
Many individuals have advised me in this study of computer music analysis, par-
ticularly graduate students and colleagues at the University of California at Santa
Cruz (notably Paul Nauert, Ben Carson, and Daniel Brown). I also owe immense
gratitude to Nico Schüler for forwarding a copy of his dissertation (2000), without
which I would have floundered more than I already have in the first chapter’s his-
tory of computer music analysis. Keith Muscutt, Eric Nichols, and Irene Natow pro-
vided much-needed advice (editorial and otherwise). Many of the ideas in this book
originated from my teaching over the years, and I thank the many classes of stu-
dents who acted as guinea pigs for my theoretical explorations. Without support
from colleagues and students such as these, this book could not have been
completed.
Hidden Structure describes a few of the ways in which I believe computer pro-
grams can contribute significantly to the analysis of music, particularly to the often
difficult-to-understand post-tonal music of our time. Although computer programs—
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page xxv

PREFACE xxv

through their incredible speed and accuracy—certainly aid our ability to analyze
this music in more or less traditional ways, creating programs that analyze music
using more indigenous computational approaches offers greater challenges and po-
tential rewards. I hope that my descriptions of several such programs here will en-
courage others to continue to seek newer and more revealing analytical processes
in the future.

Acknowledgements
The author and publisher are grateful for the use of the following material:

Figure 1.14 Theodore Presser Company

Figure 1.15 Permission granted by Nico Schuller.
Figure 1.17 Bach Chorale 002 from the Music Theory Workbench version 0.1
(http://pinhead.music.uiuc.edu/~hkt/mtw/pdf/), Heinrich Taube, author.
Figure 1.18 Permission granted by IRCAM.
Figure 1.19 © Copyright 1923 by Hawkes & Son (London) Ltd
Reproduced by kind permission of Boosey & Hawkes Music Publisher Ltd
Figure 2.9 © Copyright 1940 by Hawkes & Son (London) Ltd
Figure 2.15 Reproduced by permission of Spectrum Press (spectrumpress.com)
Figure 3.8 © 1913, 1940 by Universal Edition A.G. Wien/UE 5069
Figure 3.17 Used by permission of Belmont Music Publishers, Los Angeles
Figure 3.19 © 1937 by Universal Edition A.G. Wien/UE 10881
Figure 3.21 © 1955 by Universal Edition (London) Ltd. London/UE 12267
Figure 3.23 © 1948 by Universal Edition A.G. Wien/UE 11830
Figure 3.25 © 1948 by Universal Edition A.G. Wien/UE 11830
Figure 3.27 Heugel and Co. 1950
Figure 3.29 1975 Carl Fischer Music Publishers
Figure 3.31 1975 Carl Fischer Music Publishers
Figure 3.32 1975 Carl Fischer Music Publishers
Figure 4.7 G. Ricordi and Co.
Figure 4.11 G. Ricordi and Co.
Figure 4.12 © 1913, 1940 by Universal Edition A.G. Wien/UE 5069
Figure 4.23 Dover Publications 1995
Figure 5.12b © 1913, 1940 by Universal Edition A.G. Wien/UE 5069
Figure 5.13b © 1913, 1940 by Universal Edition A.G. Wien/UE 5069
Figure 5.14b Belmont Music Publishers, Los Angeles 1910
Figure 5.19 © Copyright 1923 by Hawkes & Son (London) Ltd
Reproduced by kind permission of Boosey & Hawkes Music Publisher Ltd
Figure 6.5 Reproduced by permission of Spectrum Press (spectrumpress.com)
Figure 6.10a © Copyright 1940 by Hawkes & Son (London) Ltd
Figure 6.11a © Copyright 1940 by Hawkes & Son (London) Ltd
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page xxvi

xxvi HIDDEN STRUCTURE

Figure 6.14a © Copyright 1940 by Hawkes & Son (London) Ltd

Figure 6.18a © Copyright 1940 by Hawkes & Son (London) Ltd
Figure 7.7 Lendvai, Ernö. 1983. The Workshop of Bartók and Kodály. Budapest: Editio
Musica.
Figure 7.8 G. Ricordi and Co.
Figure 7.15 Reproduced by permission of Spectrum Press (spectrumpress.com)
Figure 7.16 Belmont Music Publishers, Los Angeles 1910
Figure 7.19 © Copyright 1923 by Hawkes & Son (London) Ltd
Reproduced by kind permission of Boosey & Hawkes Music Publisher Ltd
Figure 7.20 © Copyright 1923 by Hawkes & Son (London) Ltd
Reproduced by kind permission of Boosey & Hawkes Music Publisher Ltd
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page xxvii

Description of the
CD-ROM

The CD-ROM that accompanies this book contains materials that augment the text
of Hidden Structure.

MP3S ON THE CD-ROM

The CD-ROM contains MP3s of all of the musical examples presented in this book,
labeled according to figure number. These MP3s are machine performed using MIDI
(Musical Instrument Digital Interface) files and Macintosh Quicktime Musical Instru-
ments. The results are thus mechanistic in performance and relatively shallow in
timbral quality and are not intended as masterful performances. They simply pro-
vide simple aural replications of the figures in the book for those unable to play or
otherwise hear them.

SOFTWARE ON THE CD-ROM

The software found on the CD-ROM includes all of the programs described in this
book. Details on how to operate this software are found below, in the chapters of
this book, and in the files containing the software. Triangulation of these three
sources should provide readers who lack a background in computer programming
with the ability to run the code.
The software found on the CD-ROM is written in Common Lisp in source code
form (readable as opposed to object code; unreadable compiled binary code). The
source code requires Common Lisp in order to function. For programs labeled (CL),
any form of Common Lisp (of which there are many free available forms for use on
any computer platform) may be used. For such programs, simply download any
Common Lisp, load the program into the CL program by any of several methods
(see the CL program itself for a description of these processes), and then follow the
instructions provided at the top of each source code file, which is readable in any
text program. For programs labeled MCL (Macintosh Common Lisp), you must
download a form of MCL that works with your particular computer (see the digi-
tool.com Web site for more information) and follow the instructions provided.
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page xxviii

xxviii HIDDEN STRUCTURE

The source code consists of a series of variable declarations (preceded by “def-

var”) and function definitions (preceded by “defun”) enclosed in various parenthe-
ses that help connect variables and processes. Objects (denoted by “defclass”)
define windows, menus, and so on. No knowledge other than this general prescrip-
tion is required to run the code provided here.
I created the software here primarily for demonstrating the principles expressed
in the associated chapter of this book. All of the program runs in the figures, for ex-
ample, result from using this software. However, the code is not bulletproof. In
other words, loading some music (e.g., performed music) may create problems for
some of the functions and thus cause programs themselves to fail. Ensuring that
code can endure any input not only would take hundreds of hours of testing and
even then be incomplete, it would also require several times the amount of code
provided here, making the result unreadable except by experts. Problems encoun-
tered with these programs should be reported to the author at [email protected].

SOFTWARE ON THE CD-ROM BY CHAPTER

Chapter 1
Sets (CL/MCL)
Visualize (MCL)
Chapter 2
Comparison (CL)
Multigraph (MCL)
Chapter 3
SetMath (CL)
Set Multiples (CL)
Set Database Analysis (CL)
Register (CL)
Chapter 4
Scale Analysis (CL)
Chapter 5
Root (CL)
Structure Map (CL)
Structure Graph (MCL)
Chapter 6
Markov (CL)
Extend (CL/MCL)
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page xxix

DESCRIPTION OF THE CD-ROM xxix

Chapter 7
Neural Net (CL)
SPEAC (MCL)
Muse (CL)
01_DAS_FM_ppi-xxx 1/29/09 1:06 PM Page xxx
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 1

ONE
Background

Throughout the centuries, the arts have undergone transformations that paral-
leled two essential creations of human thought: the hierarchical principle and
the principle of numbers. In fact, these principles have dominated music, partic-
ularly since the Renaissance, down to present-day procedures of composition.
Iannis Xenakis, Formalized Music (1971), 204

This chapter begins with definitions of the principles and terms that act as founda-
tions for the ideas, processes, and programs to follow. It continues with a survey of
historical algorithmic music analyses (often called paper algorithms) and a brief
look at the computer music programs in use today and some representative exam-
ples of particular analytical processes. It concludes with a description of two sim-
ple programs, one for visualizing music of all types and the other for delineating
pitch-class sets, particularly in post-tonal music.

PRINCIPLES AND DEFINITIONS

As I mentioned in the preface, Hidden Structure does not build a single theory of
post-tonal music, but rather describes several diverse methodologies for computer
analysis. However, this book does focus on four central principles:
1. all music consists of patterns;
2. all pitch patterns can be reduced to scales;
3. all elements of scales have different functions; and
4. all patterns, scales, and functions in music are best understood by modeling
their processes.

The first three principles listed here seem plain enough. The last, however, de-
serves elaboration. The term “modeling” as used here refers to the building of
accurate replicas of objects or phenomena in order to better understand them.
The concept of modeling as an alternative to reverse engineering, though not new,

1
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 2

2 HIDDEN STRUCTURE

has—with the advent of computational technology—gained a significant following

in recent years. For example, Stephen Wolfram (2002) has argued that traditional
scientific methodology is like a salmon losing ground against a strong current, wag-
ing a futile battle in its attempts to answer the important scientific questions of our
day. He further argues that the only true revelations of the universe will come from
computationally modeling it. Without either agreeing or disagreeing with Wolfram’s
point of view on science, I argue vigorously for the usefulness of modeling as an ap-
proach to musical analysis, as later chapters in this book will testify.
Each of the chapters in this book relates to all of the above principles to some
extent. However, Chapters 1, 2, 3, 6, and 7 concentrate particularly on Principle 1,
Chapters 4 and 7 on Principle 2, Chapters 5 and 7 on Principle 3, and Chapters 6
and 7 on Principle 4. These principles then provide a framework on which a more
general theory might be developed. As the processes described in this book unfold,
I will remind readers of how the material under consideration exemplifies these
four principles.
Because the word atonal is somewhat vague and its invocation often has deroga-
tory inferences (see Straus 1990, v), I use the term post-tonal here instead. There
are several current definitions of this word, ranging from “not explicitly tonal” to
“twelve-tone” or “serial.” In this book, I will use a more liberal definition. This
means that terms such as polytonal, octatonal, and serial all fall under the umbrella
term post-tonal. This broad definition will make it unnecessary to differentiate be-
tween terms for music by, for example, Igor Stravinsky, Béla Bartók, Edgard Varèse,
Arnold Schoenberg, and Anton von Webern.
The terms music theory and music analysis seem interchangeable to some. In this
book, however, I define music theory as a more or less descriptive account of the
principles of musical structure, and music analysis as the more or less factual ac-
count of what literally occurs in music itself. Anthony Pople has it right: “There is a
broad historical distinction between music theory—which studies musical works in
order to deduce ‘more general principles of musical structure’—and music analy-
sis, in which the interest is focused on individual pieces of music” (Pople 2004,
127). Therefore, whenever I refer to music theory, I am speaking of basic tenets that
govern all music. When I refer to music analysis, I am speaking of specific instances
of those principles in a more or less restricted body of music. For the most part, if
not stated otherwise, this book deals with music analysis rather than music theory.
To some extent, then, music analysis as referred to here resembles cryptogra-
phy, or code breaking. As Nicholas Cook puts it:

Music is a code in which the deepest secrets of humanity are written: this heady
thought assured musical studies their central place in ancient, medieval and re-
naissance thought. And though the study [analysis] of music no longer occupies
quite so elevated a role in intellectual circles, some of today’s most important
trends in the human sciences still owe it a debt. (Cook 1987, 1)
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 3

BACKGROUND 3

We have already partially broken many such musical codes; after all, the discovery
of function in tonal music, along with its ancillary, harmonic syntax, represents
hundreds of years of cryptographic investigations. Yet, in the view of some ana-
lysts, more code breaking remains. For example, no convincing corollary has yet
been found for function in post-tonal music, nor has a clear process for defining
musical structure been revealed in post-tonal music, nor has a defined syntax been
presented for most of the post-tonal music of the twentieth and twenty-first cen-
turies. Even tonal music holds many mysteries yet undiscovered. Hopefully, the
code breaking that remains can be enhanced with the aid of computers.
The term algorithm will be used often in this book and requires a clear defini-
tion. An algorithm (see Cope 2000, 1ff., for a more complete definition) is a recipe
for achieving a goal in a finite number of clearly defined steps. For example, the in-
structions “extract the dysfunctional bulb by turning it counterclockwise, insert
new bulb, turn new bulb clockwise several times until tight, and turn on the light”
constitutes a simple algorithm for replacing a burned-out light bulb. Deoxyribonu-
cleic acid (DNA) replication, blinking, breathing, and so on also represent algo-
rithms. Note that algorithmic analysis does not depend on computer hardware or
software.
The most interesting example of the notion of an algorithm is Alan Turing’s 1952
experiment with computer chess. Turing—the oft-credited father of artificial
intelligence—had so tired of waiting to challenge the first computer chess-playing
program that he invited a friend to play a game of computer chess, with Turing as-
suming the role of computer. Turing spent days creating a paper algorithm, a list of
rules required to respond to whatever imaginable move his opponent might make.
Turing and his opponent’s subsequent game lasted three or so hours, during which
Turing rigorously followed the rules he had compiled. Turing eventually lost the
game. He argued, however, that he lost because his program, called Turochamp,
had not been complete or accurate enough. Thus, his point was made—by follow-
ing his algorithm exactly he had fulfilled his role as computer. If a human computer
could play chess algorithmically, then certainly a hardware computer following
human-programmed rules could do so as well (see Standage 2002, 226–229, for
more information).
Obviously, however, a computer can play such games only as well as its human-
provided rules allow it to play. My point here is simple: computers simply obey the
dictates of their programmers; they are limited only by the programmer’s skill in
developing the right algorithm. Computers are tools, nothing more and nothing
less. Thus, although computers process data faster and more accurately than hu-
mans, algorithms for analyzing music have existed for centuries.
In tonal music, using dominant-tonic cadences, avoiding parallel fifths, resolving
dissonances, creating formalistic canons and fugues, and other creative processes
result from algorithms that composers invoke—whether they realize it or not—in
order to plan their compositions and participate in the musical style of their day.
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 4

4 HIDDEN STRUCTURE

One could also argue the algorithmic foundation of much post-tonal music on the
basis of the revelations of set theory alone. Serial music requires less substantia-
tion. Even Cage’s indeterminate music often relies on rigorously applied algorithms.
In short, algorithms play significant roles in all music.
Interestingly, the traditional processes of tonal analysis that pervades music
academia today represent an algorithm. The rules by which students reduce Bach
chorales to registerless triads, reveal their inversions, and ultimately use roman
and arabic numerals for describing functions and their inversions within keys
clearly denote a set of instructions to achieve a particular goal—the standard defi-
nition of an algorithm.
Because computers offer the fastest, most accurate, and most efficient manner
of applying algorithms to data, using computers for music analysis is a natural con-
sequence of these ideas. Indeed, computers can process information exponentially
faster than humans. For the most part, computers offer users the ability to extend
themselves in ways that no other means can match. Given the right algorithm,
then, computers present the best opportunity for music analysts to understand the
music of the future and to better understand the music of the present and the past.
Before beginning a brief history of related music research and analysis, and be-
cause Hidden Structure conjures up very different meanings to different people, I
would like to preface my historical survey with what I hope is a clear definition of
what the title means for this book. Before stating this definition, however, I feel
obliged to articulate what the title does not mean, along with my reasons for ex-
cluding such definitions.
Many readers, composers in particular, may take the title Hidden Structure: Music
Analysis Using Computers to mean analysis of musical timbre. After all, most jour-
nals dealing with computer music composition (Computer Music Journal, Perspec-
tives of New Music, Journal of New Music Research, etc.) routinely publish articles
devoted to constructing and deconstructing musical sound, called synthesis. I sus-
pect that were one to take a survey, the great majority of books and articles cover-
ing computer music analysis do so from the perspective of the analysis of timbre,
rather than from the perspective of the analysis of compositional technique. Analy-
sis of musical timbre, however, will not appear in this book, because the analysis to
which I refer in its title covers the relationships between sounds and not the
sounds themselves.
The title of this book may indicate to some the computer analysis of musical per-
formance. Creating software to perform in the style of human performers has long
been a focus of computer music research. Indeed, the performer’s nuances of
rhythm, meter, dynamics, articulation, and other parameters are a critical compo-
nent of human appreciation of music, and their analysis would seem vitally impor-
tant. Because this book can only cover so much territory, however, I have opted to
not include computer music analysis of performance. In fact, including perfor-
mance analysis would most likely obscure many of the points I wish to make.
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 5

BACKGROUND 5

Computers have also proven valuable in the translation of notated music (either
in autograph or printed form) into digital data for computer performance, engraved-
quality scores and part printing, storage, and other forms of digital coding. Such
digitizing, however, though extremely important, does not in itself provide any true
analysis of music and therefore does not represent a logical subject for this book.
Music perception is an absolutely critical component to understanding music,
and, as such, one might expect to find music cognition an actual cornerstone of a
book titled Hidden Structure: Music Analysis Using Computers. My reasons for exclud-
ing it here are many, but two are salient. First, studies of music cognition have only
recently begun, and many of the initial results conflict with one another to the
point of seriously confusing the current state of research. Second, this book is
based, at least in part, on the assumption that the analytical process discussed can
be programmed. Research in music cognition often results from studies relating to
statistics of how human subjects react to musical input, rather than objective—and
thus programmable—principles and data.
Readers interested in the analysis of electronic music (see Simoni 2006), impro-
vised music, music of oral traditions, and other such topics will also be disap-
pointed, because this book will not shed much light on these important topics. At
the same time, most of the principles described here can be applied to these con-
nected areas of research. I leave such application, however, to other, more qualified
individuals.
What, then, does the title Hidden Structure: Music Analysis Using Computers actu-
ally mean? In brief, this book focuses on the analytical study of Western classical-
music notated scores that can be represented digitally. Although a number of
individuals have attempted to quantify important aspects of music such as perfor-
mance practice (Lawson and Stowell 1999; Rink 1995) and other nonnotatable infor-
mation that composers may have intended beyond the actual score, printed music
remains the most reliable resource available for pitch, rhythm, dynamics, and other
of its parameters. Thus, I have limited the investigations presented here to notated
scores only.
Given my exclusion of several very important aspects of music, my actual defini-
tion of computer music analysis for this book may seem to many readers quite lim-
ited. This is intentional. The programs that accompany the chapters on this book’s
associated CD-ROM require such limitations. My work here is intended not to con-
jecture, but to prove. In order to provide these proofs, I must divide and conquer—
a mantra I repeat ad nauseam to my students. Hidden Structure thus will, I hope,
provide useful grist for others interested in understanding music, particularly mu-
sic of the twentieth and twenty-first centuries.
This book’s original title, “Algorithmic Music Theory,” better describes the his-
tory that follows this definition of the subject. At least half of this history covers
theories of music prior to the advent of computers, theories that nonetheless share
a single important process—the algorithm. Thus, despite the revised title, I ask all
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 6

6 HIDDEN STRUCTURE

readers to consider much of this chapter’s contents in particular and the historical
sections’ contents in general as guided by my original title, focusing on algorithms
rather than on the computer’s ability to process data more quickly and more accu-
rately than humans.
Before progressing to subsequent sections of this chapter, I should mention a
few other caveats. Throughout the book, the processes described focus almost ex-
clusively on pitch. I include rhythm (Chapters 2 and 5, for example), timbre, dy-
namics, articulations, and so on, but only occasionally. I do not intend this pitch
bias to suggest that pitch represents the nadir of musical parameters, but rather to
allow development of one aspect of music more thoroughly. I would rather not dis-
cuss, say, rhythm at all than discuss it shallowly just to ensure balance.
I also do not consider composer intent a factor in the music analysis discussed
here. Although music analysts should not ignore intent, especially when the com-
poser has made this intent public, composers are often unaware of many of the
processes they use while composing. To ignore these other processes while focus-
ing on the processes the composer has claimed as central to a composition would
render analysis practically impotent. In short, at least for this book, composer in-
tent will not take precedence over any other logical process that may be equally or
more revealing.
Finally, although listening should inform analysis, the reverse should also be
true. In fact, because I cannot predict what readers will hear in any of the examples
in this book, I take the viewpoint that the analyses presented here inform listening
and not vice versa. Listeners are biased by the music that they have heard, by their
own unique biological and chemical constitutions, by cultural and other learned
musical habits, and, of course, by personal musical sensibilities and aesthetics.
For millennia, astronomers and other celestial observers assumed that the sun
and stars revolved around the earth. However, in the Renaissance, three noted as-
tronomers, Nicolaus Copernicus (1473–1543), Johannes Kepler (1571–1630), and
Galileo Galilei (1564–1642), independently formulated, developed, and defended a
very different concept of the heavens: heliocentrism, the idea that the earth and
other planets revolved around the sun. So objectionable was this new idea to most
of the Western world that Galileo suffered severe scientific, social, and, in particu-
lar, religious persecution. Today, of course, we know that Copernicus, Kepler, and
Galileo were correct, and their theories have subsequently been applied to star
clusters, galaxies, and even clusters of galaxies called superclusters. In this in-
stance, our original analytical approach—observing the sun and stars move across
our sky—was actually an impediment to understanding the real physics involved.
Although the original theories seem almost silly today, they can provide a very im-
portant lesson for music analysts: analytical methods for better understanding mu-
sic, no matter their apparent worth in the short run, may in fact actually prevent
our ability to really understand the music we study. Although I do not pretend to
know the truth of this possibility, or, if true, a better way to proceed, I profoundly
believe that alternate ways to analyze music should not be ignored, no matter how
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 7

BACKGROUND 7

improbable they may seem. I hope that readers will bear these thoughts in mind as
the book unfolds.

A BRIEF HISTORY OF ALGORITHMIC ANALYSIS

Almost all music analysis is algorithmic in the sense that it compares musical
processes in a work under study to a corpus of known rules. I use two strategies to
focus the following survey on particular algorithms: choosing only those algorith-
mic processes that have made significant and quantifiable changes in the manner
in which music analysts analyze music, and wherever possible pointing readers in-
terested in pursuing more information on those processes I do mention—or even
processes that I do not explicitly mention but to which I implicitly refer—toward
important general and specific sources. My use of the word algorithm in precompu-
tational analysis also relates to analyses that are clearly programmable in some
meaningful way. It is in this spirit, then, that I begin the history of computer music
analysis long before the advent of actual computational hardware.
Furthermore, the following narrative in no way fully covers the subject of algo-
rithmic music theory, which itself could fill a very large book. Indeed, if my premise
is true—that most music analysis is algorithmic—then most volumes on the gen-
eral subject of the history of music analysis written in the last millennium them-
selves cover this subject. Therefore, particularly because history is not my primary
subject here, I include this brief outline in order to provide a broad context for the
discussion that follows and, hopefully, direct readers toward many sources that
they themselves can and should research further.
Note how, as the following historical outline proceeds, the reference to three of
the four principles described at the outset of this chapter follow in incremental or-
der, with early analytical approaches involving issues of pattern (shape, repetition,
cadence, etc.), medieval studies interested in scale (modes, etc.), and eighteenth-
and nineteenth-century analysis concerned with function (chord roots, etc.). The
section that follows that will then deal with a similar incremental ordering, but,
with the advent of computer analysis, include Principle 4—modeling in addition to
reverse engineering.
Although Pythagoras (582–500 BCE; see Cazden 1958), Plato (427–347 BCE, espe-
cially The Republic), and Aristotle (384–322 BCE, especially The Politics) postulated
many new ideas about the theory and analysis of music, most of these ideas were
philosophical or mathematical in regard to tuning (the ratios of Pythagoras princi-
pally) and thus suggest algorithmic frequency determination not particularly re-
lated to the purpose of this history (Winnington-Ingram 1929). Aristoxenus (b. 354
BCE), in his two major books Harmonics and Rhythmics, attempted to define inter-
vals (Principle 1), scales (Principle 2), keys, melody, and consonance, as well as
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 8

8 HIDDEN STRUCTURE

various related ideas (Winnington-Ingram 1932). Although Aristoxenus avoided

whenever possible defining his subjects in terms of tuning ratios—a very popular
approach during this period—he was also an astronomer, and he found it advanta-
geous to use numerical measurements when describing musical phenomena such
as intervals, scales, and consonance. Figure 1.1 presents Aristoxenus’s division of
the tetrachord into thirty-two equal parts, each part being a twelfth of a tone. An
early twentieth-century translation by Macran (1902) presents a series of enhar-
monic graphs representing the types and subtypes of the intervals and scales de-
scribed more fully in the book itself.
Without the groundwork established by Aristoxenus, Claudius Ptolemy (second
century CE, especially his Harmonics) could not have described the full comple-
ment of the Greek modes (Book III) and detailed them in ways that today could eas-
ily be programmed for analysis and composition (Shirlaw 1955). Ptolemy’s limiting
of the overall range of modes to two octaves (systema teleion or “perfect system”)
forced the modes in transposition to lose members at the upper limits of their
range and regain them at the lower limit of their range, a particularly computable
process. He also accepted the notion of seven modes (Principle 2), because that
number produced the central octave of the male vocal range.
Ptolemy’s studies, among others, prompted Aristides Quintilianus in the third
century CE to divide the study of music into three principal categories: theoretical,
practical, and numerical. Quintilianus’s Perí mousikês (On Music) further separates
the purely theoretical study of music into four distinct groupings (scientific, techni-
cal, critical, and historical). Although these categories do not strictly apply to mu-
sic theory, analytical processes are attached to aspects of each. Quintilianus’s
treatment of music as a numerical art connected it directly to mathematics and
involved patterns (Principle 1) such as triads, which he considered an important
concept allied with beauty.
Anicius Boethius (480–524), notably in his De institutione musica, inherited much
of Ptolemy’s modal concepts while continuing to advocate traditional Greek analyti-
cal processes: “Greatly influenced by Greek writers such as Nicomachus, Ptolemy,
Euclid, Plato, and Aristotle, the young Boethius set out to write works treating arith-
metic, music, geometry, and astronomy as disciplines that lead the soul to its first
encounter with incorporeal knowledge” (Bower 2002, 141). Boethius’s De institutione
musica (c. 500) follows the ideal of the Greek quadrivium, and its conservatism
deeply influenced the music theory of Western Europe for nearly the next millen-
nium (Patch 1935). The five books of De institutione musica cover harmonics, pro-
portions (Principle 1), semitones, and scales (Principle 2), and ends with a review of
harmonics as seen particularly through Ptolemy’s eyes. Boethius’s seminal work
provides a foundation for the compositional rules of his time (Boethius 1967). His
prescriptive formulae allow for many computational translations (see Figure 1.2).
Centuries passed before Hucbald (840–930), in his De harmonica institutione,
described the nomenclature necessary to realize many of Boethius’s theories.
Whereas Boethius’s treatise is primarily theoretical, Hucbald’s concentrates on the
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 9

BACKGROUND 9

FIGURE 1.1 From Aristoxenus, The Harmonics of Aristoxenus, trans. Henry S. Macran
(New York: Oxford University Press, 1902), 249.
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 10

10 HIDDEN STRUCTURE

FIGURE 1.2 A woodcut of Boethius with an instrument designed for methodologi-

cally deriving tunings (an algorithm).

practical. De harmonica institutione assumes its readers’ knowledge of Gregorian

chant—see the example in Figure 1.3—and thus is directed to musicians and not to
laymen (Weakland 1956).
The Musica enchiriadis, originally attributed to Hucbald but now considered
anonymous, appeared concomitantly with De harmonica institutione and provided
the first known instruction in organum (two-voice parallel-motion music, an exam-
ple of which appears in Figure 1.4; Principle 1 and Principle 2) and thus demarcated
a major turning point from monophony into an early type of polyphony. Like
Hucbald’s De harmonica institutione, the Musica enchiriadis had a completely practi-
cal intent as well as an algorithmic perspective in the sense of providing clear rules
for composition and analysis:
For example, he [Hucbald] (like Musica enchiriadis) uses no numerical interval
ratios or monochord division. Instead, the scale, and the intervals that structure
it, are taught empirically by means of concrete examples drawn from the plain-
chant melodies and intonation formulas of the cantus tradition, demonstrating by
direct experience the connection between the two. This characteristically Car-
olingian pragmatic approach is evident throughout in Hucbald’s continual cita-
tion of specific chant melodies to exemplify theoretical concepts. The concepts
themselves, however, are adapted from late Roman writings on the ancient ars
musica, especially Boethius’s De institutione musica. (Cohen 2002, 318)
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 11

BACKGROUND 11

FIGURE 1.3 An example of Gregorian chant, Parce Domine, notated in neumes; the
second staff is a continuation of the first.

& ww ww ww ww
ww ww ww ww ww ww ww
Sit glor - ri - a Do - mi - ni in sae - cu - la

From the booklet edited by Gerald Abraham

accompanying The History of Music in Sound
(booklet published by Oxford University press);
examples reprinted with permission of the publisher

FIGURE 1.4 An example of medieval organum.

About this same time, Guido of Arezzo (995–1050) completed his Micrologus
(1026), a work influenced by both Boethius and Hucbald as well as many of the
theorists that came before them. The Micrologus was one of the most significant
treatises on medieval music. Written in order to train the choir of the Arezzo Cathe-
dral, Micrologus covers intervals (Principle 1), scales (Principle 2), species conso-
nance, and the proper division of the monochord, all in great detail. The book is
rich in insight, but limited by its focus—training vocalists. Guido developed clear
descriptions of phrase structure and rhythmic meanings of neumes, among other
things:

Three brilliant pedagogical ideas have traditionally been attributed to Guido,

earning him his honored place in the history of music pedagogy: staff notation,
the system of hexachords, and his “classroom visual aid” for sight-singing
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 12

12 HIDDEN STRUCTURE

performance, the “Guidonian Hand” . . . these three innovations are so towering,

that it is less often noted that the “Micrologus,” besides being in effect an early
sight-singing manual, is also one of the very first in another long line of music-
pedagogical genres: the treatise on composition. (Wason 2002, 48–49)

The Guidonian hand, a visual aid for locating the semitones in the central part of
the gamut (see Figure 1.5), serves as a kind of algorithm in itself, a simple organiza-
tion of rules for memorization. Each portion of the Guidonian hand represents a
specific pitch (in solfège, of which this is one of the first known instances) within
the hexachord. The hand here spans nearly three octaves. Conductors or instruc-
tors would point to parts of their hands to indicate a sequence of pitches that the
choir or students would then sing in proper order. More than any of the treatises
discussed so far, Guido’s Micrologus represents a perfect historical model for how
theories and algorithms from any period can transfer to computation (Crocker
1958).
Guido’s important contributions influenced the work of Johannes de Garlandia
(1195–1272). Garlandia’s De mensurabili musica proposed a new theory of conso-
nances (Principle 1), dividing them into perfect (unisons and octaves) and imperfect
(major and minor thirds) types, with fourths and fifths relegated to intermediate sta-
tus. Garlandia classified dissonances into similar categories: perfect (minor seconds,
tritones, and major sevenths), imperfect (major sixths and minor sevenths), and
intermediate (major seconds and minor sixths). Garlandia also defined classes of
organum, pitches, ligatures, and many other aspects of thirteenth-century music,
making De mensurabili musica one of the most comprehensive and enumerative
texts of its time (Crocker 1962). De mensurabili musica, though often overly exhaus-
tive and confining in its limitations, provides clear algorithms for the practical use
of rules in analyzing the music of its day.
Jacques de Liège’s (1270–1340) Speculum musice (1340), an encyclopedic de-
scription of the music theory of the early fourteenth century, followed and built
upon the work of Guido and Garlandia. Liège classified music theory into five
categories: heavenly (celestis), cosmic (mundana), human (humana), instrumental
(sonorus), and analysis (practica). However, most of his treatise concerns sonorus
and practica, which involved modes—Principle 1—and measured (metered) music.
Philippe de Vitry (1291–1361), especially in his Ars nova (1320), developed a series
of signs that represented the division of notes into various short durations:

Several of the texts which describe these new mensurations refer to the com-
poser Philippe de Vitry as the inventor of the new system. Whether this is true
or not, de Vitry was certainly a well-known advocate of the “new art” of musical
composition in the fourteenth century, of which mensural innovations play such
a prominent role. Indeed it is from the title of one of his treatises that the term
ars nova was taken to describe this new style. . . . (Berger 2002, 635)
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 13

BACKGROUND 13

FIGURE 1.5 The Guidonian hand, in which a specific note is assigned to each part
of the hand. By pointing to a part of the hand, the conductor can indi-
cate to a group of singers which note to sing.
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 14

14 HIDDEN STRUCTURE

Both Liège’s and de Vitry’s writings have clear mathematical (programmable) as-
pects, even though, like many medieval theorists, these authors often navigate
through religious and philosophical realms as well (Werner 1956).
Johannes Tinctoris (1436–1511) honed the work of Liège and de Vitry (especially
in his Liber de arte contrapuncti of 1477). Tinctoris convincingly demonstrated that
his theories of the uses and construction of dissonant suspensions (Principle 1)
could help reveal the intricacies of the counterpoint of his time. His Proportionale
musices (1473–1474) comprehensively describes temporal relationships based on
proportions. Although Tinctoris was not a particularly original thinker, his prolific
output and careful description of the rules that apply to the various forms of
fifteenth-century music make his work vital to the history of theory and provide a
clear model for rule-based algorithmic analysis. Figure 1.6 presents an example of
florid counterpoint—free rhythmic values in all voices—from Tinctoris’s era.
As sixteenth-century polyphony developed and flourished, the work of Gioseffo
Zarlino (1517–1590), particularly his Le istitutioni harmoniche (1558) and Dimonstra-
tioni harmoniche (1571); Henricus Glarean (1488–1563), notably his Dodecachordon
(1547); and Nicola Vicentino (1511–1572), especially his L’antica musica ridotta alla
moderna prattica (1555), helped to decipher the developing complex counterpoint
and introduced chromatics into the diatonic modes (Principle 2). Zarlino was par-
ticularly influential in describing just intonation (see Figure 1.7) and the proper use
of chromaticism. Many of the principles these music theorists espoused translate
well to algorithms, particularly the rules of modal counterpoint (see Cope 2004).
Building on the work of Zarlino, Glarean, and Vicentino, several late sixteenth-
and early seventeenth-century theorists (e.g., Thomas Morley in A Plaine and Easie
Introduction to Praticall Musicke [1597]; Thomas Campion in A New Way of Making
Fowre Parts in Counter-point [1618]; and Michael Praetorius in Syntagma musicum
[1618]) developed algorithms for both composition and performance. Christoph
Bernhard in his Musica autoschediastika (1601) and Musica poetica (1606) actually
developed a system of classifying musical style. His descriptions and naming of
several embellishments (Principle 1) and expressions in music were the first of
their kind (Boorman 1980):

Bernhard, working in North Germany in the decades just after the publication of
Kircher’s Musurgia universalis in 1650, made style the very foundation of his clas-
sificatory system. He retained, perhaps unconsciously, an underlying link to the
Burmeister tradition, in that he saw musical figures as ornamenting a plain, dia-
tonic musical style. But now, that plain style is explicitly identified as an actual
style in the real musical world. (McCreless 2002, 862)

In the early eighteenth century, Joseph Johann Fux (1660–1741) completed his
seminal Gradus ad Parnassum (1725), a book that survives to this day as a guide for
writing species counterpoint. Fux had studied the writings of Zarlino and others,
and his dialogue between master and pupil in Gradus lays out both the didactic and
the aesthetic rules for sixteenth-century counterpoint (Principles 1 and 2) in ways
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 15

BACKGROUND 15

À& C w ˙ ˙ ˙ ˙ œ œ ˙ ˙ ˙
˙ ˙ ˙ ˙
Et resurrexit
˙ ww w ẇ ˙ ww w
?C w ˙
à w

FIGURE 1.6 An example of three-voice counterpoint around the time of Tinctoris.

FIGURE 1.7 The division of the octave according to Zarlino.

02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 16

16 HIDDEN STRUCTURE

that make computational implementation not only possible but also practical (see
my description of the Gradus program in Cope 2004):

Whereas the rules of Cochlaeus, Zarlino, and Aaron deal with surface situations,
Fux’s four rules operate abstractly, at a high level of generality. True, these rules
do not cover all of the situations addressed by earlier theorists. For example, be-
ginning and end, proximate location, and so on, are left to later discussion in
specific contexts. In effect, the rules cover an infinitely wider range of situations.
Taking two classes of consonance and mapping them on to three types of move-
ment result in a regulation of great power and memorability. (Bent 2002, 560–561)

The four rules mentioned here refer to the four ways that perfect and imperfect
consonances can precede and follow one another. Documented evidence exists
that many composers of note (e.g., Beethoven, Brahms, and others) used Fux’s
Gradus for studying counterpoint. Figure 1.8 presents a manuscript page based
on this text, attributed by some to Beethoven, who, these scholars believe, tran-
scribed and embellished it for his student Archduke Rudolph.
Around this same time, Jean-Philippe Rameau (1683–1764), building on the theo-
retical studies of many of the theorists discussed thus far, described chords as
emanating from a single source pitch (root).

Rameau acknowledged that the inspiration for this breakthrough came from
Descartes’ method, which was to build a system of natural law on a self-evident
principle. In his Traité de l’harmonie réduite à ses principes naturels (1722)
Rameau identified this first principle as the first six divisions of the string; these
could be shown to generate all of the consonant and dissonant intervals and
chords as well as the rules for their interconnection. But it was first necessary to
recognise as an a priori fact that a note and its octave-replicates were identical.
From this ensued the principle of inversion. (Boorman 1980, 756)

Thus, Rameau laid the foundation for the now traditional functional analysis (Prin-
ciple 3) of tonal music, the fundamental bass—a foundation that has proven emi-
nently computational, as demonstrated in the next section.
Figure 1.9 presents two diagrams from Rameau’s treatise on harmony (see
Rameau 1971) describing (a) the major (perfect) triad and its inversions, and (b) the
dominant (fundamental) seventh chord and its inversions (Principle 3). Rameau
comments that “the largest triangle will contain the perfect chord, the source and
the root of the other chords; these others will be contained in the two smaller tri-
angles” (Rameau 1971, 41). The pitch names here refer to solfège scale degrees, and
the numbers refer to partials of the overtone series projected from the bass pitch
(root) of the chord. Thus, Do in Figure 1.9a is partial 4 of the C overtone series, two
octaves above the fundamental C, and La in Figure 1.9b is partial 20 of the F over-
tone series. Although Figures 1.5 and 1.9 are separated by more than six hundred
years, the algorithmic notation for both suggest at least one common perspective:
methods of automatic generation and analysis of music.
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 17

BACKGROUND 17

FIGURE 1.8 An exercise by Beethoven for his student Archduke Rudolph based on
Fux’s Gradus ad Parnassum.

Johann Philip Kirnberger (1721–1783) continued to develop many of Rameau’s

ideas, especially the triad and its inversions, and extended these ideas into a no-
tion of melodic function. He also divided dissonance into two categories: essential
(such as the dominant seventh, Principle 3) and incidental (such as the suspen-
sion). His many books on theory (especially Die Kunst des reinen Satzes in der Mu-
sick of 1774–1779) could almost be used in classrooms to this day, so closely do
they approximate current approaches to tonal analysis.
Kirnberger also invented the musikalisches Würfelspiele, or musical dice games,
algorithmic combinatoria that also interested C. P. E. Bach, Haydn, and Mozart
(Cope 1996). Figure 1.10 presents the first two pages of a published version of a
musikalisches Würfelspiel attributed to Mozart, with the two matrices representing
phrases of music and each column giving numerical choices of measures of music
(not shown).
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 18

18 HIDDEN STRUCTURE

a.

b.

FIGURE 1.9 Two diagrams from Rameau’s Treatise on Harmony describing (a) the
major (perfect) triad and its inversions, and (b) the dominant (funda-
mental) seventh chord and its inversions.

Like Kirnberger, Joseph Riepel developed musikalisches Würfelspiele, often in-

cluding such algorithmic exercises in his treatises (see Grundregeln zur Tonordnung
insgemein [1755] and Gründkliche Erklärung der Tonordnung insbesondere [1757]):
Especially noteworthy is the wide variety of ways in which he [Riepel] character-
izes the organization and content of phrases. He thus distinguishes them on the
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 19

BACKGROUND 19

a.

a.

FIGURE 1.10 (a) First published after Mozart’s death in 1793 by J. J. Hummel in
Berlin, this musikalisches Würfelspiel instruction page shows the first
two phrases of number selections (chosen by throws of the dice)
that were keyed to measures of music. (b) Also shown, an example
page of music by Haydn for this own musikalisches Würfelspiel.

basis of their rhythmic activity (a concern rarely addressed by eighteenth-

century theorists), their overall melodic contour, their underlying harmonic sup-
port, their degree of melodic closure, and their length in terms of measure num-
bers. (Caplin 2002, 671)

Riepel further proposed the creation of major works using combinatorial proce-
dures. Ratner comments that “Riepel proceeds along these lines as he works out
melodic combinations in the construction of minuets, concertos, and symphonies.
Within a given model he seeks to achieve optimum effects by substituting figures,
phrases, and cadences” (Ratner 1970, 351).
Influenced primarily by Rameau, Moritz Hauptmann (1792–1868) in his Die Natur
der Harmonik und Metrik of 1853 outlines a system of harmony based on logic.
Although adopting Rameau’s system of roots and inversion, Hauptmann developed
harmonic successions based on common tones rather than exclusively on root pro-
gression (Principle 3). He avoided relating his theories to the overtone series,
primarily due to its inclusion of both consonant and dissonant intervals, and also
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 20

20 HIDDEN STRUCTURE

b.
Menuet
Haydn

## 3 œ œ œ œ œ œ ‰ j
1

œœ
2 3
œ œ œ
4
œ. œœ œ
& 4 œ # œ œ œ œ œ . œ

? ## 3 œ œ œ Œ œ œ œ œ œ œ
4 œ œ œ ˙. œ œ œ
œ
œ œ
## 5
œ œ œ œ œ 6 7 œ ˙.
J
8

& ˙. bœ œ œ.
œ œœ œœ
? ## œ œ œ œ Œ ‰ œ œ
œ œ œ œ œ œ J œ œ
Ÿ œ œœœœœ
## œ
9
œ œ 10
œ œ œœœœ œ œœœ
11 12
œœœ
& œ œ œ œ # œœ …
œ œ œ œ œ
? ## œ œ œ œ œ œ œ œ Œ
œ œ œ œ
# # œœ .. œœ œœ j œ œœœœ
˙.
13 14 15 16

& … J œœœœœ œ ˙
J
œ œ œœ œœ
? ## Œ œ Œ œ œ #œ œ œ œ œ œ œ œ œ
œ œ
3 3 3

FIGURE 1.10 continued

02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 21

BACKGROUND 21

because the series is infinite: “In Hauptmann’s dualistic model, there are three
‘functions’ assigned to pitches that constitute major and minor triads (or as we
will call the, following Hauptmann, ‘klangs’): unity (Einheit); duality or opposition
(Zweiheit); union (Verbindung)” (Klumpenhouwer 2002, 460). Hauptmann also ap-
proached rhythm and meter in innovative ways for his time. His overall analytical
system represents one of the more highly computable and algorithmic approaches
to musical analysis of the nineteenth century.
Hugo Riemann (1849–1919) developed many ideas of previous theorists but
honed them in ways that allowed for different uses. For example, Riemann’s refine-
ment of antecedent-consequent phrase sets (see his System der musikalishchen
Rhythmik und Metrik [1903]) led to their adoption by the theorists of his time. Rie-
mann also provided broader definitions of tonic (T), dominant (D), and subdomi-
nant (S) functions (TDS) such that their applications in late romantic music made
clearer sense to those dedicated to continuing to analyze tonal function where oth-
ers had attempted to abandon it (Harrison 1994). Henry Klumpenhouwer notes that

the function labels S , T, and D representing subdominant, tonic, and dominant,

respectively . . . in Riemann’s conception of them . . . have both a dynamic (that
is, transformational) and topographical modality. The latter modality on its own
is not Riemann’s: he himself explicitly traces the origins of this concept of chord
function to the work of Fétis. In Riemann’s view functions also have a syntactic
aspect, since complete harmonic phrases must have the structure TSTDT. More-
over, the syntactical functions may be served not only by the primary klangs in a
tonality but also by the secondary klangs . . . that relate to the primary klangs.
(Klumpenhouwer 2002, 468)

Interestingly, the TDS approach has achieved a revival of sorts in recent years,
with many theorists applying it to analyzing functions in music hitherto considered
nonfunctional (Principle 3).
Heinrich Schenker’s (1868–1935) extraordinary contributions to music theory
and analysis (notably his Neue musikalisches Theorien und Phantasien, created be-
tween 1906 and 1935) proposed major new perspectives on tonal music (see
Schenker 1979). Relying on the work of many of the theorists previously discussed
here, Schenker created a reductive process that presents foreground (Vordergrund ),
middleground (Mittlegrund ), and background (Hintergrund ) layers, each with suc-
cessively more musical detail removed in order to reveal the music’s fundamental
structure (Salzer 1962). Schenker further defined a fundamental framework (Ursatz)
and melody (Urlinie—descending diatonic scales from the mediant, dominant, and
tonic at the octave). Many of Schenker’s ideas appear in the Experiments in Musical
Intelligence (Principle 4) computer program (see Cope 1991 and 1996). Figure 1.11
presents a sample Schenker analysis of a chorale prelude by J. S. Bach showing the
3̂–2̂–1̂ upper line and the basic I–V–I harmonic structure.
Paul Hindemith, in The Craft of Musical Composition (1942), details an overtone-
series approach to defining chord roots, key centers, and as models for composing
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 22

22 HIDDEN STRUCTURE

and analyzing harmonic tension (Principle 3). Although many of the theorists dis-
cussed up till now developed similar ideas, Hindemith’s approach is nonetheless
distinctive, if for no other reason than that he composed using his ideas as well as
theorized about them. (See the analyses of parts of his symphony Mathis der Maler
[1934] in Hindemith 1942.) Hindemith fought, with some success, for the universal
application of his acoustically based ideas to the works of other composers. His
principles form the basis of many of the computational ideas in Chapter 5.
The Schillinger System of Musical Composition (1946) by Joseph Schillinger repre-
sents a kind of benchmark for algorithmic music theory, analysis, and composition,
containing as it does descriptions of many arithmetic formulae and other easily
computable processes. Unfortunately, many of his ideas lack clarity, and his system
has been severely criticized, fairly or not, for its misuse of certain mathematical
terminology, axioms, and theorems. Nonetheless, Schillinger’s system holds many
interesting surprises, not the least of which is his approach to composite rhythm
(the rhythm of all moving lines simultaneously) as a source for timbre.
Twentieth-century theorists other than Schenker and Schillinger—notably
Rudolph Réti, Nicholas Ruwet, Jean-Jacques Nattiez, Leonard Meyer, Milton Babbitt,
Allen Forte, Fred Lerdahl, and others—will be discussed in the next section of this
chapter as well as in later chapters. Suffice it to say that most contemporary analy-
sis techniques, especially as they relate to twentieth- and twenty-first-century mu-
sic, include algorithmic processes that make them computationally inviting. Indeed,
many current theories of post-tonal music have current computer implementations.

A BRIEF SURVEY OF COMPUTATIONAL MUSIC ANALYSIS

A brief word concerning the evolution of computers here may help provide per-
spective and context for the program descriptions to follow. Charles Babbage
(1792–1871), often credited with creating the first computing machinery, recalled
his first thoughts on this subject:

One evening I was sitting in the rooms of the Analytical Society, at Cambridge,
my head leaning forward on the table in a kind of dreamy mood, with the table of
logarithms lying open before me. Another member, coming into the room, and
seeing me half asleep, called out, “Well, Babbage, what are you dreaming about?”
to which I replied, “I am thinking that all these tables (pointing to the logarithms)
might be calculated by machinery.” (Babbage 1864, 42)

Babbage’s first Difference Machine is shown in Figure 1.12 (1833). Babbage de-
scribed its purpose: “I considered that a machine to execute the mere isolated op-
erations of arithmetic, would be comparatively of little value, unless it were very
easily set to do its work, and unless it executed not only accurately but with great
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 23

BACKGROUND 23

3̂ 2̂ 1̂

# j
œ œ œ
P P
& cœ œ œ œ ˙ œ œ ˙ œ œ ˙
N P

?# c ˙ œP œ œP œ
N
œ œ œ œ œJ œ ˙
œ œ ˙
I V I

FIGURE 1.11 A sample Schenker analysis of Bach’s organ prelude Wenn wir in
höchsten Noten sein.

rapidity, whatever it was required to do.” The engine relies on notched wheels
whose teeth interlock in successive orders of ten. As cumbersome as this proto-
type is, certainly by today’s standards, it operated quite effectively in its time. The
central processors of contemporary computers, although exponentially smaller and
faster than Babbage’s original device, still use many of its same basic principles.
Ada Lovelace (1815–1852), a devoted associate of Babbage and the machine’s
first programmer, wrote in her notes:

Again, [the Analytical Engine] might act upon other things besides number, were
objects found whose mutual fundamental relations could be expressed by those
of the abstract science of operations, and which should be also susceptible of
adaptations to the action of the operating notation and mechanism of the engine.
. . . Supposing, for instance, that the fundamental relations of pitched sounds in
the science of harmony and of musical composition were susceptible of such ex-
pression and adaptations, the engine might compose elaborate and scientific
pieces of music of any degree of complexity or extent. (Lovelace 1843, 694)

Lovelace’s notes were published in 1843 in volume 3 of Richard Taylor’s Scientific

Memoirs, with the author’s name given as AAL. This passage may well be one of the
first, if not the first, to mention music and computers (analytical engines) in the
same breath.
Even with analytical engines and other such machines, Allen McHose, for over
thirty years the chairman of the theory department of the Eastman School of
Music, developed a statistical method for creating Bach chorales without such
assistance (see McHose 1947 and 1951). His method would now be easily adaptable
to computer software. McHose hand-counted the functions, inversions, and so on
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 24

24 HIDDEN STRUCTURE

FIGURE 1.12 Charles Babbage’s first Difference Machine (1833), a forerunner of the
modern-day computer.

in Bach chorales and ensured that students used these functions in proper propor-
tions when writing exercises in four-part harmony. Over the years McHose’s reputa-
tion by way of his books spread through American academia, and for a brief time
his approach became the de facto standard for analytic instruction at the college
level.
Speed and accuracy continue to be the most significant attributes of today’s
computers, which still, at their most fundamental level, only add, subtract, and
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 25

BACKGROUND 25

store numbers. However, the sophistication of the layer upon layer of complexity
added to this core has made computers nearly indispensable in today’s world.
Born almost exactly one hundred years after Babbage’s first dreams of computing
technology, Alan Turing (1912–1953) had similar prophetic visions. In 1937 he pub-
lished his first important paper describing his theory of the Turing machine. He sub-
sequently played a significant role in the development of computers in Britain and
became a strong supporter of artificial intelligence. Together with his friend David
Champernowne, who, interestingly, later developed computer composing programs
(Herik 2000; Hofstadter 1979), Turing invented one of the first chess-playing pro-
grams discussed earlier in this chapter.
Computer music analysis programs have existed in various forms for decades.
Most of these programs follow more or less standard processes for analyzing tonal
and post-tonal music; many are simple tools for making analytical tasks faster and
more accurate. A relative few of these programs have full-fledged analysis packages
capable of a variety of musical representations and—at least to some extent—are
able to evaluate many different musical parameters (melody, harmony, rhythm,
form, etc.).
As with the preceding brief history of algorithmic analysis, the following overview
of computational music analysis is incomplete and could easily expand to several
times its current size. I apologize to those whose programs I have omitted due to
space limitations as well as to those whose programs I have mentioned but the
descriptions of which are overly brief. With the availability of programs on the In-
ternet, software-accompanying books on CD-ROMs, and to some degree programs
available commercially, clearly one could devote an entire book to a proper discus-
sion of this topic. My coverage here represents at least a fair sampling of the his-
tory of and general types of programs available at this time.
I begin with computer music visualizations. Many current computer music pro-
grams have such ancillary visualizations that allow users to see some sort of image
usually generated in synchronicity with the sound produced. The resultant kaleido-
scopic images reveal the dynamics and occasionally the pitches of their sources
but provide little of use for serious analysis. In fact, most of these visualizations are
artificially enhanced to make the images more visually attractive and thus lack au-
thenticity. However, serious music visualizing began to develop around the same
time as computer graphics (the 1950s). Visualization can be a very useful technique
for comparing or otherwise generalizing the form of musical works. Images based
on tessitura, texture, register, dynamics, rhythm, or other musical parameters can
provide interesting and useful insights for music analysts. The most important as-
pect of such visualizations is that images can be captured and reproduced in
spaces small enough for the eye to grasp the entirety of a work or movement,
where more detailed representations require far more space and thus cannot be so
easily seen at once. Issues of balance, shape, direction, points of arrival, and so on
can often be easily seen in such visualizations, where they might otherwise be
hidden.
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 26

26 HIDDEN STRUCTURE

Straightforward temporal diagrams of (usually) a single parameter of music can

provide interesting and even valuable insights into music that other techniques
may not so easily reveal. Such revelations are particularly important when viewing
several such diagrams simultaneously, as shown, vertically aligned, in Figure 1.13.
Here, the highest pitch (top) is followed in succession by texture, duration, dy-
namic, and relation to beat. The contrapuntal image thus produced reveals comple-
mentary points and points of disagreement that can provide useful information for
analysts. However, one should not confuse such diagrams with actual analysis. Vi-
sual diagrams are useful tools for music analysts in that they provide information
in ways that make possible comparisons that may otherwise defy analysis. Never-
theless, these diagrams in no way actually analyze music—such analysis remains
for human users of computer programs to accomplish. Several visualization pro-
grams accompany this book on its CD-ROM and appear in many figures in later
chapters.
Although computer composition programs can generate music by way of mathe-
matics and many other processes (see Xenakis 1971), music analysis programs re-
quire that music be coded in some way that gives these programs access to its
various parameters. Thus, the earliest programs of real interest to music analysts
were those that presented logical music encoding methods. The MUSIC I–V series
of programs (1957–1969; see Mathews et al. 1969) made significant strides with mu-
sic encoding, and MUSIC V continues to act as a model for many music notation
programs that, owing to their numerical abstractions, offer possibilities for compu-
tational analysis. Other programs, such as MUSTRAN, developed by Jerome Wenker
beginning in 1962 (see Byrd 1970), include most of the common symbols for music
notation.
The Digital Alternative Representation of Music Scores (DARMS), developed in
1963 by Stefan Bauer-Mengelberg (see Bauer-Mengelberg 1970), continues today as
an important computer representation program based on the physical placement of
information on a page of music. DARMS code is extremely accurate and currently
appears in several dialects (Schüler 2000; Selfridge-Field 1997a). The Plaine and
Easy Code (Brook 1965), devised by Murray Gould around 1964, and ALMA (Gould
and Logemann 1970), created by Gould and George Logemann in the mid-1960s, are
also excellent early examples of music digital representation. Michael Kassler de-
veloped IML (Intermediary Musical Language; see Robison 1967) and MIR (Musical
Information Retrieval; ibid.) in the 1960s as well. The introduction of MIDI (Musical
Instrument Digital Interface) code in the mid-1980s helped create a common inter-
face between various electronic music instruments as well as links between elec-
tronic performance and digital storage (Schüler 2000; Selfridge-Field 1997a).
Although many of the above-named programs are idiosyncratic in that they
served particular needs of the time, they have all contributed to the more universal
availability of digital representations of music. ISMIR (the International Society for
Music Information Retrieval) currently offers enormous potential for the establish-
ment of a common protocol for music storage and retrieval for future computer
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 27

BACKGROUND 27

Highest Pitch

Texture

Duration

Dynamic

Relation to
Beat

FIGURE 1.13 The author, Horizons for Orchestra, in graphic format. Time moves left
to right.

music analysis. Combining hitherto large but differently configured databases with
a unified interface will furnish a vast storehouse of available music for analysis by
any of the processes described in this book. Indeed, a common data structure and
robust software residing atop such a structure can provide the perfect domain for
all types of analytical programs to study any style of music.
The mathematician Frederick Brooks used the Computation Laboratory at Har-
vard University in 1957 to analyze thirty-seven hymn tunes using first- through
eighth-order Markov chains (see Chapter 6 for more information on Markov
chains), thus becoming one of the first researchers to use computers to analyze
music (Brooks et al., 1957). Brooks then used these analyses to create new hymn
tunes. By many accounts, the resulting melodies were quite successful in emulating
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 28

28 HIDDEN STRUCTURE

the basic style of the original tunes, thus becoming one of the first attempts to
model as well as reverse-engineer music as a method for analysis (Principle 4).
Joseph Youngblood’s work (1958) also stands out among those who attempted
to use computational means prior to the wider acceptance and availability of com-
puters for studying music. Youngblood used first-order Markov chains (Principle 1)
in attempting to differentiate musical styles, particularly melodies by Schubert,
Schumann, and Mendelssohn. He then based style identification on redundancies in
terms of chromaticism and probabilities of scale-degree usage. Although Young-
blood’s results were not particularly persuasive, owing to his use of a small set of
musical works, his research bears noting for its early date and utilization of charac-
teristically computational analytical techniques.
Lejaren Hiller and Leonard Isaacson (Hiller and Isaacson 1959), in their landmark
1959 book Experimental Music, lay out several potential goals for computational mu-
sic analysis: (1) the use of the Monte Carlo method (a process using laws of chance
controlled by statistical norms) to experiment with and compare musical forms;
(2) the analysis of musical styles based on entropy; (3) the generating and cata-
loguing of tone rows as cantus firmi for counterpoint (Principle 1); and (4) the
analysis of musical timbres (see especially Hiller and Isaacson 1959, 165–170). Al-
though Experimental Music focuses primarily on music composition rather than
music analysis, much of their program descriptions, particularly those aimed at
information theory and statistics, can easily be translated to analysis. Reverse-
engineering the processes used to create the Illiac Suite (1956, named after the Il-
liac computer at the University of Illinois), for example, can prove enlightening
when analyzing this work (which appears complete in an appendix to Experimental
Music). Figure 1.14 presents Experiment 2 from the Illiac Suite.
Hiller and Isaacson also focus on modeling (Principle 4; see also Chapter 6 and 7
in Hidden Structure), where musical principles are tested for their veracity. Hiller
and Isaacson comment that

as a consequence of coding aspects of the problem as numerical information and

generating experimental results by means of a computer, a computer is made to
behave as a specialized, but unbiased composing apparatus existing in a com-
pletely isolated environment, subject only to the controls and information the
music analyst might wish to supply. In this application, a computer is an ideal
instrument by means of which analytical ideas can be tested. (Hiller and Isaac-
son 1959, 166)

Many of the other early attempts at the computer analysis of music (see Schüler
2000 for a complete discussion and detailed listing) have to do with information
theory—the mathematical and physical study of information flow in a system. As
applied to music, information theory involves the probabilities of repetition of mu-
sical patterns (Principle 1). If a musical pattern repeats many times, its probability
increases, with its “information” (see Chapter 2 and the Glossary for definition) de-
creasing by inverse proportion (also see Shannon and Weaver 1949 and Pierce 1980
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 29

BACKGROUND 29

˙ b˙ U
˙ b˙ #˙ ˙
& ˙ b˙ ˙ #˙
Violin I
˙ ˙
b˙ b˙ b˙
˙ b˙ ˙ b˙ ˙
b˙ nU˙
Violin II & ˙ n˙
#˙ U
˙
˙ ˙ ˙ #˙
#˙ b˙ #˙
Viola B ˙ ˙
˙
b˙ n˙ bU
˙
? b˙ ˙ ˙ ˙ ˙
Cello
b˙ ˙
˙ ˙
FIGURE 1.14 From Experiment 2 from Hiller and Isaacson, Illiac Suite, using the Il-
liac computer at the University of Illinois, ca. 1956.

for more details). As an example, during the 1960s Hiller collaborated with several
of his students (including Calvert Bean, Robert Baker, and Ramon Fuller) in the
computer analyses of sonata expositions by Mozart, Beethoven, Hindemith, and
Berg (Schüler 2000) using information theory. Hiller and his students calculated
their findings based on tempo and note densities. Another of the analyses under-
taken by Hiller (see Hiller 1964) revealed redundancies in Webern’s Symphony
Op. 21 using higher-order Markov chains to track and predict intervals (Principle 1).
Milton Babbitt’s work with musical set theory (see Babbitt 1955, 1960, 1961, and
1965), although not in itself actual computer music analysis, set the stage for an ex-
traordinary number of later programs based on his computational approach to
analyzing post-tonal music. Allen Forte’s many articles and particularly his book
The Structure of Atonal Music (1973) have continued to popularize this important
approach to post-tonal music analysis. A number of chapters in Hidden Structure
also involve set theory because of its value in comparing harmonic, melodic, and
harmonic/melodic pitch patterns (Principle 1) in search of understanding post-
tonal music (see also Forte 1993).
The 1960s continued with a series of notable experiments in computer music
analysis. Eric Regener’s design for his System for Analysis of Music (SAM; see
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 30

30 HIDDEN STRUCTURE

Regener 1967) includes an elaborate assembly-language program for a state-of-the-

art (at the time) IBM 7090 computer. Though SAM seems to have been centered
more on computational routines to define data types than on incorporating actual
musical analysis, the software clearly demonstrates a serious programming attempt
to stabilize what hitherto had been idiosyncratic efforts. The following year, Ray-
mond Erickson’s article on computer music analysis (Erickson 1968) summed up
this problem: “There are as yet no standards for the encoding of music, no gener-
ally available musical analysis programs (as there are, for example, statistical pro-
gram ‘packages’), and no comprehensive theoretical systems for computer-aided
analysis” (242). Although this and Erickson’s later (1969) article paint a more posi-
tive future for the field, they do little to deter his pessimism about the then-current
state of disarray in computer music analysis approaches.
Also in 1969, Stephan Kostka (1969, 1971) coded a series of programs to analyze
the style of Paul Hindemith’s string quartets. Kostka’s programs defined the roots
and classes of chords, definitions based to a degree on Hindemith’s own principles
as described in The Craft of Musical Composition (1942), discussed earlier in this
chapter and in more depth in Chapter 5. Despite only partial success in his har-
monic analysis, Kostka extended his research to include Hindemith’s melodic style.
Kostka’s results were mixed, depending as they do on pattern matching for fre-
quency of both literal and varied appearances (Principle 1). Ultimately, his at-
tempts were limited by the sheer numbers of punch cards—required at the time for
computer input—to represent even minimal musical data relating to musical style.
Harry Lincoln’s 1970 book The Computer and Music contains many important
articles on the computer analysis of music, including Youngblood (1970) on root
progressions (Principle 3), Gerald Lefkoff (1970) on twelve-tone rows, and John Lof-
stedt and Ian Morton (1970) on the frequency of chord roots (Principle 3). The
Computer and Music also contained one of the first articles on computer analysis of
non-Western (Javanese) music by Fredric Lieberman (1970; see also Steinbeck 1976;
Suchoff 1968). Owing possibly to the complexity of analyzing music of oral tradi-
tions as well as its origination as an academic field at roughly the same time com-
puters became available, ethnomusicological research has often depended on
computer processes for analyzing its musical data.
Nico Schüler quotes a particularly apt passage from Jan LaRue’s article:

May I recommend the computer to you as an instrument without human preju-

dices. It has its own prejudices, numerical and procedural. But these often act as
stimulants and correctives, as healthy balances and supplements to human atti-
tudes. With this new aid, the coming generation of musicologists should develop
a style analysis that is comprehensive rather than selective, broad rather than
personal, and rich in musical insight. (LaRue 1970, 197)

Given the few articles and books on the subject of computer music analysis in the
1960s and early 1970s, these remarks provided an incentive for others to follow in
future decades.
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 31

BACKGROUND 31

Even given these many historical computer programs and theoretical descrip-
tions of computer programs, computer music analysis research has remained sec-
ondary to more traditional paper analyses. As Nicholas Cook points out,

computational approaches to the study of music arose just as the idea of com-
paring large bodies of musical data—the kind of work to which computers are
ideally suited—became intellectually unfashionable. As a result, computational
methods have up to now played a more or less marginal role in the development
of the discipline. (Cook 2004, 103)

The 1970s, however, produced some very interesting ideas and computer pro-
grams for music analysis. Dorothy Gross (1975, 1980, and 1984) contributed signifi-
cantly to the development of computer analysis, particularly in the 1980s, defining
her projects rigorously and dividing them into logical categories such as grouping,
analysis, and interval counting programs (Principle 1). Gross then combined the re-
sults of these analyses in comprehensive ways so that her results had both theoret-
ical and applied value. Unfortunately, she used her computer programs to analyze
small samplings of musical data typically involving just one piece each by com-
posers such as Bach, Haydn, Chopin, and Dallapiccola. Fred Hofstetter (1973 and
1979) also made great strides in computer music analysis in the 1970s by describ-
ing procedures for identifying stylistic influences. His research, primarily using
compiled interval counts, proved especially valuable in revealing cultural musical
influences across various European political boundaries.
Otto Laske, better-known for his recent research in music cognition but nonethe-
less important for his publications in the 1970s, made inroads (see particularly
1972, 1973, and 1974) with what he calls a “generative theory of music” (1993). This
generative theory follows the relationships between the syntactic structure of mu-
sic (Principle 3) and its semantic representation—an early example of music cogni-
tion. Laske’s KEITH, for example, is a rule-based system that initiates analytical
discoveries by dividing music into three categories—what is said (linguistic), what
is heard (sonological), and what is understood (analysis). Laske has also postu-
lated several principles for cognitive musicology (1992) as well as generative gram-
mars and music (1993; see also Lerdahl and Jackendoff 1983).
The 1980s produced activity in the computational analysis of both tonal and
post-tonal music. For example, Bo Alphonce (1980) created programs that com-
puted the transformations of prime forms of all possible pitch-class sets, produced
software for the analysis of invariant sets and their subsets, and so on (Principle 1).
Interestingly, Alphonce’s research into computational grouping possibilities for
post-tonal and serial music produced much the same kind of prodigious output as
my own (see Cope 2005), including incredible numbers of both irrelevant and rele-
vant possibilities. Without question, however, the work of Alphonce, like that of
Gross, has had an immeasurable influence on those who followed.
Ann Blombach developed a computer program in 1982 that analyzed fifty
major-key Bach chorales to determine whether harmony or counterpoint is more
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 32

32 HIDDEN STRUCTURE

essential in their composition (Principle 1). Blombach used several parameters,

such as statistical frequencies of pitch, duration, patterns, and rhythm, among oth-
ers, in her analyses. Blombach’s results demonstrate a subtle balance between the
contrapuntal and harmonic aspects of the selected chorales. The Bach chorales—
staples of most collegiate music theory curriculums—have subsequently been used
for a wide variety of computational research, having proved their worth as a kind
of benchmark for tonal voice-leading (Blombach 1981).
Dean Simonton (1980) analyzed transitions within the first six notes of 5,046
classical music themes by ten composers in order to discover the musical charac-
teristics that tended to make them well-known (Principle 1). Simonton defined
“well-known” as those works that had the most recordings, most performances,
and/or most citations in the literature. He limited his studies to pitch alone (exclud-
ing rhythm and other parameters). Although Simonton attempted to correlate cre-
ativity and originality in these works as well, his research reached its limits and
ultimately has not stood the test of time well. Nonetheless, the mere fact that he
directly attempted to use computer analysis for such subjects as creativity and
originality make his research historically important.
One of the first programs to use Allen Forte’s approach to set theory (1973), the
Computer-Assisted Set Analysis Program (CASAP), was developed by Charles Rug-
giero and James Colman (1984). This program calculates prime forms, discovers
similarities, and returns complex relations of sets identified by users. The Contem-
porary Music Analysis Package (CMAP) created by Craig Harris and Alexander
Brinkman (Harris and Brinkman 1989) extends the CASAP program by Ruggiero and
Colman by including invariance, ordered and unordered interval-class vectors, sub-
sets, complements, adjacent interval vectors, and so on in its output. CMAP has
subsequently been expanded and adapted to several platforms by Peter Castine
(see Castine, Brinkman, and Harris 1990) and others (Principle 1).
Stephen Smoliar (1980), one of the first theorists to describe the transfer of Hein-
rich Schenker’s structural analysis techniques for tonal music to a computer pro-
gram, developed computational processes for the identification of the three
structural levels (foreground, middleground, and background) and even computa-
tional definitions of the Ursatz (fundamental structure) in tonal music (Principle 3).
Although some of Smoliar’s work remains descriptive and has not yet been imple-
mented, his detailed outlines of a full-scale program provide an excellent basis for
an algorithmic representation of Schenker’s ideas.
The MUSANA program developed by Nico Schüler and Dirk Uhrlandt (see
Schüler and Uhrlandt 1994) analyzes pitch, duration, and meter statistically for
mean, standard deviation, frequency correlation, autocorrelation, entropy, and so
on (Principle 1). The program makes comparative analyses of musical styles based
on personal, genre, and/or period. MUSANA, though somewhat limited in its scope
both musically (to classical music) and computationally (to statistics), furthers
many important analytical possibilities for research into musical style. Figure 1.15
shows two flowcharts, one of MUSANA as a whole (a), and the other the analysis
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 33

BACKGROUND 33

FIGURE 1.15 Flowcharts for MUSANA and its analysis module.

02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 34

34 HIDDEN STRUCTURE

FIGURE 1.15 continued

02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 35

BACKGROUND 35

portion of the MUSANA program (b). Flowcharts such as these represent algo-
rithms and demonstrate points of decision, data input, beginning and end points,
and computational activity.
John Schaffer (1994) created a program that allows users to define criteria dur-
ing the analytical process. The program consists of a relational database and allows
user interaction during analysis. Schaffer also incorporates a form of fuzzy logic in
his system so that matches of nonequal but related patterns are not overlooked
(Principle 1). Schaffer developed his program primarily for the pitch-class set
analysis of post-tonal music.
Mira Balaban has described a number of programs based on computational
processes designed to analyze music for its hierarchical structure (see especially
Balaban 1992). Balaban’s approach involves explicit and implicit descriptions of
music based on variable grouping sizes. Peter Desain and Henkjan Honing devel-
oped a program in the POCO software package primarily for analyzing and generat-
ing expression in music (Honing 1990). Their extensive papers on this subject
belong to a larger broad-based project named “Music, Mind, Machine.” Most inter-
esting about this research is its reliance on digital sound as data, rather digital
representations of music notation. Kemal Ebcioglu worked for several years devel-
oping his CHORAL (1992) system for harmonizing Bach chorales (Principle 3). His
work is particularly important because it uses a form of predicate calculus (sym-
bolic logic) to indicate the fundamental rules generating the music. Jamshed
Bharucha has used neural nets (see Chapter 7) to analyze harmony. One of his pro-
grams, MUSACT (Bharucha 1993), incorporates standard connectionist architecture
for modeling harmonic structures and then properly identifying them (Principle 3).
In the 1990s several very different research projects developed that had an im-
portant influence on the field of computer music analysis. For example, the Hum-
drum Toolkit, developed by David Huron (1995 and 1999), a powerful Unix-based
music information-retrieval computer program, searches for particular motives,
compares voice-leadings in various repertories, counts suspensions, defines and
catalogues harmonic progressions, analyzes dissonance in relation to metric posi-
tion, and performs other such operations (Principles 1, 2, and 3). Conceived as a
set of utilities rather than a single large program, Humdrum offers a command-line
approach to extracting particular data and statistics. The program requires music
in Humdrum format; currently available databases include traditional and folk mu-
sic, a limited number of popular songs, and a somewhat small—given the amount
available—number of more or less complete standard classical repertories (e.g.,
Bach’s Brandenburg concerti, all of the Beethoven symphonies, Bartók’s complete
Mikrokosmos, the Corelli trio sonatas, the Debussy piano preludes, Mozart’s com-
plete string quartets, etc.). Written in cross-platform Unix, Humdrum Toolkit is
available as freeware for a wide variety of computer platforms.
In some ways, the Humdrum Toolkit is more suited to locating, identifying,
counting, and classifying provided instances of musical or textual (lyrics) data than
to actually analyzing these data. This is not to say the program does not provide
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 36

36 HIDDEN STRUCTURE

information that can prove invaluable for analysis. However, most of the analysis
must take place elsewhere (e.g., with users or another computer program). Hum-
drum Toolkit serves as an important resource for both musicologists and music
theorists requiring reliable statistical data for their research, and as such it repre-
sents a valuable tool for computer music analysis. Figure 1.16 presents ten com-
mands from the many possibilities to give readers a sense of the breadth of the
program.
In 2001 David Temperley and Daniel Sleator created the Melisma Music Analyzer,
which derives meter, voicings, phrases, key, and chord function from tonal music.
Melisma (for Modular Event-List-Input System for Music Analysis) consists of sepa-
rate modules that can be detached or added as needed during analysis. The pro-
gram runs in a Unix environment with source code available from the authors
(Temperley and Sleator 1999). In 2002 Guerino Mazzola and Oliver Zahorka devel-
oped a program called RUBATO that is likewise modular. RUBATO analyzes, com-
poses, and/or performs music, and the modular architecture of the program allows
for third-party developers to create modules for specialized tasks. The creators of
RUBATO pride themselves on its scientific basis, though many of the processes
they use have traditional musical roots. RUBATO’s graphics are quite elegant and
helpful during the analytical process, and the weighting system produces interest-
ing results. However, the output typically requires a further analysis that Mazzola
calls an “analysis of analysis” (Mazzola 2002, 855).
The Tonalities program, created by Anthony Pople at the University of Notting-
ham, was designed for analysis of Western tonal music of the late nineteenth and
early twentieth centuries. Tonalities allows users to analyze passages of music in
terms of differing keys that may be detailed from a range of supplied options. Inter-
estingly, Tonalities is an add-on to Microsoft Excel spreadsheet software. Excel
has the advantage of being a professional, well-supported business program and
the disadvantage of requiring that all data have a specialized format. Tonalities
therefore requires that users be conversant both with Excel and with certain types
of rhythmic notation such as those explained in The Rhythmic Structure of Music, by
Cooper and Meyer (1960), and A Generative Theory of Tonal Music, by Lerdahl and
Jackendoff (1983).
The Music Theory Workbench (MTW), developed by Heinrich Taube at the Uni-
versity of Illinois, is a powerful program for analyzing tonal music, particularly the
371 Bach chorales as collected by Riemenschneider (1941). MTW takes a short mu-
sical work and outputs an analysis in the form of an annotated graphic score like
the one shown in Figure 1.17. The analysis includes chord and inversion classifica-
tion (Principle 1), nonharmonic tone determination, primary and secondary tonal
center identification (Principle 2), and a functional harmonic analysis (Principle 3).
The encoded database of Bach chorales consists of automatically translated MIDI
files downloaded from the Web.
The analysis process for the MTW program follows a straightforward series of
steps. First, information is parsed from an ASCII (American Standard Code for Infor-
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 37

BACKGROUND 37

cents translate selected Humdrum pitch-related representations to cents

diss calculate the degree of sensory dissonance for successive spectra (**
fin2hum translate Finale files to Humdrum
hint determine harmonic intervals between concurrent pitches for Humdru
key estimate the key for a Humdrum passage
mint determine melodic intervals between successive pitches for Humdrum
nf determine normal form for successive vertical sonorities in Humdrum
pattern exhaustively locate and count user-defined patterns in a Humdrum in
record record live MIDI input in Humdrum **MIDI data format
solfg translate selected Humdrum pitch-related representations to French s

FIGURE 1.16 Ten commands found in Humdrum.

mation Interchange) score into a time line of vertical sonorities that can be identi-
fied as triads or seventh chords. These sonorities are then classified as to type
(major, minor, diminished, augmented, seventh chord, etc.) and inversion. Sonori-
ties that are not classifiable as chords are subjected to nonharmonic tone analysis,
with the resulting nonharmonic tone or tones identified as to their type (passing,
neighbor, suspension, etc.). MTW uses harmonic rhythm to determine points to as-
sign harmonic function to two or more sonorities with—possibly—different roots.
Given a tonal center confirmation, the analysis proceeds with key identification and
produces a functional harmonic analysis of all the sonorities in the time line. Pivot
chords between two keys are identified when a single sonority serves two or more
functions in its harmonic context.
Many current algorithmic music composition programs (also called computer-
assisted composition programs) offer valuable analysis tools. For example, Open-
Music (OM) from IRCAM (Institut de Recherche et Coordination Acoustique/
Musique) in Paris, Elody from Grame in Lyon, France, Symbolic Composer, Alice
(ALgorithmic Interactive Composing Environment), Common Music from CCRMA
(Center for Research in Music and Acoustics) at Stanford, and many more have
analysis components. Some (particularly OM and Alice) offer users the ability to al-
ter the program to increase its analytical capabilities. Figure 1.18 presents an
overview of OM’s graphic interface and various forms of music notation.
Many other computer analysis programs exist than the ones I have mentioned
here, including those available on the Internet, through university research centers,
and, in some cases, commercially.
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 38

38 HIDDEN STRUCTURE

FIGURE 1.17 Bach Chorale 002 from the Music Theory Workbench, version 0.1
(http://pinhead.music.uiuc.edu/~hkt/mtw/pdf/), by Heinrich Taube (ac-
cessed 24 July 2007).
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 39

BACKGROUND 39

FIGURE 1.18 An example of the OpenMusic visual programming environment of

music for composition and analysis.

MUSICAL EXAMPLES
In each chapter of this book I will present its analytical process as applied to the
third piece of Stravinsky’s Three Pieces for String Quartet (1914) in order to pro-
vide readers with a sense of continuity and to involve them in a somewhat more
complete analysis of one work. The beginning of this music appears in Figure 1.19.
The Three Pieces for String Quartet (1914) were completed during the period in
which Stravinsky was occupied with composing his larger work Les noces (1914–
1916). Stravinsky later twice returned to these pieces to form the first three of his
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 40

40 HIDDEN STRUCTURE

III.
À & 45 46 œ b œ
Stravinsky (1914)
45
V1 b˙ œ œ œ œ bœ ˙ œ œ- ˙ œ
π
V2 & 45 46 45
b˙ œ œ œ œ bœ ˙ œ œ bœ œ- ˙ œ
π
B 45 œ nœ bœ œ 46 œ 45
Va
b˙ œ b˙ nœ œ b œ- ˙ nœ
π
Vc à ? 45 π˙ bœ œ œ bœ œ ˙ œ 46 b œ œ œ- ˙ œ 45

À & 45
4
≤ ≤ ≤ ≤
C œ œ œ œ
3
ϲ w
œ bœ œ œ œ- ˙. ˙ F p
3
≤ ≤ ≤ ≤ ≤
& 45 C œ œ œ œ œ w
œ bœ œ œ œ- ˙. ˙ F p
3
≤ ≤ ≤ ≤ œ≤ w
B 45 C œ œ œ œ
bœ bœ œ n œ b œ- b˙. ˙
F p
≤ ≤ ≤ ≤ b œ≤
3

à ? 45 œ œ bœ œ œ- ˙. ˙ C #œ œ œ œ #w
F p
FIGURE 1.19 The first few measures of Stravinsky, Three Pieces for String Quartet
(1914), 3rd movement.

Four Studies (1928 and 1952, the fourth being Madrid ) for orchestra. Because this
chapter does not focus on a single analytical approach but rather presents a brief
history of algorithmic and computer analysis, I here present this music with several
historical methods of analysis that will not otherwise be covered in this book.
These methods include tonal, polytonal, and formal techniques.
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 41

BACKGROUND 41

Tonally analyzing the music in Figure 1.19 requires a liberal definition of tonal
harmony, for often chords apparently built in thirds require imaginary or altered
members. The first chord, for example, could be classified as an E-flat seventh
chord in third inversion with a missing third. The second chord could then trans-
late to a D seventh chord, also in third inversion (D–F–A–C-sharp, respelled). The
third chord then poses problems, requiring a creative solution—possibly a C major
altered ninth chord (C–E–B-flat–D-sharp). As the phrase continues, such chordal
analysis becomes less and less clear, particularly when one attempts to discern the
key as well as the resulting functions.
Polytonality would seem to offer more opportunity for logical analysis of this
piece. The opening phrase in Figure 1.19 contains all pitches but A-flat. The only
two keys that make sense for a polytonal analysis of this opening passage that omit
A-flat are B-flat major and D major, the initiating notes that represent the alto and
bass voices beginning in Figure 1.19. Unfortunately, although the upper two voices
and many notes from the viola and cello fit nicely into B-flat major, D major does
not appear likely in any of the music here. The more likely analysis suggests C ma-
jor and B-flat major with the lone G-flat (F-sharp) acting as a chromatic nonhar-
monic tone.
Formally, the third of Stravinsky’s Three Pieces for String Quartet resembles a
rondo with three returning fragments, of which Figure 1.19 presents two. The music
throughout this piece repeats with different rhythms, meters, accents, and so on
characteristic of his style in this and other periods of his compositional life. The
enigmatic solo viola ending the piece (not shown here) suggests that Stravinsky
intended a less-than-convincing set of key relationships. Yet through his use of dy-
namics, phrase lengths, and other musical parameters, Stravinsky achieves a quite
convincing movement. As we shall see in later chapters, this work provides many
surprises beyond this simple narrative analysis.

PROGRAM DESCRIPTION
The computer code accompanying this book on CD-ROM exemplifies a number of
the analytical processes described in this chapter. For example, the Visualize code
presents a kind of seismic graph display of several parameters of the music in its
database. The graphs this software creates (the Macintosh Common Lisp version
only) produces images like that shown in Figure 1.13, with severe fluctuations indi-
cating highly variable information content and stable horizontal lines representing
unchanging content. Although these graphs are very useful for visually accessing
and overviewing large segments of music, they do not represent actual musical
analysis. In other words, although not nearly as precise as more traditional musical
scores, the graphs represent a useful alternative by presenting the same music in
far less space.
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 42

42 HIDDEN STRUCTURE

The Sets Analysis code on the CD-ROM accompanying this book represents a
simple example of the kind of program available from many different sources on
the Internet in various forms (Java, etc.). The Macintosh version of this program on
the CD-ROM presents a straightforward “Identify Set” window when choosing
Lookup from the menu at startup (see Figure 1.20). The program then provides the

FIGURE 1.20 Two screens from the Sets and Vectors program on the CD-ROM that
accompanies this book.
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 43

BACKGROUND 43

pitch-class set name, vector, lookup process, and any Z-related pairs for input sets
from three to nine pitches, as described in Forte (1973) and in the introduction to
set theory at the beginning of Chapter 3 of this book. The Common Lisp version of
this program works in the same way, but without menu or window shortcuts. Infor-
mation about how to access the various forms of sets appears in the header docu-
mentation in the code file.

CONCLUSIONS
This chapter has briefly covered an enormous time span—from Pythagoras to the
current day, almost 2,500 years of music theory and analysis. Although it obviously
provides a brief overview rather than a detailed history, this background does es-
tablish the context for the more experimental processes described in the chapters
to come. For example, Chapter 2 will explore the potentials of information theory
and one of its subfields, algorithmic information theory. These fields, both sub-
sumed under the general umbrella of discrete mathematics, have enormous poten-
tial for revealing important information about all disciplines, not the least of which
is music.
02_DAS_Chap1_pp1-44 1/29/09 1:31 PM Page 44
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 45

TWO
Lisp, Algorithmic
Information Theory,
and Music

Estimates of the relative degrees of order and disorder of different samples of

music or different sections of given musical structures could be attempted. This
is suggested since entropy seems to be a more useful variable than less well-
defined concepts such as “harmonic tension.”
Lejaren Hiller and Leonard Isaacson, Experimental Music (1959), 167

In the epigraph at the beginning of this book I quote Saint Augustine’s statement
that “Beauty is the splendor of order.” This notion of order—also mentioned by
Hiller and Isaacson in the epigraph to this chapter—represents a particularly im-
portant perspective in computer-assisted analysis, for all of the techniques I de-
scribe in this book depend, in one way or another, on the order present in music.
For example, this chapter explores the very essence of order and its contradiction,
chaos, by describing a means—computer programming—and one of the fundamen-
tal types of computational analysis of order and chaos: information theory. Creating
computer programs for analysis typically requires a computer programming lan-
guage. I begin this chapter by describing a computer language called Lisp. All of the
software accompanying this book on CD-ROM is written in Lisp, and much of this
software references information theory in one form or another. Understanding a
few of the basics of Lisp will help readers to negotiate their way through the source
code and to run the software presented here. I continue this chapter by introducing
algorithmic information theory (AIT), a subset of information theory, that analyzes
data for the amount of information (non-redundant data) it possesses. Musical algo-
rithmic information theory (MAIT) applies AIT to musical works to determine their
inherent information content for comparison to other works and as a revelation of
the dynamic music information content (DMAIT). As such, this chapter concerns
Principle 1 (music as pattern), as discussed in Chapter 1.

LISP
Lisp (a malformed acronym for List Processing) is one of many programming lan-
guages designed to create computer programs. Conceived and first implemented in

45
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 46

46 HIDDEN STRUCTURE

the late 1950s by John McCarthy and his associates at MIT, Lisp has had many
devotees, especially in the field of artificial intelligence. Originally available in many
flavors, the most often used form of Lisp today is Common Lisp, available on every
standard computer platform and uniform in use through a standard reference man-
ual (Steele 1990). Some commercial versions of Common Lisp come with ancillary
code to aid users in creating their own applications. These versions of Common
Lisp require a second, platform-dependent manual.
Booting—initializing—Common Lisp typically produces a Listener window ready
for command-line input. This input in the Listener window is evaluated—interpreted—
by Lisp whenever users follow input with the return key. Code may also be typed
into a text window. However, code typed into a text window will not be interpreted
until users explicitly invoke it. Invoking interpretation usually results from typing
some combination of implementation-dependent keyboard and/or mouse commands.
Although many important types of information exist in Common Lisp, the two
most important types are functions and data. Functions are operatives that create
or alter data in useful ways. Data represents information upon which functions op-
erate. In Lisp, functions are typically used individually or combined with other
functions to produce more complex processes and output. User-defined functions
can be included alongside built-in functions; the results are—as the axiom goes—
often greater than the sum of their parts.
In order to keep complicated processes in Common Lisp clear, the language re-
quires parentheses, in part to distinguish between data and functions, as well as to
clarify different functionally initiated actions. In other words, when entering func-
tions or data into a Listener window, users must partition their boundaries and fir-
ing order with parentheses. Furthermore, information entered into the Listener
window will be evaluated by Lisp unless preceded by a single quote (’). The follow-
ing code provides a simple example of these concepts, where the “?” represents a
typical prompt provided by the form of Common Lisp used:
? (first '(1 2 3))
Note that the data here—a simple list of numbers—is surrounded by parentheses
to indicate its inclusiveness, with the single quote identifying the data as a list not
to be evaluated (although Lisp can evaluate numbers, it cannot evaluate lists of
numbers) until the function first has had an opportunity to act on that list of
data. The data that functions act upon is also called that function’s arguments. The
Lisp function first in the above code returns the first element of its argument,
which must be a list. Therefore, when users hit the return key after the entry of the
above code, Lisp produces the number 1. Although these procedures may at first
seem arcane, they make more sense with practice.
In effect, Common Lisp evaluates the function first, with which it is familiar,
and then applies that function to the data with which it is not familiar. Functions al-
ways appear to the left in a parenthetical representation, with data to the right.
Thus, a simple mathematical expression such as
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 47

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 47

? (* 1 2)
equates to “1 times 2,” the result of which will be 2 when, again, users hit the re-
turn key. Neither the number 1 nor the number 2 requires single quotes here be-
cause Lisp recognizes individual numbers.
Lisp operations can also be nested to produce more interesting results. The
code, for example,
? (first (rest '(1 2 3)))
produces the number 2, because Lisp first evaluates the function rest (meaning all
but the first), which returns the rest of its argument ‘(2 3), and then evaluates the
function first, which returns the first element of the result of the rest operation.
This nesting of parentheses can produce some very complicated-looking code.
However, one can always unwrap the parentheses to see how the functions operate
on the output of other functions by separating the sublists, beginning with the
most deeply nested one. Typically, this means that the functional operations are
read from right to left rather than vice versa. Fortunately, Lisp provides another
primitive—a built-in function that accomplishes very simple things—in this case
called second, that produces the same result but using just one function call, as in
? (second '(1 2 3))
Data can also be nested in sublists according to whatever plan the programmer
feels will reveal the information best. For example, I often use sublisted data for
musical notes in lists such as
((0 60 1000 1 127)(1000 62 1000 1 127)(2000 64 1000 1 127)
(3000 65 1000 1 127) . . .
for the first four notes of a C-major scale played one after the other. Here, the en-
tries in each sublist pertain to ontime (in thousandths of a second), pitch (where
middle C is 60 and increments in both directions refer to half steps up or down),
duration (in thousandths of a second), channel (1–16), and loudness (0–127, with 0
representing silence). Because very large works contain very long lists of note-
events (i.e., thousands of separate notes), these note-event lists can often become
quite difficult to read. Lisp therefore provides a method of assigning symbols to
represent data. By using the Common Lisp function setf, for example, all the notes
of a musical work can be placed in a kind of container (variable), the name of
which users can then call without having to write out or read all of the separate
events and their contents. For example, the code
? (setf musical-work '((0 60 1000 1 127)(0 62 1000 1 127)
(0 64 1000 1 127)(0 65 1000 1 127)
. . . .
places all of the data that follows its name into the musical-work container. Typing
? (first musical-work)
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 48

48 HIDDEN STRUCTURE

and pressing the return key then produces the note-event (0 60 1000 1 127). Note
that one need not use the single quote here because musical-work has already
been defined, and thus Common Lisp recognizes and can interpret its meaning.
Another Common Lisp primitive, cons, creates lists from—typically—an atom
(number or letter) and a list, as in
? (cons 1 '(2 3))
which returns the list (1 2 3) following the return key. Like the function setf, de-
scribed previously, cons requires two arguments, “constructing” (the word from
which cons derives) a new list combined from its two arguments. Unlike setf,
cons does not place the data or results in a container (variable).
Some functions in Common Lisp—called conditionals—test data for certain at-
tributes. For example, the function if tests its first argument and returns its sec-
ond argument if it proves true, or its third argument if the first argument proves
false. The following code provides a simple example:
? (if (numberp 1) t nil)
where the first argument to the function if acts as the test here (if 1 is a number),
the second argument t indicates that true should be returned if the first argument
is true, and the third argument indicates that nil (or false) should be returned if
the first argument is not true. This combination is often termed an if-then-else
clause. In this case, hitting the return key will produce t because 1 is a number.
The function numberp is called a predicate (a function that typically returns true or
false as applied to a single argument).
Lisp is widely known for its use of recursion, functions that call themselves dur-
ing execution with lesser and lesser elements of their arbitrary-length list argu-
ments. Using recursion therefore means that users need not know the actual length
of list arguments in order to access them, a great advantage when manipulating
vast amounts of information. The following code describes a user-defined recursive
function called add-one that adds 1 to each of the elements in its list argument.
Note that the function defun defines functions in Common Lisp.
(defun add-one (list-of-numbers)
(if (null list-of-numbers) ()
(cons (+ 1 (first list-of-numbers))
(add-one (rest list-of-numbers)))))
The interpretation of this function can be described in the following way:
(line 1) the first line of code names the function and declares the vari-
ables that will represent its argument(s) in a list (in this case only one
named argument exists). Function and argument variables should be
named according to what they do or represent in order to render the
code as readable as possible;
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 49

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 49

(line 2) the function if in this case tests the argument list-of-

numbers to ascertain if it contains data, and if empty it then returns
an empty list (second line of code);
(line 3) if (else) the list still contains entries, the code constructs a list
consisting of 1 added to the first of add-one‘s list argument and (line
4) the result of add-one’s continued application to the remainder of
the list.

The terminating code (or second line) in the definition of add-one must occur
first, because if the function add-one were to attempt to add 1 to an empty list, it
would fail (because (), read also as nil, does not contain a number in its final re-
cursive call).
Lisp novices often find the recursion in functions such as add-one difficult to
comprehend. There are several ways to explain recursion beyond the code itself.
Possibly the clearest explanation involves nesting the application of actions that
take place on a sample argument. Using ‘(1 2 3) as a simple example, add-one first
tests its argument to ensure that it contains numbers. Our argument at this point
consists of ‘(1 2 3), and so it defaults to lines 3 and 4 of the add-one definition.
The cons function interprets its first argument as 1 plus the first number of ‘(1 2
3), computes this as 2, and proceeds as in
(cons (+ 1 1) (cons (+ 1 2) (cons (+ 1 3) ())))
which when read from right to left produces the following reduction
(cons 2 (cons 3 (cons 4 ())))
When computed, add-one produces the list (2 3 4), exactly the outcome desired.
The first call to cons (on the right here) returns the list (4), the second call to
cons returns the list (3 4), and the final call to cons produces the list (2 3 4). At
this point, novices usually ask why not use the above code rather than creating a
complicated recursive function called add-one to perform it. The answer, of
course, is that recursive functions in Lisp can operate on arbitrary-length lists,
making add-one capable of adding 1 to all of the elements of lists of any length—
lists that would otherwise require the typing of immense numbers of numbers and
subsequent calls to cons.
Although this brief description of Common Lisp fails to describe many of the
thousands of built-in functions and most of what programmers can really accom-
plish with the language, readers should be able to load, use, and at least partially
understand the code that accompanies this book if they carefully read the descrip-
tions that appear near the top of each file. I will also occasionally place Lisp code
in the text of this book to clarify points made and reinforce the processes de-
scribed here. For a more detailed description of Lisp, see Cope 1991, chapter 3;
Graham 1995; Steele 1990; Touretzky 1990; and Wilensky 1986.
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 50

50 HIDDEN STRUCTURE

DEFINITIONS
Because this chapter covers many subjects dependent on the concept of informa-
tion, I begin with a clear definition of that word in the context used here. “Informa-
tion,” when it pertains to information theory, means the non-redundant portion of
any string of numbers. As example, the floating point representation of the fraction
1/ , or 0.333333333333 . . . , holds very little information, because the redundant rep-
3
etition of 3s extends infinitely and can be expressed more succinctly as a fraction.
In contrast, the number ␲, or 3.141592653589793 . . . , with its apparently endless
non-patterned and non-repeating sequence of numbers, contains a very high level
of information. Of course, being an infinite sequence of the arrangements of the
numbers 0 through 9 means that, when exploded to individual digits, ␲ does con-
tain patterns and number repetitions (as seen here in Figure 2.1, the first two thou-

3.1415926535897932384626433832795028841971693993751058209749445923078164062
862089986280348253421170679821480865132823066470938446095505822317253594081
284811174502841027019385211055596446229489549303819644288109756659334461284
756482337867831652712019091456485669234603486104543266482133936072602491412
737245870066063155881748815209209628292540917153643678925903600113305305488
204665213841469519415116094330572703657595919530921861173819326117931051185
480744623799627495673518857527248912279381830119491298336733624406566430860
213949463952247371907021798609437027705392171762931767523846748184676694051
320005681271452635608277857713427577896091736371787214684409012249534301465
495853710507922796892589235420199561121290219608640344181598136297747713099
605187072113499999983729780499510597317328160963185950244594553469083026425
223082533446850352619311881710100031378387528865875332083814206171776691473
035982534904287554687311595628638823537875937519577818577805321712268066130
019278766111959092164201989380952572010654858632788659361533818279682303019
520353018529689957736225994138912497217752834791315155748572424541506959508
295331168617278558890750983817546374649393192550604009277016711390098488240
128583616035637076601047101819429555961989467678374494482553797747268471040
475346462080466842590694912933136770289891521047521620569660240580381501935
112533824300355876402474964732639141992726042699227967823547816360093417216
412199245863150302861829745557067498385054945885869269956909272107975093029
553211653449872027559602364806654991198818347977535663698074265425278625518
184175746728909777727938000816470600161452491921732172147723501414419735685
481613611573525521334757418494684385233239073941433345477624168625189835694
855620992192221842725502542568876717904946016534668049886272327917860857843
838279679766814541009538837863609506800642251252051173929848960841284886269
456042419652850222106611863067442786220391949450471237137869609563643719172
8746776465757396241389086583264599581339047802759

FIGURE 2.1 The first 2000 digits of ␲.

03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 51

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 51

sand digits of ␲). However, these patterns and repetitions appear in small groups
of numbers, at least when ␲ is represented—as it is here (and elsewhere)—short of
its assumed infinite size.
In order to make sure that all readers understand the difference between com-
pression (reducing out redundant information and replacing it with place holders)
and simple representation, note here that the symbol ␲ does not indicate a useful
form of compression, even though anyone who knows its meaning could resurrect
some particular part of ␲ if it were discovered in data. Representations for data
compression must contain code that generates the original numeric forms of what
they represent, not just other convenient symbols. Otherwise, we would require
special symbols for every conceivable number sequence and would need to refer-
ence a large lexicon each time such symbols appeared in compressed data. Al-
though I will soon discuss a possible case for such a compression model in relation
to music, as a general rule, this kind of compression would require more symbols—
more likely partial symbols—and lookup time than any possible usable system
could manage.
The terms information and entropy are often used interchangeably. Although
these two may seem somewhat unrelated, they do share one critical dimension. In-
formation measures the amount of non-redundant information in data, while entropy
measures the amount of improbability or unpredictability in data. A high degree of
information therefore signifies a high level of improbability, unpredictability, or
non-redundancy. A high level of entropy in data indicates the same thing. For the
purposes of this book, however, I will use the term information rather than entropy
to avoid any possible confusion with entropy’s meaning in physics—more related,
among other things, to heat or energy exchanges between hot and cold bodies.
This definition of information does not indicate preference or aesthetic value.
One string of numbers is no better or worse than another string of numbers based
on its higher or lower level of information content. One might especially assume
preference or value when the string of numbers represents music—that in a musi-
cal work somehow high information content is preferable to low, or vice versa. In
actuality, calculating the information content of a string of numbers—musical or
not—simply indicates its level of redundancy versus what some refer to as random-
ness (I prefer the term unpredictability).
The notion of redundancy or non-redundancy plays very important roles in
defining information as used here. In effect, redundancy, although the complement
or opposite of information, can provide immensely important clues that allow un-
derstanding of a message even when that message is garbled. For example,
Every sentence in any language is highly redundant. A sentence of English—or of
any other language—always has more information than you need to decipher it.
This redundancy is easy to see. J-st tr- t- r—d th-s s-nt-nc-. The previous sentence
was extremely garbled, all of the vowels in the message were removed. However,
it was still easy to decipher it and extract its meaning, The meaning of a message
can remain unchanged even though parts of it are removed. This is the essence
of redundancy. (Seife 2006, 11)
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 52

52 HIDDEN STRUCTURE

Of course, omitting vowels is just one method of garbling messages. For example,
Hebrew and Arabic lack explicit vowels, and thus speakers of these languages
would not find Seife’s example at all disconcerting.
Information theory is a branch of communications theory that deals with the
amount and accuracy of information when transmitted from a source through a
medium to a destination. The information referred to in information theory can be
discrete, such as alphanumeric information (letter-and-number combinations), or
continuous, such as speech and music. Information theory has many practical
uses, including cryptography, data compression, error correction, and the like.
Suppose we have a message source which produces messages of a given type
such as English text. Suppose we have a noisy communications channel of speci-
fied characteristics. How can we represent or encode messages from the mes-
sage source by means of electrical signals so as to attain the fastest possible
transmission over the noisy channel? Indeed, how fast can we transmit a given
type of message over a given channel without error? (Pierce 1980, 44)

Founded by Claude Shannon (Shannon and Weaver 1949), information theory

depends heavily on statistics and probabilities. Shannon developed a theory that
actually produces relatively believable text from even a pseudo-random group of
letters and/or spaces—a mock example of a message received from a very noisy
medium. Shannon began with a series of six processes (the “orders” here indicate
the levels of probabilities taken into consideration):
1. Zero-order symbol approximation as in ZZFYNN PQZF LQN, where errors and
correct symbols exist side by side
2. First-order symbol approximation (including letter frequency), as in ID AHE RENI
MEAT, where letter frequency is taken into consideration as well as which letter
likely follows which letter
3. Second-order symbol approximation, as in RENE ID AHA MIET, where letter or-
dering two steps back is taken into consideration
4. Third-order symbol approximation, as in HE ARE ID TI NEAM, where letter order-
ing three steps back is taken into consideration
5. First-order word approximation, as in I DARE HE IN TAME, where letters and now
words are taken into consideration
6. Second-order word approximation, as in I DARE HE NAME IT, where letters, words,
and probable word order are taken into consideration (Pierce 1980)
Using simple letter and word probabilities such as these enables communication
systems to reconstruct information from noisy or corrupted messages.
The importance of information theory has become more apparent with time:
The theory of information did not seem all that important at first. True, it changed
the way cryptographers and engineers thought about their work; true, it set the
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 53

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 53

groundwork for building the computers that would soon become part of every-
day life. But even the founder of information theory, Claude Shannon, had no
idea just how far-reaching his idea would become. (Seife 2006, 56)

Morse code can provide a more concrete example of how information theory works,
particularly how information theory works with unknown languages. Samuel Morse
invented Morse code in 1837. This code, which survived for many decades as the
most popular type of telegraph, Teletype, or other form of communication, consists
of combinations of short and long signals (dots and dashes) representing letters of
the alphabet. Letters that occur frequently are represented with single dots or
dashes or, at most, combinations of two dots and/or dashes. Conversely, rarely
used letters typically consist of longer combinations of dots and dashes. This cod-
ing system makes it easier for telegraphers to quickly type and decode messages.
In fact, Morse’s code gained some 15 percent in speed over other methods of com-
munication in its heyday, the pre-telephone period, and concomitantly enabled tele-
graph communications that connected most of the civilized world. Dots, dashes,
and spaces are also much easier to use computationally because they can be ex-
pressed in binary using just three representations (00, 01, 10, for space, dot, and
dash, respectively), whereas the English-language alphabet requires at least twenty-
six separate binary representations, not including spaces, numbers, or special
characters.
Morse code offers a good example for demonstrating one use of information the-
ory. For instance, not knowing the language used (Morse code), how much noise is
present in the medium, or where this noise occurs in the output, among other
problems, makes the task of interpreting information very difficult. One encounters
these same difficulties in, for example, attempting to understand messages from
past civilizations or decipher radio messages in the search for extraterrestrial intel-
ligence (SETI). Music, when represented in numbers, actually poses similar prob-
lems, as we shall see. The process of interpreting noisy transmitted messages is the
very essence of communications theory and the origins of information theory—the
study of information loss and recovery.
Using Morse code instead of letters and spaces allows us to more easily imagine
the code as a language about which we have no prior knowledge (most readers will
likely find themselves in exactly that position here). According to Shannon, our
ability to translate a message—even with attendant noise—lies in direct proportion
to our understanding of the probabilities involved. For example, Shannon’s
previously discussed six-step process can be applied with the assumed notion that
all languages have inherent similarities, with much the same basic letter and word
probabilities. No doubt not precisely true in all cases, this assumption enables a
logical—if not accurate—translation, even without knowing which dots, dashes,
and spaces appear as a result of noise in the medium. The following code interpre-
tation follows these same six steps:
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 54

54 HIDDEN STRUCTURE

(1) --•• --•• ••--• --•-- --• --• •--• --•-- --•• ••--• •--•• --•-- --•

(2) •• --•• •-- •••• • •--• • --• •• -- • •-- --

(3) •--• • --• • •• --•• •-- •••• •-- -- •• • --

(4) •••• • •-- •--• • •• --•• -- •• --• • •-- --

(5) •• --•• •-- •--• • •••• • •• --• -- •-- -- •

(6) •• --•• •-- •--• • •••• • --• •-- -- • •• --

Each line of Morse code here represents exactly the same letters in the examples
included in the letters-and-words presentation of Shannon’s six steps. The concept
of using probabilities to create order from chaos will also prove quite valuable in
Chapter 6 of this book, which presents programs using Markov chains to model
post-tonal music in varying styles.
Figure 2.2 shows the basic Morse code symbols for each letter of the alphabet,
selected special characters, and the numbers 1 through 10. Interestingly, I pro-
duced the above code translation of English into Morse using the following simple
recursive Lisp function..

(defun translate-into-morse-code (letters)

(if (null letters)()
(cons (second (assoc (first letters) morse-code))
(translate-into-morse-code (rest letters)))))

The Lisp primitive assoc in the third line of code here associates letters with their
Morse code equivalents stored in the variable morse-code.
Information theory can also prove useful in exposing what is not present in a
message. For example, in 1939 Ernest Wright authored a book titled Gadsby, some
239 pages long, in which there is not a single instance of the letter “e”—the most
commonly used letter of the English alphabet. Of course, this omission wreaks
havoc on the probabilities inherent in overcoming noise in message transmission
using Shannon’s theory. The following first paragraph from Wright’s book provides
a good example of his narrative:

If youth, throughout all history, had had a champion to stand up for it; to show a
doubting world that a child can think; and, possibly, do it practically; you would-
n’t constantly run across folks today who claim that “a child don’t know any-
thing.” A child’s brain starts functioning at birth; and has, amongst its many
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 55

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 55

Letters
•- -••• -•-• -•• • ••-• --• •••• •• •--- -•-
a b c d e f g h i j k
•-•• -- -• --- •--• --•- •-• ••• - ••- •••-
l m n o p q r s t u v
•-- -••- -•-- --•• •-•- •--•- ---- ••-•• --•-- ---• ••--
w x y z ä á ch é ñ ö ü
Punctuation
•-•-•- --••-- ••--•• •----• -•-•-- -••-•
period comma question apostrohe explanation slash
mark mark
-••••- -••-• -•--•- •-••-•
hyphen fraction parentheses quotation

sign mark
Numbers
•---- ••--- •••-- ••••- ••••• -•••• --••• ---•• ----• -----
1 2 3 4 5 6 7 8 9 0

FIGURE 2.2 Morse’s code for English letters and numbers.

infant convolutions, thousands of dormant atoms, into which God has put a mystic
possibility for noticing an adult’s act, and figuring out its purport. (Wright 1939, 1)

Because the frequency of no other letter of the alphabet can replace the frequency
of the letter “e,” Shannon’s six-step process will reveal its absence without much
difficulty. Few composers, for example, have refused to use the pitch C, or six-
teenth notes, or a piano dynamic marking. However, the fact that many composers
did omit certain pitches from a composition, did not use certain duration combina-
tions, or made other such choices can be as useful to musical analysis as determin-
ing what these same composers included in these same compositions. Obviously,
any analytical approach that did not make such omissions clear would be far
less useful than one that did. After all, although Gadsby certainly has many other
attributes as a novel, any analysis not recognizing its missing “e”s would certainly
be incomplete.
Although most of the examples of information theory presented thus far have in-
volved language and not music, and although these examples display clear relevance
to messages and the meaning derived from such messages, the term information as
used from now on in this book will not imply message or meaning, but only the non-
redundant information present in data. I have relied primarily on language con-
structs thus far because they more easily convey the points being made.
I have also described information theory here primarily to set the stage for a def-
inition of algorithmic information theory (a subset of information theory that will
be described shortly). However, one can imagine many uses for information theory
itself in music. For example, as mentioned in Chapter 1, most early attempts at
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 56

56 HIDDEN STRUCTURE

FIGURE 2.3 An example of the kind of three-dimensional modeling of musical style

by Böker-Heil.

computer music analysis involved information theory. Norbert Böker-Heil’s re-

search with information theory and musical style analysis (see Böker-Heil 1972), for
example, produced interesting three-dimensional modeling of the type shown in
Figure 2.3. His plots of the information content of Palestrina, Rore, and Marenzio
particularly demonstrates their similarities and differences. Although the mathe-
matics are too complex to present here, suffice it to say that the large amount of
data that information theory analyzes and characterizes can be usefully repre-
sented and compared in such graphic outputs.
Furthermore, by better understanding how composers convey their intentions
through information in the form of music notation to performers who present this
information to audiences using voices and instruments, and how audiences inter-
pret this performed information, analysts can more thoroughly clarify their various
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 57

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 57

relationships. Each of these areas (intention, performance, and cognition) repre-

sents a significant potential for noise—mistakes—to enter into the process of com-
munication, prompting questions about information flow—the basic principle of
information theory. Therefore, information theory can lead us to a better under-
standing of how we perceive music, one of the foci of music cognition.
Algorithmic information theory (AIT), a branch of information theory, concen-
trates less on the communication accuracy of information and more on the precise
amount of non-compressible information contained in a message. In fact, algorith-
mic information theory assumes that the information content in the data it ana-
lyzes is completely accurate. Compressing a photograph, for example, typically
involves replacing the many thousands of bits representing the various colors with
symbols, thus identifying—so far as possible—the redundancy present. Because
photographs often have several areas consisting of one color, using symbols to rep-
resent these colors and their general locations can reduce the bit size of a photo-
graphic file significantly. Analyzing such compressed files can reveal a great deal
about their constitution—principal colors, balance of colors, and so on. Although
compression alone cannot usually determine the subject of a photograph, the ab-
stract analyses that result from compression provide information not immediately
available from viewing the photograph in its original state. As we shall see, what-
ever compression coding AIT uses must be able to restore, or decompress, the
photograph to its precise original state from the compressed data. Such compres-
sion/ decompression algorithms are referred to as codecs. AIT is the basis of many
other operations, including such processes as cryptography, information entropy,
and data compression.
The more compact we make data in AIT, the better the compression model. The
reason for this is not so much to save storage space, but to produce the best algo-
rithm for reducing the data. In effect, we have matched the data to the most effi-
cient algorithm. Of course, the algorithm used must remain consistent when
applied to different data of the same type, or comparisons between data will be un-
even. Also, uncompressed data file sizes reflect very little about their makeup,
while the use of appropriate and consistent algorithms to compress these same
data files reveals a great deal—their relative information content.
Gregory Chaitin (the originator of algorithmic information theory) commented
on how AIT defines
the complexity of something to be the size of the simplest theory for it, in other
words, the size of the smallest program for calculating it. This is the central idea
of algorithmic information theory (AIT), a field of theoretical computer science.
Using the mathematical concept of program-size complexity, we exhibit irre-
ducible mathematical facts, mathematical facts that cannot be demonstrated us-
ing any mathematical theory simpler than they are. (Chaitin 2005, 175)

As example, earlier in this chapter we discovered that the fraction 1/3 represents
information more succinctly than 0.3333333 . . . . Of course, our second example, ␲,
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 58

58 HIDDEN STRUCTURE

is just a symbol. We have not actually reduced the level of redundancy, but simply
replaced the infinite sequence with a convenient symbolic representation. However,
we can compress such numbers effectively using mathematics, as in
␲ = A/R2
where A represents the area of a circle and R represents the radius of that same cir-
cle. Even the formula just described can be slightly more compressed to
␲ = C/D
where C represents the circumference of a circle and D represents the diameter of
that same circle. AIT searches for the most economical way of expressing data. In
Lisp, ␲ exists as a constant (an unalterable variable), but both formulae above can
still be expressed as
(/ A (* R R))
and
(/ C D)
where the terms A, R, C, and D refer to the previously described characteristics of
a given circle. The latter formula much more concisely expresses 3.141592653589793
. . . , and thus even the numerical representation of ␲ can be compressed to a frac-
tion of its size.
Number sequences such as 2, 4, 8, 16, 32, 64 . . . can often reduce easily as well,
because these continuing powers of 2 can be expressed as
ƒn = 2n
meaning that the sequence results from incrementally increasing powers of the
number 2, with n beginning at 1. Thus, how we compress information greatly influ-
ences how much information a series of numbers actually possesses. Although it
may seem highly unlikely that musical data will easily reduce to mathematical for-
mulas, no method should be overlooked as a possible compression technique.
What may seem improbable as a tool for reducing an entire composition often
turns out to be exactly the right approach to compressing smaller sections of the
same composition, sections that might otherwise not be reducible by any other
means.
Another method of reducing number strings involves rules, a subject I discuss in
more detail in Chapter 6. Rules that can regenerate their source material exactly
can be extracted from musical data in many ways. For example, the rules of two-
dimensional cellular automata (algorithms that produce simple graphic output that
follow constraints) can prove very useful. As I will explain in Chapter 6, rules can
relate to musical attributes and can actually generate missing material as easily as
they regenerate compressed material.
A rule in a two-dimensional graphic automata takes the form of eight constraints
that govern whether or not a position in a given row will remain empty or be filled,
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 59

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 59

based on the state of positions in a preceding row (Wolfram 2002). For example, the
following rule coded in Lisp
(defvar *rules*
'(((* * *) *)
((* * o) o)
((* o *) o)
((* o o) *)
((o * *) o)
((o * o) *)
((o o *) *)
((o o o) o)))
dictates (line 2 above) that if a currently open position (“o”) is preceded by three
filled (“*”) positions (i.e., each of the three contiguous positions directly above it),
the new position will be filled, and so on. There are 256 (28) possible rules such as
the one presented above. Each of these rules leads to a predictable outcome. Fur-
ther variations arise, however, with different initial rule settings of the automata.
For example, Figures 2.4a and 2.4b use a single filled position in the initiating line.
Quite different immediate results occur with different rule settings, although many
of these eventually settle down into patterns similar or identical to those provided
by a single entry in the initiating line.
The recursive Lisp code presented below produces graphs similar to those
shown in Figure 2.4 using the rule appearing above.
(defun produce-automata (number start rules)
(if (zerop number) ()
(cons start
(produce-automata
(1- number)
(cons 'o (butlast (create-new-row start rules)))
rules))))
(defun create-new-row (old-row rules)
(if (null old-row) ()
(cons (apply-rule (firstn 3 old-row) rules)
(create-new-row (rest old-row) rules))))
(defun apply-rule (group rules)
(let ((test (second (assoc group rules :test #'equal))))
(cond (test)
(t 'o))))
(defun firstn (n list)
(if (or (zerop n)(null list))()
(cons (first list)
(firstn (- n 1)(rest list)))))
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 60

60 HIDDEN STRUCTURE

In Figure 2.4a, the output appears much as it will in any Common Lisp Listener win-
dow when running the above code, although I have enhanced the image here by re-
moving all of the empty spaces and parentheses to make the graph more readable.
In Figure 2.4b, I have turned this output on its side so that time is represented left
to right (rather than top to bottom) to better resemble musical data.
The asterisks here could translate to pitches on a vertical twelve-pitch grid with
the circles representing rests, or vice versa. Although I suspect that no complete
work will reveal cellular automata origins, I have often found that small sections of
many works reduce quite effectively following such processes.
Many types of compression techniques other than those I have described here
have been and are currently in use. Large pattern lexicons, for example, offer the
ability to place markers wherever certain patterns occur, requiring a lookup se-
quence in that same lexicon during regeneration. Musically such pattern lexicons
make sense, because it would be quite possible to store, say, 10,000 three- to ten-
pitch patterns in a lexicon, each pattern having its own designation that then repre-
sents that same pattern every time it occurs in a piece of music.
Such formulae, sequences, lexiconic lookups, and so on prove surprisingly ro-
bust in reducing small segments of musical data. Any process, no matter what its
likelihood of success, that eliminates the need to prosaically list data verbatim will
aid in compressing data and thus produce more accurate levels of information con-
tent. Even using several methods simultaneously can be effective, the only draw-
back being the obvious need for a more complicated process to resurrect the data
precisely to its original form.
Interestingly, only highly random number strings not reducible by pattern redun-
dancy, mathematical formulae, sequences, or lexicons approach 100 percent infor-
mation content. Some mathematicians conjecture that completely irreducible
numbers exist; they call them ⍀ (omega). These ⍀ numbers would consist of infinite
sequences of bit combinations in which no repetitions, patterns, or correlations ex-
ist (see Chaitin 2005, 76). Chaitin believes that ⍀ numbers exist even though no one
to date, including him, has discovered one. Indeed, no one has actually presented a
single such number or revealed the mathematics that would produce it. (⍀ pro-
vokes numerous important scientific debates, such as the existence of fundamental
[quantum] randomness; see Cope 2005, chapter 4; Chaitin 2005, 201–203).
In summation, then, AIT seeks to compress strings of numerical information to
their smallest possible size by using symbols to represent repeating numbers or
patterns of numbers. Thus, a sequence such as 2, 4, 5, 7, 8, 2, 4, 5, 7, 8 could reduce
to 24578r, where “r” represents the repeating sequence of numbers. The informa-
tion content of this simple string of ten numbers would thus equate to 0.6 or 60
percent, because 24578r is six symbols, rather than the ten symbols used in the
original sequence.
Note that using symbols in the manner just described, though suited for dis-
covering musical information content, does not follow precisely AIT binary compres-
sion processes. For example, symbols can often contain nearly as much physical
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 61

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 61

a)
a.
oooooooooooooooo*oooooooooooooooo
ooooooooooooooo***ooooooooooooooo
oooooooooooooo*o*o*oooooooooooooo
ooooooooooooo**o*o**ooooooooooooo
oooooooooooo*ooo*ooo*oooooooooooo
ooooooooooo***o***o***ooooooooooo
oooooooooo*o*ooo*ooo*o*oooooooooo
ooooooooo**o**o***o**o**ooooooooo
oooooooo*ooooooo*ooooooo*oooooooo
ooooooo***ooooo***ooooo***ooooooo
oooooo*o*o*ooo*o*o*ooo*o*o*oooooo
ooooo**o*o**o**o*o**o**o*o**ooooo
oooo*ooo*ooooooo*ooooooo*ooo*oooo
ooo***o***ooooo***ooooo***o***ooo
oo*o*ooo*o*ooo*o*o*ooo*o*ooo*o*oo
o**o**o**o**o**o*o**o**o**o**o**o
oooooooooooooooo*oooooooooooooooo
ooooooooooooooo***ooooooooooooooo
oooooooooooooo*o*o*oooooooooooooo
ooooooooooooo**o*o**ooooooooooooo
oooooooooooo*ooo*ooo*oooooooooooo
ooooooooooo***o***o***ooooooooooo
oooooooooo*o*ooo*ooo*o*oooooooooo
ooooooooo**o**o***o**o**ooooooooo
oooooooo*ooooooo*ooooooo*oooooooo
ooooooo***ooooo***ooooo***ooooooo
oooooo*o*o*ooo*o*o*ooo*o*o*oooooo
ooooo**o*o**o**o*o**o**o*o**ooooo
oooo*ooo*ooooooo*ooooooo*ooo*oooo
ooo***o***ooooo***ooooo***o***ooo

b.
(b)

FIGURE 2.4 Cellular automata output generated by the code in the text.
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 62

62 HIDDEN STRUCTURE

data as the data they represent, causing little actual compression (e.g., five-character
symbols replacing five-digit patterns hardly reduce number strings in terms of ac-
tual size). Therefore, the process described above is not so much intended to
reduce size significantly, but to direct us toward information content.
An important component of AIT is that the files compressed must be “lossless,”
that is, they must be restorable (by reverse processing) to their precise original
form. This principle—the recovery of exactly the data compressed—is essential to
the AIT process described in this chapter. Were it not for this reverse engineerabil-
ity, compression would not require such rigorous techniques, and our resultant
information-content percentages would be suspect. The function called com-
pression described later in this chapter makes restoration possible by having the
symbols that represent musical repetitions and patterns contain instructions that
allow restoration. This regenerative ability is a key element of AIT processes, and
although only the Comparison program has that capability, the Multigraph program—
both described later in this chapter—lacks it simply so that the information can be
read into physical graphs for visual presentation. As one might imagine, the more
complex the music, the more difficult full regeneration—pitch, duration, dynamics,
articulation, metrical positioning, and so on—becomes. Unlike text, or pixels in a
digital photograph, each parameter of music alters the other related parameters in
ways that make compression very difficult. As we shall soon see, separating these
musical elements from one another allows for more flexibility in the AIT process.

MUSICAL ALGORITHMIC INFORMATION THEORY

Algorithmic information theory did not reach a level of interest or success compa-
rable to that achieved in the early years of computer music analysis by information
theory. However, there have been several somewhat inadvertent uses of musical al-
gorithmic information theory. For example, MIDI presents an interesting example of
compression. Compared to complex notation program files, MIDI produces accu-
rate results with a fraction of the storage space required. However, MIDI also loses
a significant amount of information owing to its score representation approach (see
Selfridge-Field 1997b).
Musical algorithmic information theory involves specialized processes. What ap-
pear to the eye and ear of a musician as obviously related motives will not easily
submit to standard AIT compression. Thus, musical AIT as I define it here incorpo-
rates several forms of musical variation that help reduce strings of numbers to
much smaller amounts of information than traditional mathematical compression
could. For example, in music, the number string “2 4 5 7 3 5 6 8” can be reduced to
“2 4 5 7 t1,” where “t” means “transpose” and “1” signifies that adding 1 to each of
the first set of four numbers creates the second set of four numbers. This process
effectively reduces otherwise apparently non-reducible strings. Retrogrades, inver-
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 63

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 63

sions, inversion retrogrades, and all their possible transpositions represent yet fur-
ther information algorithms that, though not typically used to reduce strings in AIT,
have appeared in music for centuries. The terms inversion, retrograde, and inversion
retrograde mean precisely what the words signify. Inversion indicates that the pat-
tern appears inverted (upside down). Figure 2.5b presents an example of inversion
in musical notation, the inversion of Figure 2.5a. Figure 2.5c shows an example of
retrograde (backward), and Figure 2.5d shows retrograde inversion (backward and
upside down). These variation types have been used for centuries in Western clas-
sical music. However, the manner in which they are presented here results from
Arnold Schoenberg’s groundbreaking work with serial music in the early twentieth
century.
In order to reveal all of the possible transpositions of the inversion, retrograde,
and inversion retrograde, Figure 2.6 presents two matrixes of these types of musi-
cal variations and their transpositions of the four-note motive of Figure 2.5. Figure
2.6a lists the original motive and all of its eleven transpositions from left to right,
along with their retrogrades from right to left. Figure 2.6b lists the inversion of the
original motive and all of its eleven transpositions from left to right, along with
their retrogrades from right to left. Typically, motives and variations such as these
are referenced by “P” for prime (the original motive), “I” for inversion, “R” for retro-
grade, and “RI” for retrograde inversion.
Figure 2.6 requires slightly more explanation, for the entries appear in what
are called their pitch-class form. Pitch classes—as discussed in more detail in
Chapters 3 and 4—are pitches devoid of their register, or 0 through 11 (in this case
10 and 11 are represented by “t” and “e” respectively). In effect, pitch class states
that all Cs (0) belong to the same pitch class regardless of the register (octave) in
which they appear in music. Using pitch classes instead of pitches or numbers that
include octaves makes a matrix like that in Figure 2.6 much easier to read.
Another important kind of musical variation occurs often in tonal music, where,
for example, a simple downward scale can appear beginning on different pitches,
thus producing different arrangements of whole and half steps. Such tonal varia-
tions of simple motives require that a pattern matcher recognize similarities be-
tween like-designed motives that vary not only by transposition but by locations of
internal half steps. Figure 2.7 provides a simple example of such relationships. In
this figure, the downward scale appears three times in sequence, with the location
of the half step sliding to the left by one interval on each recurrence. Clearly such a

a. b. c. d.

&c œ œ bœ œ œ bœ œ œ œ œ
œ œ bœ bœ œ œ

FIGURE 2.5 Examples of retrograde (b), inversion (c), and retrograde inversion (d)
of a motive (a).
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 64

64 HIDDEN STRUCTURE

a.

0 5 3 2
1 6 4 3
2 7 5 4
3 8 6 5
4 9 7 6
5 t 8 7
6 e 9 8
7 0 t 9
8 1 e t
9 2 0 e
t 3 1 0
e 4 2 1

b.
b.

0 7 9 t
1 8 t e
2 9 e 0
3 t 0 1
4 e 1 2
5 0 2 3
6 1 3 4
7 2 4 5
8 3 5 6
9 4 6 7
t 5 7 8
e 6 8 9

FIGURE 2.6 Two matrixes of the four-note motive of Figure 2.5. The first matrix
provides the motive and its eleven transpositions (down from right
to left) along with its retrograde and its eleven transpositions (read
right to left). The second matrix provides the inversion and its eleven
transpositions along with its retrograde inversion and its eleven
transpositions.
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 65

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 65

&c œ œ œ œ œ œ œ œ œ œ œ œ
1/2 step 1/2 step 1/2 step

FIGURE 2.7 A simple downward scale beginning on different pitches, thus produc-
ing different arrangements of whole and half steps.

presentation to the ear would sound like a statement and two tonally transposed
repetitions. To a computer program unaware of such tonal musical forms, however,
the sequence would appear as twelve incompressible notes. Therefore, the code
provided with this book on CD-ROM contains a series of functions that take into ac-
count such tonal variations.
There are many other ways in which musical motives can be varied and yet re-
tain a representable connection to the original (or “prime”) upon which it is mod-
eled. For example, interpolated pitches, excised pitches, and raised or lowered
pitches can all give further opportunities for algorithmic compression. Although
such variations may be more difficult to represent and regenerate than simple in-
version, retrograde, and so on, they nonetheless provide compression possibilities
that most nonmusical approaches do not offer. This is the essence and value of
what I term MAIT.
The information content of a musical work—the part that cannot be further
reduced—consists of material that does not repeat sufficiently, exactly, or in recog-
nizable variation, such that a symbol can replace it to compress that passage. In ef-
fect, the higher the information content of music, the more it tends toward the
random (chaotic); conversely, the lower the information content, the more the mu-
sic tends toward organized repetition and variation (order). Such observations
might tend to suggest that lower-information-content music is more developed and
well formed, while higher-information-content music is more erratic and disorga-
nized. These comments, of course, further suggest that better music has lower infor-
mation content and vice versa. However, making such distinctions implies aesthetic
rather than objective values. What information content in music identifies for ana-
lysts is how that music relates to other music by the same composer, the music of
other composers, the style of the time, the form of the music, and many other fea-
tures soon to be discussed. Information content in music does not imply aesthetic
value, although, arguably, it might contribute to personal aesthetic evaluations.
My use of Musical AIT began in 2003, when I first utilized an algorithm to discover
the information content of works by such diverse composers as Bach (eighteenth
century) and Webern (twentieth century) in order to see if outwardly contrasting
styles might be comparable in terms of their algorithmic information content. This
research revealed that several composers of different centuries, whose styles differ
significantly, have more-similar information contents than many composers of
similar styles of the same century. Making such comparisons may indicate that
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 66

66 HIDDEN STRUCTURE

like-information-content works have more in common than one might initially imag-
ine. What that in-commonness is, of course, remains to be seen.
For example, the information contents of various segments of works by J. S.
Bach (0.51), Mozart (0.45), Beethoven (0.50), Brahms (0.64), Webern (0.88), Ernst
Krenek (0.87), and David Cope (0.63), determined using a very simple compression
scheme, demonstrate interesting apparent contradictions, with Brahms closer to
Cope than to Beethoven, and Bach closer to Beethoven than to Mozart. Of course,
most of the information figures here are expected. The Bach, Beethoven, and
Mozart examples have relatively low information content, and Brahms, Cope,
Krenek, and Webern have relatively high information content, covering a range of
0.45 to 0.88 overall. These numbers result from analyzing brief monophonic sam-
plings of the music only (i.e., a short melodic phrase typical of the composer’s
work; see Figure 2.8), chosen in order that all of the music appear here rather than

a.

? 3 .. œ œ œ œ b œ œ œ œ #œ œ œ œ #œ œ
Cello b4 œ œ œ œ œ œ
œ

? b œ œ œ œ bœ
5
œ œ œ #œ œ œ #˙
œ œ œ œ œ
œ

b c Ó Œ œœ œ œœœ œœ œ œ Œ œœ œ œœœ œœ œ
b.

& b œ

c.
b 2 j ‰ œj œ œ ‰ œj œ œ œœœ ˙
&b b 4 ‰ œ œ œ ˙ ˙ ˙ œœœœ œ

FIGURE 2.8 Brief monophonic samplings of music by (a) J. S. Bach, Suite no. 1 in G
Major for Violoncello Solo (1720), Minuet, mm. 1–8; (b) W. A. Mozart,
Symphony no. 40 in G Minor, K. 550 (1788), 1st movement, mm. 1–5;
(c) Ludwig van Beethoven, Symphony no. 5, op. 67 (1808), 1st move-
ment, mm. 1–5.
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 67

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 67

just referencing larger works which, interestingly, remain relatively faithful to these
information-content results.
Musical segments and works also retain individuality. For example, it is note-
worthy that each musical work has its own information signature. That is, no two
works contain precisely the same amount of information. Thus, one could, for ex-
ample, identify entire works by their MAIT information content. Determining the
information content of a musical work can also identify many other of its important
aspects. For example, comparing levels of information between two works can help
determine the general style of the music. Minimalism, for example, often produces

d.

& c ˙. œ œ. w ˙. œ w ˙. #œ w ˙. œ ˙
w

œ
e.
bœ œ-
2
&4 Œ ‰ nœ œ ‰ j j J ‰
#œ # œ-
œ.

f.
b œ- œ b œ œ
?c œ
J b˙ j ‰ Œ œ. bœ œ. J ˙ œ œ œ.
bœ.
p
˙ œ F P
g.
Jazzy
q = 120 > >
3 œ œ œ œ ≈ ≈ bœ bœ œ nœ nœ
# œ œ bœ œ nœ nœ œ ‰ bœ
& 4 œ œ œ œ #œ n œ b œ bœ bœ
> . > > fl >
f
FIGURE 2.8 continued
(d) Johannes Brahms, Symphony no. 1, op. 68 (1876), 4th movement,
mm. 30–38; (e) Anton Webern, Variations for Piano, op. 27 (1936), 2nd
movement, mm. 1–4; (f) Ernst Krenek, Suite for Violoncello, op. 84
(1939), 1st movement, mm. 1–4; (g) The author, Three Pieces for Solo
Clarinet(1965), 1st movement, mm. 1–3.
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 68

68 HIDDEN STRUCTURE

a low level of information, while indeterminism generally contains a high level of in-
formation. Most music, of course, falls between these two extremes; music with
subtle variations is lower in information, and music containing surprising contrasts
is higher in information.
Interestingly, information in MAIT consists of two general types. The first con-
tains data that occurs repeatedly in various forms throughout the remainder of the
work being analyzed, as previously discussed. The second contains data unique to
its single appearance in a work. To differentiate these two types of information re-
quires a program that can mark the two types in some way. Here, I use a very sim-
ple number that precedes the actual patterns that vary (type 1) and thus serves
two purposes: marking the repeating patterns and informing analysts of the num-
ber of occurrences found in one form or another. Unique information (type 2)
poses some very interesting questions. First, what overall role does unique infor-
mation play in a work of music? Second, is unique information locationally sensi-
tive, and if so, does it serve particular musical purposes (e.g., cadences, phrase
initiations, etc.)? Third, what percentage does each type of information occupy?
Even broader questions such as “Why does such information even exist in a work?”
seem appropriate. My initial studies of these two types of information have pro-
duced quite interesting results: Webern and Bach, for example, have virtually no
unique information, while Beethoven and Brahms have significant amounts. Unfor-
tunately, these results are preliminary and incomplete, and I shall leave it to others
to discover whether such interesting relationships exist.
Furthermore, and what is possibly more important, examining the information
and redundancy portions of data more closely can prove quite rewarding. Analyz-
ing music for its inherent structure typically means stripping away music that is
structurally less important in order to reveal the more significant structural mate-
rial. Because it would seem natural for variations to develop from source material
rather than vice versa, removing redundancy to reveal musical sources seems a
logical analytical process for music. MAIT accomplishes exactly this because it
uses the most effective algorithm possible for compression of music. For example,
Figure 2.10 presents a simple example of Figure 2.9’s music with all of the redun-
dant information removed from the score. Although clearly this smattering of music
does not equate to, say, the kind of structural analysis described by Heinrich
Schenker (Schenker 1979), the pitches here do offer quite interesting and revealing
views of the non-redundant data present in the work they represent. Note that most
of the music remaining in Figure 2.10 appears near the beginning of the work. This
is typical for such representations because motive originals occur before their vari-
ations. Only motives that originate late in a work or non-varied note-events unre-
lated to motive originals appear later in these reduced representations.
Traditional music analysis—including post-tonal analysis—does not require
restoration, as does MAIT. In fact, if music analysis required the same lossless regen-
erative abilities as MAIT, music analysis—even tonal music analysis—would look
very different than it does today. It is worth keeping this in mind when evaluating
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 69

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 69

œ œ #œ œ œ œ œ œ œ #œ œ nœ bœ œ
2
&4 œ œ nœ œ œ ˙ œ œ ‰
P J

j
& 42 Œ œ œ nœ œ œ œ œ œœ œ œ œ n œ œ # œ œ œ # œ . œj # œ # œ ‰
p
8 œ œ nœ œ œ œœ œœ œ œ œ nœ œ #œ œ œ #œ. œ œ nœ
&Œ J J ‰ Œ œ œ
p p
& œ œ #œ œ œ œ œ œ œ #œ œ nœ bœ œ œ œ œ œ œ ‰
˙ œ. œ œ #œ #œ
P p
16
œ
& #œ #œ œ œ œ #œ œ #œ #œ œ œ œ œ #œ œ œ bœ œ œ nœ œ œ bœ œ
più p P
& œ nœ œ œ #œ œ œ œ Œ
? #œ #œ œ œ œ #œ ‰ œ #œ œ
bœ œ #œ œ nœ
J
più p P
23

& œ œ nœ œ œ œ bœ œ œ œ bœ œ œ œ ˙ œ œ #œ œ
p più p
? b œ œ ‰ # œJ Œ œ nœ œ b˙ œ bœ
J œ
J bœ œ œ œ œ
p più p
29

&œ œ œ œ œ #œ œ œ n œ œ 43 œ œ œ œ œ 42 œ œ œ œ œ ˙ ˙
π P
#œ œ œ œ œ œ œ œ œ bœ œ b œ 2 œ œ # œ œ œ ˙œ œ ˙˙
?œœœ 43 4
π P
FIGURE 2.9 Bartók, Mikrokosmos, no. 81.
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 70

70 HIDDEN STRUCTURE

Wandering
Non troppo lento q = 76

2 œ œ #œ œ œ œ œ
&4 ∑ ∑ ‰ œ nœ Œ ∑ ∑

j
& 42 ∑ ∑ ∑ Œ ‰ œj j ‰
nœ Œ ∑ #œ #œ ‰
8 œ
&Œ J ‰ ∑ Œ ‰ œ ∑ ∑ ∑
J

& ∑ ∑ ∑ ∑ ∑ ∑

14
∑ ∑ Œ œ ∑ ∑ ∑
& œ #œ Œ ∑

& ∑ ∑ ∑ Œ ‰ j ∑ ∑ ∑ ? Œ ‰ œJ
œ
22
∑ Œ œ ‰ ∑ j
& J œ ‰ Œ ∑ ∑ ∑

bœ œ
? ∑ ‰ # œJ ∑ ∑ ∑ ∑ œ Œ

29

& ∑ ∑ œ œ Œ 34 ∑ 42 ∑ ∑ ∑

? ∑ ∑ ∑ 34 Œ b œ Œ 42 ∑ ∑ ∑

FIGURE 2.10 AIT structural analysis of the music presented in Figure 2.9.
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 71

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 71

the results of music compression. Without reverse-engineering capability, music

analysis could take a wide variety of speculative forms that would stand only the
scrutiny of logic, not the absolute value of accurate accountability to the original
form from which it derives.
Thus far in this chapter, we have concentrated our analytical methods on MAIT,
which provides us the ability to compare music of radically different styles in pos-
sibly useful ways. Although this comparison can prove useful for some analytical
processes, most analysts will likely consider it superficial rather than central to se-
rious musical analysis. I will not contest this perception because, though interest-
ing, the process has yet to prove valuable in realistic ways for my own research. In
contrast to this interesting but somewhat tangential analytical approach, I there-
fore now present a more meaningful use of MAIT called Dynamic Musical AIT
(DMAIT).
When I listen to music, my mind’s ear shifts from one musical parameter to an-
other. Melody, harmony, dynamics, timbre, rhythm, and other parameters all jostle
for my attention, with each prevailing at one time or another. When analyzing a
work, however, I find myself concentrating on just harmony, just form, and so on,
rather than allowing my analysis to shift in focus as my listening does. There are
probably many reasons I isolate one aspect of the music in this way. Primarily this
single-mindedness is due to my not having an easy method of analytically deciding
when, where, and how to represent my shift of focus.
Algorithmic information theory often yields quite convincing reasons why my
focus shifts when I simply listen to the work I am also analyzing. By indicating the
parameter—pitch, rhythm, meter, and so on—that offers the most information at
any given point in a work, MAIT can provide distinct points where the priority of in-
formation shifts from one musical parameter to another. This ability of DMAIT to
clearly indicate the points where one might move one’s attention from listening to
one aspect of the music to another could therefore have significant value.
I know of no empirical data to verify my contention, however. Rather than de-
velop this subject—more rightfully belonging to the field of music cognition—I will
restrict my comments here to the analytical implications of this process. I will con-
centrate on the interplay of information and redundancy between several, but not
all, musical parameters that suggests an analysis of any one parameter alone can-
not sufficiently address the true implications of the music.
The following DMAIT approach to analysis therefore incorporates and integrates
running commentaries on aspects of pitch, rhythm, meter, timbre, dynamics, and
other related parameters in an attempt to understand the full import of a musical
work. Breaking information into parameters in this way can reveal a great deal
about a work as it develops over time. Such analyses can often agree with more
traditional analyses and at times disagree with them, providing interesting new in-
sights into the music.
DMAIT systematically refigures MAIT for several musical parameters simultane-
ously. For instance, information content for pitch and rhythm computed anew from
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 72

72 HIDDEN STRUCTURE

the beginning of a composition for each added beat provides a continuous map of
both their individual information levels and their interplay over time. Using such
processes reveals which parameter might produce the most useful analytical re-
sults at any particular point in time. As well, dynamically plotting the information
content of several parameters of music at once can yield many insights into form,
structure, and other aspects of a musical work or passage. Figure 2.11 provides a
simple graph showing the pitch (register), rhythm (meter), dynamics (articulation),
and texture (timbre) of a work by Brahms, presented in standard notation at the
top of the figure. The parenthetical comments after each parameter above repre-
sent associated but not incorporated parameters that figure to one degree or another
into the analytical process.
The graph in Figure 2.11 also represents a kind of DMAIT fingerprint possibly
identifying the work and, to some extent, indicating the composer of that work (if
not already known). Although graphs of some works may appear similar, in fact
every work produces a unique DMAIT fingerprint, especially when reading the
floating-point numbers that produce the graph. Even two or three vertical slices of
such graphs demonstrate their individuality in terms of information content (or
redundancy, depending on your point of view).
Note that the graph in Figure 2.11 shows all four parameters initiating at the
highest points of information content—consistent in all such graphs because re-
dundancy cannot exist without preceding information. Interestingly, each musical
parameter here tends to eventually flatline into either horizontal or slightly rising
or falling lines. This flatlining results from the fact that so much data has been col-
lected that new information is scarce and has little impact. To avoid this flatlining
in longer works, I employ a moving DMAIT aperture of several beats preceding the
already calculated information content, rather than recomputing information con-
tinually from the very beginning of the work. This produces more distinctive active
lines that rarely settle into invariant configurations. Such apertures can lead to
somewhat misleading graphs when attempting to use DMAIT for formal and struc-
tural analysis.
Choosing intervals over pitches or vice versa in DMAIT is not a trivial decision.
Pitches provide important information about chromaticism in tonal music. DMAIT
will spike at each secondary chromaticism and ascend with modulation. Post-tonal
music typically rotates so quickly through the twelve pitches of the chromatic
scale that information often remains high. Using intervals, on the other hand, pro-
vides more insight into post-tonal music, informing analysts more directly about
the structure of the music. Figure 2.12 presents pitches reduced to intervals for
pattern matching. Note that the information here appears quite different than its
pitch counterpart.
A chromatic scale presents an excellent and simple example of the differences
between using pitch and interval when calculating information content in music.
From the perspective of pitches, a chromatic scale—with no repeating pitches—
represents a consistently high level of information. From the perspective of intervals,
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 73

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 73

& c ˙. œ œ. w ˙. œ w ˙. #œ w ˙. œ ˙
Ó
w

FIGURE 2.11 A simple graph showing four parameters of a melody by Brahms

(presented at the top of the figure).
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 74

74 HIDDEN STRUCTURE

b 2 j j
&b b 4 ‰ œ œ œ ˙ ‰ œ œ œ ˙ ˙
interval
representation: 0 0 -4 2 0 0 -3

pitch
representation: 67 67 67 63 65 65 65 62

FIGURE 2.12 Pitches reduced to intervals for pattern matching.

however, a chromatic scale—containing only one interval type, the minor second—
represents a very low level of information. Thus, data representation can have sig-
nificant impact on the derivation of information content. Neither view—pitch or
interval—is correct or incorrect. However, the contrasting results of the two ap-
proaches requires that one gauge the output within its appropriate context and
choose the best approach for particular genres (e.g., tonal versus post-tonal music).
Figure 2.13 presents a DMAIT reduction of Bartók’s Mikrokosmos no. 81 of
Figure 2.9. The vertical lines represent larger-than-a-beat time increments to allow
complete works to appear in full on a typical computer screen. As this graph
shows, DMAIT accomplishes far more than just revealing which musical dimension
is particularly emphasized at a given time; it also demonstrates aspects of form
(spikes with slow diminishment, particularly in rhythm), points of arrival (spikes
with little or no diminishment, particularly in pitch), and hints of many musical de-
velopments such as variation or changes in accompaniment figures.
For those keeping track of Lisp code to this point, I have included a single
change to the variable *horizontals* in the code file “Piano Roll” in the folder
called “multigraph” for this chapter on the CD-ROM accompanying this book to al-
ter the time increment size. Note that I use the Lisp macro defvar to define ini-
tially a variable and the Lisp macro setq to change a variable. This is an important
differentiation: attempting to change a previously defined variable with defvar
will prove unsuccessful. Also note that data presented between a #| and a |# is
considered documentation in Lisp, just like data following a semicolon. None of this
material will be evaluated by Lisp.

MUSICAL EXAMPLES
The Multigraph program (described in more detail in the next section of this chap-
ter) produces output like that shown in Figure 2.14 for the beginning of Stravinsky’s
Three Pieces for String Quartet (1914) and confirms what those who know this
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 75

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 75

FIGURE 2.13 Bartók, Mikrokosmos, no. 81, from Figure 2.9, in a Dynamic Musical
AIT graph.

work might well anticipate—information resides primarily in pitch, with texture, dy-
namics, and rhythm quickly leveling to nearly zero information content. Although
these observations may be predictable, the spikes in the pitch information repre-
sent useful indications of the manner in which new material is introduced in this
work when the surface of the music seems otherwise highly repetitious.
When I first began using AIT, it provided a useful though certainly not compre-
hensive tool for gauging the stylistic authenticity of works created by my program
Experiments in Musical Intelligence. In other words, comparing the AIT analyses of
a work produced by Experiments in Musical Intelligence with one from the music in
a database by the composer being emulated offered an interesting way to evaluate
style effectiveness. Such comparisons, although not explicitly involving the inter-
vals, harmonies, rhythms, and so on of the two works, ascertain the overall consis-
tency of the information flow.
As an example, the following two mazurkas in the style of Chopin represent the
computational and the human-created works involved in one such comparison.
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 76

76 HIDDEN STRUCTURE

FIGURE 2.14 Multigraph output for the beginning of Stravinsky, Three Pieces for
String Quartet (1914), shown in Figure 1.19.

Figure 2.15 presents these two mazurkas in full. Note that the latter Chopin
mazurka was not present in the Experiments in Musical Intelligence database dur-
ing the composition of its own mazurka. I make this clear at this point so that read-
ers will understand that the similarity in information content of the two works did
not result from their inclusion in the computer program’s output of one of the mod-
els upon which it was based. Both works are presented here in like manner; that is,
neither piece has dynamics or articulation, and neither was performed in the MIDI
file that produced the data. However, in order to make the comparison more logi-
cal, both pieces are of roughly the same length.
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 77

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 77

a.
œ bœ œ nœ. œ œ
b
& b 43 Œ œ
. œ œ œ œ œ J n œ œ œ .. œ Œ œ. œœ
J
œœ œœ œ œ œ œœœ œœœ
? b b 43 œ œ b œœœ œœ œ œœ œœ
œœ œœ
œ #œ
œ œ
œ œ

œ œ œ Œ œ
. œ œ œ bœ
J nœ œ œ bœ œ œ ‰ œ.
6
b
& b œ œ. œœ œ œ œ œ œ
J J
œœœ œœœ n œœœ œœœ b œœ œœœ œœ œœ œœ b œœœ
? bb œ œ œ œ œ œ œ
œ œ
œ nœ. œ n œ œ œ .. œ œ œ. œ. œ œ œ œ œ œ
n œ œ œ œ .. œ œ n œ
11

& bb J Œ

œ œ œœœ œœœ # œœœ œœœ œœ n œœ œœ # n œœœ

? b œ œœ œœ œ œ œ œ œ œ
b œ œ œ œ

œ. œ .. œ œ n œ ˙
16
b œ œ œ œ #œ œ ˙ œ
&b Œ J nœ œ

œ œ
? b b œ œœ œœ # œœœ n œœœ œ œ œœ # n œœœ œ
œ œ œœ œœ œ
œ œ œ œœ Œ

œœ .. œ œ œ œ
21
b
&b ˙ œ œœ .. œ œ œ œœ œœ ˙
Œ
œœ œ œ œœ ˙ œœ œœ .. œ
Œ œ Œ Œ œ

? bb Œ œ œ œœ œœ œœ œœ œœ œ œ œ
œ Œ Œ œ œ Œ œ œ Œ
œ œ œ

FIGURE 2.15 Two mazurkas in the style of Chopin for analytical comparison.
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 78

78 HIDDEN STRUCTURE

26
b œ ˙
& b œ œ nœ œ œ œ. #œ œ ˙˙ œœ œ n œ œœ œ œœ œ˙ # œ n œ œ n œœ
J Œ

? b b Œ œœ œœ œ œ œ œ œ œ œœ
œ œ Œ n œœ œœ Œ nœ Œ n œœ œœ
œ œ

n œ n ˙œ # œ n œ œ œœ œ œ œ œ # œ ˙
31
b b n ˙œ œ œœœ œ œœ
& #œ #œ œ nœ œ œ nœ œ
œ œ
#œ œ œœ œœ œœ œœ
? bb Œ n n œœœ œœ
œ Œ n œœ œœ Œ œ œ
œœ œœ
œ œ
Œ œœ œœ
#œ œ
œ œ
˙ œœœœœ œœœ
37
b
& b œ œ œ. œ œ
j œ . œ œ œ œ œ nœ bœ œ œ œ œ œ
# œ œ . œj
# œ œœ œœ œœ œœ œœ œ œ œ œ œ
? bb œ œ œ œ œ œ œ œ œ œ œ b œ œ œ # œœ n œ
œ œ

j œ bœ nœ. œ n œœ œœ œ .. œ œœ
. œ œ œ œœ b œœ œ œ œœ œ . œ.
43
b
& b Œ œœ . œ œ œ œ
J
œ œ Œ œ. œœ œœ

œœ œœ œ œ œ œœœ œœœ
? bb œ œ b œœœ œœ œ œœ œœ
œœ œœ
œ #œ
œ œ
œ œ

b œ œœ . œœ œ œ œœ œœ Œ œœ .. œ n œœ œ b œœ œ œœ œ j œ . j œ bœ
48
b
& œ œ œ œ œ œ œ œœ ‰ œ . œ œ œ œœ b œœ œ œ
J œ œ œ
œœœ œœœ n œœœ œœœ b œœ œœœ œœ œœ œœ b œœœ
? bb œ œ œ œ œ œ œ
œ œ

FIGURE 2.15 continued

03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 79

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 79

œœ . œœ œœ œ œ œœ œœ
b œœ n œœ .. œœ n œœ œœ œœ .. œœ œœ œ. n œ œœ # œœ œ .. œ œœ # n œœ
53

& b Œ œ. œ œ œ œ
J
œ œ œœœ œœœ # œœœ œœœ œœ n œ œœ # n œœœ
? b b œ œœ œœ œ
œ œ œ œœ œ œ
œ œ œ

58
b œ. œœ . œœ # n œœ œœ œœ œ œ œ ˙ œœ ˙ œœ # n œœ ˙
&b Œ œ. œ œ œ ˙ ˙ ˙
œ œ œœ # n œœœ œ œ œœ # n œœœ œ
? b b œ œœ œœ œ œ œ œœ œœ œ œ œ œœ Œ
œ œ

FIGURE 2.15 continued

03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 80

80 HIDDEN STRUCTURE

b.
# œ nœ œ
& 34 œ œ œ œ œ # œ œ # ˙ œ #œ œ œœ œ œ ˙ œ œ œ #œ œ œ œ

? # 34 œœœ œœ # œœœ n œœœ b œœ

œ œœ œœœ œœœ # # œœœ œœœ
œ œ œ œ œ

6
# œ
& ˙ œ œ œ #œ œ œ œ œ œ #œ ˙ œ œ œ #œ œ œ œ œ ‹œ #œ œ

?# œœ œœ œœ œœ œœœ œœœ œœœ œœ

œ # œœœ n œœœ
œ œ œ œ
œ œ œ œ œ

11
# nœ œ
& œ œ œ #œ œ œ ‹œ #œ nœ œ œ œ œ œ #œ œ #œ #œ #œ œ #œ ‹œ œ. œ
b œœ œœ œœ # œœ b œœœ bb œœœ #œ œ
?# œ nœ œ œ b œœ # ‹ œœ # # # œœœ œœ
œ
œ œ œ bœ #œ

16
# #œ #œ #œ #œ
& ‹œ #œ #œ #œ œ # œ ‹œ œ. œ #œ #œ ‹œ #œ # œ #œ #œ œ #œ nœ œ

?# # # # œœœ œ
# œœ # # # œœœ œœ
œ # # # œœœ œ
# œœ # œœœ œœ
œ
#œ #œ #œ #œ

20
# j
& œ . # œj œ œ œ œ œ œ # œ œ # œ . # œ n œ œ œ # œ n œ # œ œ œ œ œ ˙

?# œœ œ œœ œœœ # œœœ œœ
œ
œœ
# œœœ œœ œœ
œ œ # œœ œ œ #œ œ œ
œ œ œ

FIGURE 2.15 continued

03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 81

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 81

# œ. œ œ œ œœ ≈ œœ ˙˙˙ œ. œ œ œ
25

& n œœ . œœ œœ œœ œ œ n œœ . œœ œœ œœ œœœ ≈ œœœ ˙˙ n œœ ≈ œœ œœ œœ

œœ ˙ œœ ˙˙
? # œœœ œœ œœ
œ œ œ ˙˙ œœ
œ˙
œœ œœ
œ œ œ ˙ n œœœ ≈ œœœ œœ œ
œ œœ
˙. .

œœ .. œœ œœ œœ œœ ≈ œœ ˙˙˙
30
# œ œ œ
& n œœ ≈ œœ ˙˙ œœ ≈ œœ œœ œœ
œ œœ œ n œœ ≈ # œœœ œœ œœ nœ œ œ œ œ
˙
? # n œœœ ≈ œœœ ˙˙˙ œœ ≈ œœ œ œ œ
œœ
œœ
œ
œœ
œ
œœ
œ˙
œœ œœ
œ œ
œœ
œ ˙˙
œ œ .
# .
35
œ
& n œœœ . œœœ œœ œœ
œ œœ ≈ œœ ˙˙
œ œ ˙ n œœ ≈ œœ œœ œœ n œœ ≈ œœ ˙˙ œœ ≈ œœ œœ œœ
œ œœ œ

? # œœœ œœ œœ œœ ˙˙ n œœœ ≈ œœœ œœœ œœœ n œœœ ≈ œœœ ˙˙˙ œœ ≈ œœ œ œ

˙. œ œ œ ˙
œ œ
40
# œ
& n œœœ ≈ # œœœ œœœ œœ œ≈ œ œ
# œœ œœ œœ œœ
œ œ œ
b n œœ ≈ œœœ œœ œœ n œœœ ≈ œœ œœœ œœœ œœ ≈ œœ ˙˙˙

? # œœ œœ œœ
œ œ œ ≈œœ œ œ n œœœ œœœ œ ≈œœ œ œ ≈œ˙
œ œœ œ n œœ œœ œ œ œ˙
45
# œ œ œ œœ œœ ≈ œœ ˙˙
& n œœ ≈ œœ œœ œœ n œœ ≈ œœ œœ œœ n œœ ≈ œœ œœ œ œ œ ˙

? # n œœœ ≈ œœœ œœœ n œœœ ≈ œœœ œœœ œ

œœ
œœ
œœ
œœ
œœ
œœ
œ
˙˙
˙
œ œ

FIGURE 2.15 continued

03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 82

82 HIDDEN STRUCTURE

49
# œ œ œœ
& œ œ œ #œ œ bœ œ œ œ œ œ #œ œ œœ œ œ œ œ œ œ œ œ œ œ œ
#œ
œœœ œœ # œœœ n œœœ b œœ œœ œœœ œœœ # # œœœ
?# œ œ œœœ
œ œ œ œ œ

54
# œ
& œœ œ œ œ œ #œ œ œ œœ œ #œ ˙ œ œ œ #œ œ œ œ œ ‹œ #œ œ

?#
œœ œœ œœ œœ œœœ œœœ œœœ œœ
œ # œœœ n œœœ
œ œ œ œ
œ œ œ œ œ

59
# nœ œ
& œ œ œ #œ œ œ ‹œ #œ œ œ œ œ œ œ #œ œ #œ #œ #œ œ #œ ‹œ œ. œ

b œœ œœ œœ # œœ b œœœ bb œœœ #œ œ
?# œ nœ œ œ b œœ # ‹ œœ # # # œœœ œœ
œ
œ œ œ bœ #œ

64
# #œ #œ #œ #œ
& ‹œ #œ #œ #œ œ # œ ‹œ œ. œ #œ #œ ‹œ #œ # œ #œ #œ œ #œ nœ œ

?# # # # œœœ œ
# œœ # # # œœœ œœ
œ # # # œœœ œ
# œœ # œœœ œœ
œ
#œ #œ #œ #œ

68
# j
& œ . # œj œ œ œ œ œ œ # œ œ # œ . # œ n œ œ œ # œ n œ # œ œ œ œ œ ˙

?# œœ œ œœ œœœ # œœœ œœ
œ
œœ
# œœœ œœ œœ
œ œ # œœ œ œ #œ œ œ
œ œ œ

FIGURE 2.15 continued

03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 83

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 83

Figure 2.16 presents the two DMAIT graphs representing these two mazurkas,
with each graph roughly matching the other. The two graphs certainly differ, as
would be expected, because any two works will have somewhat differing informa-
tion identities. However, the contours of both works’ analyses—the important
characteristic—resemble one another quite closely.
Because DMAIT measures only the flow of information resulting from repetition,
variation, and contrast, the mazurka form (especially given the brevity of both of
the works in Figure 2.15 and their lack of explicit repetition) is not responsible for
the similar shapes of information content in the two graphs. Nor can one count too
heavily on the similarity of the characteristic rhythms of the right hand, factors
that result from the Experiments in Musical Intelligence recombinant mode of com-
position (see Cope 2001). Thus, the information in the graphs results from the flow
of musical ideas and the rate at which these ideas develop as each piece pro-
gresses through time. Because information content does not constitute a part of
the Experiments in Musical Intelligence‘s mode of operation, one can only surmise
that the highly similar information-content contours occur as a direct by-product of
the process of composition. The similarity of the two graphs validates the relative
authenticity of the output of Experiments in Musical Intelligence.
Were such parallel comparisons rare between the human-composed works and
replications produced by Experiments in Musical Intelligence, this similarity be-
tween the two mazurkas might be considered coincidental. However, I have made
such analyses time and again with roughly comparable results. Such comparisons
lead me to conclude that Emmy’s process of composition does not differ signifi-
cantly from that of human composers (see Cope 2005, particularly chapter 12). My
conclusions do not derive from observation of the information content alone (see
also Cope 2001). I understand the reluctance on the part of many readers to accept
this conclusion, and to some degree I share this same skepticism of the belief that
humans compose by hybridizing what they have previously heard. However, the in-
formation provided here seems to indicate that they do.
To give readers an idea of how works of non-comparable information contents
compare, Figure 2.17 presents a prelude by Chopin of comparable length to the two
mazurkas of Figure 2.15. The contour of the graph shown in Figure 2.18 varies sig-
nificantly from both of the contours presented in Figure 2.16. Remembering that
although the um-pah-pah of the left hand in the mazurka and the somewhat less
repetitious left-hand configuration of the prelude do not in themselves contribute
much to the overall information variability, the contrast between the prelude and
both mazurka analyses proves quite substantial.
Standard non-musical algorithmic information theory typically compresses bi-
nary data to binary data for efficiency. Binary data uses only two numbers—0 and
1—appearing right to left in columns following successive powers of 2 but origi-
nating with 1. To read these numbers, begin from the rightmost number and count
the columns as 1, 2, 4, 8, 16, 32, 64, 128, and so on, with 0 meaning to add 0 and 1
meaning to add that the column’s number when computing the result. For example,
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 84

84 HIDDEN STRUCTURE

g g p p g p g

FIGURE 2.16 Two DMAIT graphs representing the Chopin mazurkas shown in Fig-
ure 2.15.
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 85

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 85

FIGURE 2.16 continued

03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 86

86 HIDDEN STRUCTURE

#
& 34 # œœœœ œœœœ œœœœ .. œ œœœ œ . # œ œ # œœœœ œœœœ œœœœ .. œ œœœ œ œœœ # œ œ # œœœœ œœœœ œœœœ .. œ
.
j j
? # 34 œ œ œ ˙˙ œœ . ≈ œ œ œ ˙ œ.≈ œ œ œ
. ˙ œ.

6
# œ
& # œœœœ ≈ œœ œ œ œ œ n œ œ . œ œ œ œ # œœœœ œœœœ ..œ œœœ œ . # œ œ # œœœœ œœœœ œœœœ .. œ
œœ .
˙ œ œ nœ œ j
?# œ œ. œ œ Œ Œ œ œ ˙˙ œœ . ≈ œ œ œ
œ œ .
12
# œ
& œœ œ œœ # œ œ œ œ # œœœœ œœœœ .. œ # œœœ œ ≈ œœ œ œ œ œ
#œ nœ. œ œ œ œ
œ œ œ œ
j ˙ œ œ
?# ˙ œ. ≈ Œ œ œ œœ #œ nœ. œ œ œ œ
˙ œ. œ œ
17
# #œ œ #œ œ # œœ # œ # œœ # œœ
& # œ œ # œœ # œœ œœ œœ . # œœ # œœ # œœ œ œ œ #œ œœ # œœ # œœ # œœ

? # œ œœ œœ œœ œœœ œœœ œœ # œœ œœ œœ # œœ œœ œœ œœ
œ œ œ œ œ œ
œ œ
22
# #œ œ # œ # œ # œœ # œœ # œ # œœ # œœ #œ œ
& # œœ œœ œœ . # œ œ œ # œ œ œ # œœ # œœ # œœ œœ œœ .

? # œ œœ œœ Œ # œ
#œ œœ œœ # # œœ œœ œœ œœ œœ œœœ œœœ
œ œ œ
œ

FIGURE 2.17 A prelude by Chopin comparable in length to the pieces in Figure

2.15.
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 87

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 87

27
# #œ œ # œœ # œ # œœ # œœ #œ œ # œ # œ # œœ
& # œœ # œœ # œœ œ œ œ #œ œœ # œœ # œœ # œœ œœ # œœ œœ œœ . # œ œ

? # œœœ # œœœ œœœ œœ # œœ œœ

œ œ œ œœ œœ œœ œœœ œœœ #œ
Œ # œ
œœ
œ
32
# #œ # œ œœ œ œ
& œœ # # œœ # œœœ œ œ œ # œ n œ # œœ œ # # œœ # œ # œœ œ œœ œ # œ # œ œ

? # œœ # # œœ œœ # # œœ œœ œ #œ nœ œœ # œ # œ œ œ
#œ #œ œ œ œ #œ

37
# œ #œ
& # # œœ œœ œ ≈ # œ n œ # # œœœ œœ #œ œ œ œ #œ nœ #œ œ œ # œ
#œ œ # œ œ œ # œ œ œœ

?# # œœ # # œœ œ #œ # œ # œœ œœ œœ
#œ #œ #œ #œ œ #œ # œœ
œ
42
# #œ œ #œ œ # œœ # œ # œœ # œœ #œ œ
& # œœ œœ œœ . # œœ # œœ # œœ œ œ œ #œ œœ # œœ # œœ # œœ # œœ œœ œœ .

? # œœ œœœ œœœ œœ # œœ œœ œœ # œœ œœ œœ œœ œœ œœœ œœœ

œ œ œ œ œ œ
œ
47
# #œ œ # œœ # œ # œœ # œœ #œ œ # œ # œ # œœ
& # # œœ # œœ œ # œ œ œ # œœ # œœ # œœ œœ œœ . # œœ œ œ

? # Œ # # œœ œœ œœ # # œœ œœ œœ œœ œœ œœœ œœœ œœ # œœ œœ
œ œ œ
œ

FIGURE 2.17 continued

03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 88

88 HIDDEN STRUCTURE

œœ œœ œœ . œ
52
# #œ
& œœ # œœ # œœœ ## œœœ œ œ # œœ # œœ œœ # œœ # œœœ œœ . œ # œ # # œœ # œœœ # œœ # œ # œœ
œ# œ œ œ œ œ.
œ #œ

? # œœœ # œœœ œœœ œœ œœ œœ œœœ œœœ #œ œ œœ # # œœ œœ # œœœ œœœœ œœœœ .. œ

Œ# œ œ .
œ

œœ . œ œœ œ ≈ œ œ
58
# œœ œœ . œ œœ œ œ œ œ œœ
& œ œ. #œ œ œ
# œœœœ
œœ
œœœ œœœ . œ #œ # œœœœ œœœ œœœ .. # œœœ œ
.
? # œœœ œ . # œ œ
œ œ œ
œ œ œ
œ œ œ œ œ œ œœ œœ œœ
œ œ œ œœ œ

63
# œ œ œ
& œ #œ nœ. œ œ œ œ
œœ œœ . œ œœ œœ . œ œœ
œ # œ n œ . œ œ œ œj . ≈
œ œ
? # œ œ #œ nœ. œ œ œ œ œ œ œ œ œ œ
# œ n œ . œ œ œ œj . ≈
œ œ . œ
œ œ . œ œ œ.

FIGURE 2.17 continued

a byte of information—eight columns of binary data—creates the full range of deci-

mal numbers between 0 and 128 with, for example, 0 equal to 00000000, 1 equal to
00000001, 2 equal to 00000010, 3 equal to 00000011, 128 equal to 10000000, and so
on. Binary numbers can be added, subtracted, multiplied, and otherwise operated
on, just as decimal numbers can.
I have avoided using binary data here for two reasons. First, binary data requires
a fluency in reading that most readers do not possess. Second, and possibly more
important, this ability to read binary data becomes particularly important when
actually viewing the compressed data as described earlier in this chapter. In fact,
when analyzing music using MAIT, I represent pitches as note-events rather than
just pitch numbers. My reasons are simple: note-events incorporate location, dura-
tion, and other qualities that inform analysis more usefully than do either pitch or
binary pitch information alone.
For example, the following data provides very useful details about the type and
location of the compressed information in a work.
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 89

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 89

FIGURE 2.18 A Dynamic Musical Algorithmic Information Theory graph of the

Chopin prelude shown in Figure 2.17.
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 90

90 HIDDEN STRUCTURE

((0 60 1000 1 127)(1000 72 1000 1 127)(2000 71 1000 1 127)

(4000 72 1000 1 127)(5000 67 1000 1 127)(6000 76 1000 1
127)
(7000 74 1000 1 127) T6 T4 TI6 TR4 T0)

This forty-two-note passage poses its information only at the outset, with all re-
maining entries representing transpositions, transposed inversions, or retrogrades
of the original pitch patterns. On the other hand, the following brief nineteen-note
passage
((0 60 1000 1 127)(1000 72 1000 1 127)(2000 71 1000 1 127)
T2 T3 R6 T4 (15000 72 1000 1 127)(16000 71 1000 1 127)
(17000 67 1000 1 127)(18000 74 1000 1 127))

concludes with a completely new motivic idea after several simple variations of the
initial one. Reading even complex compressed files presented in this manner for
their information content can prove enlightening in matters of form, structure,
melodic invention, and so on. Only note-events, or a similar notation standard such
as MIDI, can provide this kind of readable and interpretable detail.
Interestingly, I have also used Multigraph, the program that created the preced-
ing analyses, in my own compositional processes. By viewing graphs of an unfin-
ished work I am able to gain important insights into how various parameters of my
work-in-progress presents itself and as a result either change aspects of the compo-
sition thus far composed or alter my views as to how the music should proceed
from that point onward. Analysis is as important for composers as it is for analysts,
and it seems to me these tools should be considered from both points of view.
I have kept the examples in this chapter short and thin-textured in order to facili-
tate easier understanding of the processes described. Readers should be able to
follow the code on the CD-ROM more easily without complex harmony and counter-
point confusing the issue. However, with several small alterations, this code will
function using any musical parameters and can analyze quite complex post-tonal
music with basic AIT processes.
Interestingly, John Pierce, one of the pioneers of information theory, asked and
then answered the question Does information theory have anything concrete to of-
fer concerning the arts? “I think that it has very little of serious value to offer ex-
cept a point of view, but I believe that the point of view may be worth exploring”
(Pierce 1980, 253). For the most part, Pierce was referring here to composing, not
analysis. However, his pessimism seemed even more pointed later in this same
source: “Perhaps in some age of bad art, man will be forced to [make] stochastic
art as an alternative to the stale product of human artisans. So much for informa-
tion theory and art” (Pierce 1980, 267). Although I understand Pierce’s point of
view, I must respectfully disagree with these comments. The amount of information
that a work of art possesses seems quite important, for it relates to the amount of
information contained in other works by the same composers, or in works by other
composers in the same style. Dynamically gauging the information content of a
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 91

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 91

time-based work of art such as music as it releases that information seems particu-
larly important.
However, AIT also produces more questions than it answers. For example:
1. What role do rests (silences) play in music information (currently, note-events
do not represent rests)?
2. How does DMAIT shed light on first, second, and subsequent hearings, where
multiple listenings resist surprise?
3. How do note ontimes figure into algorithmic information theory, given that no
two note-events in performed music can occur at precisely the same time when
ontimes figure into the compression model?
4. Does DMAIT truly provide a formal or structural model of a piece of music?
Would anyone suggest, for example, that removing redundant material from a
printed musical score could rival a Schenkerian analysis in revealing fundamen-
tal structure?
5. Do certain compression profiles for music relate in some way to what we feel are
masterworks?
Although certainly a valuable tool in analyzing music—particularly in its dynamic
form of representation, as described in this chapter—algorithmic information
theory falls short of providing the kind of deeply meaningful analysis that music
analysts seek. The unbiased but also often unmusical approach that algorithmic in-
formation theory takes requires more musical detail in both its processes and its
analyses than I have included here. The analytical approaches in the chapters to
come will hopefully provide more of this kind of musical detail.

PROGRAM DESCRIPTION
Before describing the programs accompanying this book on CD-ROM, I think it im-
portant to demonstrate how pattern matching—the core of information theory—
works, and particularly how it can work in Lisp. As an example, the following Lisp
code pattern matches two equal-length lists of pitches against one another accord-
ing to prime, retrograde, inversion, and retrograde inversion forms, as well as all
twelve transpositions for each form. Note how the process here divides into three
separate functions (get-intervals-from-pitches,compare-interval-lists,
and reverse-sign) that are then combined in ways that make the top-level func-
tion (pattern-match) quite powerful. Note that Common Lisp allows for document
strings (in “”) after the declaration of argument variables. These document strings
can be retrieved and help someone reading the code for the first time to under-
stand its purpose. The function or (in compare-interval-lists) represents a
Boolean query that returns t if any one of its arguments (arbitrary number) is true.
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 92

92 HIDDEN STRUCTURE

(defun pattern-match (pitch-list-1 pitch-list-2)

"Determines whether two pitch-lists are equal in one of 48
ways."
(compare-interval-lists (get-intervals-from-pitches
pitch-list-1)
(get-intervals-from-pitches
pitch-list-2)))
(defun get-intervals-from-pitches (pitch-list)
"Returns intervals from its pitch-list argument."
(if (null (rest pitch-list)) ()
(cons (- (second pitch-list)(first pitch-list))
(get-intervals-from-pitches (rest pitch-list)))))
(defun compare-interval-lists (list-1 list-2)
"Compares by retrograde, inversion, and retrograde inversion."
(if (or (equal list-1 list-2)
(equal (reverse list-1) list-2)
(equal (reverse-sign list-1) list-2)
(equal (reverse (reverse-sign list-1)) list-2))
t))
(defun reverse-sign (number-list)
"Turns intervals upside down."
(if (null number-list)()
(cons (* (first number-list) -1)
(reverse-sign (rest number-list)))))
Running the following argument lists with pattern-matching will return t for
prime transposition
'(60 62 64 62) '(62 64 66 64) ;
for retrograde
'(60 62 64 62) '(64 66 64 62) ;
for inversion
'(60 62 64 62) '(62 60 62 64) ;
for retrograde inversion
'(60 62 64 62) '(64 62 60 62) .
The function get-intervals-from-pitches first tests its argument (second
function in code here) to see if the rest of its list argument is empty or not, rather
than the usual test for whether the argument itself is empty. This variation on a
theme results from the fact that the last pitch in the argument cannot be compared
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 93

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 93

to a nonexistent following element for its interval. Another way to view this in-
volves the use of second in the third line of actual code for get-intervals-
from-pitches, where, with but one number remaining in the list, the call to
second would return nil. Code variations such as this are often required for
specialized cases.
Note that the function pattern-match is not very elegant in that it simply re-
turns t or nil in comparing two equal-length arguments. A more complete search
of a musical work for patterns will require further higher-level functions to collect
patterns of varying lengths and compare them with the patterns of an entire work
(a recursive function). Programmers must also decide whether to search note lists
incrementally or contiguously. Incremental pattern matching requires that patterns
begin on each new note of a work, while contiguous pattern matching requires that
patterns follow one after the other with no overlapping pitches. In essence, incre-
mental pattern matching moves the aperture one pitch at a time, whereas contigu-
ous pattern matching moves the aperture to the note following the current pattern.
Incremental pattern matching is far more thorough, but it requires more computing
time than does contiguous pattern matching.
Note that Common Lisp contains a number of simple pattern matchers as default
primitive functions. The function equal, for example, determines whether two pat-
terns are equal or not. Functions such as >, <, and so on compare numbers to de-
termine whether one is greater or less than the other. These functions, however,
offer little more than absolute matching. In other words, they contain no forgive-
ness. Forgiving pattern matchers that detect inexact matches and, for example,
measure pattern relations as mostly equal, partly equal, or in some other relation
can be extremely useful but require more intricate code than that presented here
(see Cope 1996 for examples). Such forgiving pattern matchers also must clearly
designate what the variations consist of; otherwise the matching will be of rela-
tively little value for MAIT analysis.
In order to ensure that every possible pattern is considered, the program called
Compression on the CD-ROM accompanying this book collects and labels pitch pat-
terns of length equal to, or less than, half the length of its argument incrementally
down to three pitches in length. Groups of two pitches—an interval—do not consti-
tute musical patterns worth matching. Larger patterns are matched first to avoid
the necessity of separately labeling the many smaller patterns that would match
within these larger patterns if they themselves match. Patterns are tested for equal-
ness, transposition, retrograde, inversion, and inversion retrograde and for all pos-
sible transpositions of these three variations. To accomplish these matches, pitches
are reduced to intervals and those intervals then matched (see Figure 2.12). Search-
ing by intervals rather than pitches allows the matching process to capture rela-
tionships that would otherwise require several intermediate stages. Finally, in a
separate stage, the program checks for repeated pitches.
Each discovered pattern match is labeled according to (1) its transposition;
(2) its variation type, as in “I” for inversion, and so on; (3) the length of the pattern;
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 94

94 HIDDEN STRUCTURE

(4) and—following a slash—the position of the pattern of which it is a variation.

Thus, the label
-2T4/1
represents a major second (-2) downward transposition (T) of a four-pitch pattern
(4) beginning on pitch 1 (1) of the list of pitches provided as argument to the com-
pression function. Although one might argue that this multi-component symbol it-
self contains so many aspects of the pattern that it does not represent a true
reduction of information, I remind readers that the programs I describe here should
not be measured by their capacity to compress data so much as by their ability to
identify the true information content present in that data. In fact, a large sequence
of numbers would reduce to a single number if the spaces were removed. Of
course, this number would also be unreadable without a guide of some sort to ex-
tract each individual number. The point, however, remains that each individual
member of a symbol used to represent data should be counted separately and com-
pared to the original list in order to truly discover the compression involved.
Labels such as -2T4/1 will therefore count as six rather than one member of a se-
quence. Of course, when dealing with very large repeating or repeating-with-
variation patterns, such representations can also significantly compress data.
The compression function does not analyze for rhythm, dynamics, channel, or
other parameters, in order to keep the process simple for demonstration. All
symbol-designated patterns are assumed to have the same durations, with ontimes
figured as immediately following the preceding pattern. Thus, the pitch pattern 60,
62, 64 (2, 2 in intervals) followed by a 4T3/1 indicates that the pitch pattern 64,
66, 68 (2, 2) follows it in order. I switch back and forth from pitches to intervals
here because the actual program uses pitches as input but processes the informa-
tion as intervals.
The compression function operates by command line using the format
(compression '(64 64 64 65 66 67 67 67 67))
where the term compression represents the compressing function and the list of
numbers represents the list of pitches to be compressed. The return from such a
function call, as in
(64rep3/1 65 2irt4/1 67)
indicates that the original list of nine pitches reduces to a list of four indicators—
two symbols and two non-involved intervening pitches. Note that the repeating
(rep) symbol begins with the repeating pitch 64. Note as well that because the
repeating pitches begin in place, the place indicator reads as 1.
The compression-ratio function produces a percentage ratio of the number
of pitches used as argument and the number of pitches and symbols returned.
Thus, the call
(compression-ratio '(64 64 64 65 66 67 67 67 67))
returns
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 95

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 95

0.44,
that is, the ratio of returned symbols and pitches to pitches alone.
This simple AIT compressing program functions on lists of pitches of arbitrary
length, typically representing solo melodic lines. Nonetheless, the program can also
compress sublists of, say, pitch and channel (timbre choice), pitch and duration,
and grouped pitches for harmony. Even in its pitch-alone matching state, however,
the program reveals interesting aspects of information relativity between various
works by composers spanning several centuries, as mentioned earlier in this chap-
ter. For example, the music shown in Figure 2.8 provides the following output:
(compression-ratio (get-pitches bach))
0.5128205128205128
(compression-ratio (get-pitches beethoven))
0.5
(compression-ratio (get-pitches brahms))
0.6428571428571429
(compression-ratio (get-pitches cope))
0.625
(compression-ratio (get-pitches krenek))
0.8666666666666667
(compression-ratio (get-pitches mozart))
0.45
(compression-ratio (get-pitches webern))
0.875
where the Beethoven and Mozart examples have relatively low information content,
Bach has moderate information content, and Brahms, Cope, Krenek, and Webern
have relatively high information content, covering a range of 0.4 to 0.88 overall.
The function get-pitches shown above follows the same recursive process
that add-one and translate-into-morse-code used:
(defun get-pitches (note-events)
(if (null note-events) ()
(cons (second (first note-events))
(get-pitches (rest note-events)))))
As previously described, each note-event contains five elements: ontime, pitch, du-
ration, channel, and loudness. Therefore, the above function could retrieve any of
these elements by replacing the use of second in line 3 with first, third,
fourth, or fifth, depending on the element desired, with the name of the func-
tion in lines 1 and 4 also changed to reflect the retrieved element (as in get-on-
times, etc.).
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 96

96 HIDDEN STRUCTURE

The DMAIT code called Multigraph on the CD-ROM accompanying this book
compresses pitch, rhythm, texture, and dynamics and presents a chart of these
parameters, as shown in many figures of this Chapter and in Figure 2.19. The verti-
cal lines in this graph indicate new entrance points for pitches in this music and
not equal time increments (though horizontal distance still represents linear time).
The counterpoint of the four lines here demonstrates its varying—based on all pre-
vious material presented—information content.

FIGURE 2.19 Sample Multigraph output.

03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 97

LISP, ALGORITHMIC INFORMATION THEORY, AND MUSIC 97

Multigraph uses a variable that represents the largest segment of music used for
determining the information content. Typically set to twenty onset points, this vari-
able can be increased to any number desired. Because the variable determines the
largest segment and the program then uses segments of successively smaller sizes
down to three, the larger this variable, the longer the information determination
will take. With a window of twenty the program can still take several minutes to
complete its calculations, even on very fast machines. When the window is set
higher, the program may require several hours to compute and graph its findings,
depending on the speed of the computer’s processor.

CONCLUSIONS
This chapter has concentrated on information theory—an important field in com-
munications theory and artificial intelligence, among others—and on how, in partic-
ular, algorithmic information theory proves useful for analyzing music. In contrast,
the following two chapters focus on musical set theory, a hybrid of mathematics
and music. The examples given and the code shown will be more specific than
those in this chapter and likely strain most non-musician’s abilities to maintain
their levels of understanding. I hope that the introductory material to each of these
two chapters will enable those less skilled in reading and understanding music to
follow the arguments with a minimum of confusion.
03_DAS_Chap2_pp45-98 1/29/09 1:33 PM Page 98
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 99

THREE
Register and Range
in Set Analysis

I also want to say that I did not invent the unordered pitch-class set. That was
the creation of a far higher power—and I don’t mean Milton Babbitt.

Allen Forte, “Banquet Address: SMT, Rochester 1987” (1989), 97

This chapter begins with a brief outline of mathematical and musical set theory.
Both of these representations/processes involve pattern matching, a principal fo-
cus of algorithmic information theory, discussed in Chapter 2. This chapter contin-
ues by exploring many of the important elements of music that pitch-class
notation—integral to musical set theory—ignores as analysts attempt to discover
their most effective reductions. The duodecimal process presented here resolves
such discrepancies, and I use several works from the standard post-tonal repertory
to demonstrate how it does so. Each of these processes—set theory and duodeci-
mal notation for register inclusion—fall under Principle 1, as described at the out-
set of Chapter 1. The modest goals of the current chapter allow both those familiar
with mathematics but unfamiliar with music, and those familiar with music but un-
familiar with mathematics, to find common ground before the book takes more ad-
venturesome directions in Chapter 4 and beyond.

BASICS OF SET THEORY

Set theory belongs to a branch of mathematics known as discrete mathematics. Dis-
crete mathematics focuses on fixed, discontinuous numbers in contrast to, for ex-
ample, algebra and calculus, which cover the continuous domain of all real
numbers. The study of prime numbers represents a good example of discrete math-
ematics in that only numbers that are divisible by themselves and 1 qualify as
prime. Discrete mathematics has many branches, such as number theory, game
theory, and group theory, as well as set theory.
Mathematical set theory—along with logic and predicate calculus—represents
one of the axiomatic foundations of mathematics. A mathematical set is denoted by
numbers enclosed in braces (curly brackets) and separated by commas, as in

99
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 100

100 HIDDEN STRUCTURE

{0,1,2}. Set theory relates sets in many ways. For example, the set {0,1,2} is a subset
of the set {0,1,2,3,4}, in that all three members of the first set belong to the second
set (common membership). The set {5,6,7} is not a subset of the set {0,1,2,3,4}, be-
cause none of the members of the first set belong to the second set. Such simple
comparisons represent one of the manifold ways in which set theory relates sets
according to membership.
Figure 3.1 presents several methods in which sets can be more formally compared.
The notation 12 ∈ {8,10,12}, for example, indicates that 12 is a member of the set
{8,10,12}. Conversely, the notation 11 ∉ {8,10,12} states that 11 does not belong to
the set {8,10,12}. The notations ⊂ and ⊆ indicate two forms of subsets—sets whose
members all belong to another set. A proper subset (⊂) refers to a range of possible
subsets belonging to another set that do not include exactly and only that set itself.
In contrast, the second notation for a subset here (⊆, referred to as improper)
means that the range of possible subsets of one set does exactly and only include
another set. The symbol ⊄, in contrast, indicates that a set is not a subset of an-
other set. The special symbol ∅ defines the empty set {}. Two of the many opera-
tive mathematical set-theory notations, presented last in Figure 3.1, represent
important relationships between two sets: ∪ indicates a union of two sets, creating
a third set that contains all the elements of both original sets, while ∩ indicates an
intersection between two sets, creating a third set containing only the elements that
both the original sets have in common. The following five examples present logical
applications of these symbols:

is an element

is not an element
 is a proper subset
 is a subset
 is not a subset
 the empty set; a set with no
 union
 intersection
FIGURE 3.1 Symbols representing several ways in which sets can be more formally
compared.
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 101

REGISTER AND RANGE IN SET ANALYSIS 101

{1,4,5} ⊂ {0,1,2,3,4,5}
{1,4,5} ⊆ {1,4,5}
{1,4,5} ⊄ {0,2,5}
{1,4,5} ∪ {0,4,5} = {0,1,4,5}
{1,4,5} ∩ {0,4,5} = {4,5}
Combinations of these and other relationships in mathematical set theory produce
extremely valuable results and principles that impact all of the various forms of
mathematics. The following example provides a simple demonstration, with sets
here given variable letter names so that the examples extend beyond particular
sets to sets in general:
<if
A⊆B∩C
then
A⊆B
and
A⊆C
In order to facilitate understanding of such set-theory operations, in 1880 John
Venn (1834–1923) invented a visual process that demonstrated set containments
and overlaps. Figure 3.2 presents a Venn diagram of the logical deductions presented
above. Because A is a subset of the intersection of B and C (A ⊆ B ∩ C), then A is
also a subset of B (A ⊆ B) and a subset of C (A ⊆ C).

B {A} C

FIGURE 3.2 A Venn graphic of set theory logical deductions.

04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 102

102 HIDDEN STRUCTURE

This brief introduction to mathematical set theory does not do justice to this ex-
traordinary field of study. I have limited my discussion of it simply because musical
set theory does not typically use these symbols or operations. However, musical
set theory not only invokes the principles of mathematical set theory, but also in-
cludes many of its own unique comparison techniques, as we shall soon see. For
those musicians interested in pursuing the mathematical side of set theory in more
detail, several dozens of good books await your study, a number of which I include
in the bibliography (Devlin 1993; Ferreirós 1999; Halmos 1974; Lawvere and Rose-
brugh 2002; Levy 1979).
Musical set theory was initially introduced by Milton Babbitt (1960 and 1961),
embellished by Allen Forte (1973), and further developed by John Rahn (1980) and
George Perle (1991). Babbitt was

concerned primarily with those set properties—pitch class and intervallic, order-
preserving and merely combinational—and those relationships between and
among forms of the set which are preserved under the operations of the system,
and which—in general—are independent of the singular structure of a specific
set. Here, to the end of discovering certain compositional consequences of set
structure, the concern will be with those attributes of set structure which main-
tain under the systematic operations only by virtue of the particular nature of a
set, or of the class of sets of which it is an instance, together with a particular
choice of operations. (Babbitt 1961, 129)

Musical set theory follows many of the same tenets of mathematical set theory.
However, several special conditions apply in musical set theory that do not apply
in mathematical set theory. For example, musical set theory invokes the notion of
pitch class (presented briefly in Chapter 2), where register (octave) no longer ap-
plies and where the pitch C-natural equals 0, C-sharp equals 1, and so on to B-
natural, which equals 11. The process of reducing out the register of a pitch (i.e., all
C-naturals belonging to the pitch class 0) is often termed modulo 12, modulo being
a mathematical process that returns only the remainder when dividing one number
by another number. All doublings as well as registration disappear when reducing
pitches to pitch-class sets. Representing pitch in this way initiates a process that
attempts to reveal similarities between sets of pitches that otherwise may appear
quite different in number form and/or musical notation. No pitch class should ap-
pear twice in a pitch-class set.
To help differentiate musical set theory from mathematical set theory, musical
set theory typically uses brackets rather than curly braces for set notation, as in
[0,4,7], the set for a C major triad in root position. Other easily recognizable sets in-
clude the C minor triad [0,3,7] and the C dominant seventh chord [0,4,7,t] where “t”
represents the number 10 (“e” represents the number 11), which allows a single
digit/ letter symbol to be used for each of the eleven pitch classes.
In mathematical set theory, the order of the elements within a set is irrelevant.
However, the notion of ordered and unordered sets is important in musical set the-
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 103

REGISTER AND RANGE IN SET ANALYSIS 103

ory. Ordered sets follow the order of pitches and pitch classes found in the music
under analysis. On the other hand, unordered sets follow ascending order, ignoring
the order of pitches and pitch classes found in the music being analyzed. A simple
example may be useful here. As noted in Chapter 2, pitches are typically repre-
sented as 60 for middle C, 61 for middle C-sharp, and so on. Thus, an ordered pitch
set could appear as [66,69,62], given that this represents the order of these pitches
in the grouping selected from the music for this set. The ordered pitch-class ver-
sion of the [66,69,62] set would then be [6,9,2]. The unordered pitch-class version
of the [66,69,62] set would then be [2,6,9].
In traditional music theory, triads are typically analyzed from their root pitches
upward. Root positions of triads generally share a common feature: they have a
smaller range between their outer pitches than do triads in inversion. Figure 3.3
provides an example of this. Here we see a D major triad in its three incarnations:
root position, first inversion, and second inversion. Note that the perfect fifth be-
tween the outer pitches of the root-position triad spans a smaller interval than the
minor sixth and major sixth that form the outer boundaries of the two inversions.
The same is true in differentiating what is called the “normal” form of pitch-class
sets, with the exception that smaller intervals between inner pitch classes can also
count (described shortly), though the outer pitches of the set take precedence.
As an example, the ordered set [6,9,2] has a distance of eight between its outer
pitch classes, 6 and 2. This is figured by incrementally counting upward from pitch
class 6 to pitch class 2, beginning again at 0 after 11 (or “e”): 7, 8, 9, t, e, 0, 1, 2, or
eight steps. This counting upward is important, because counting downward from 6
to 2 does not include pitch class 9, necessary because we want to include all mem-
bers of the set in our computations.
Reviewing Figure 3.3 while reading the following description may help readers
understand these machinations more clearly. Unordering the set [6,9,2] creates a
total of three sets with the farthest-separated numbers inclusive of the remaining
pitch class: the original [6,9,2], [9,2,6], and [2,6,9], counting upward from the low-
est pitch class in each case. The pitch-class set [6,9,2] has a distance of eight be-
tween its outer pitch classes, 6 and 2, as previously shown. The pitch-class set
[9,2,6] has a distance of nine between its outer pitch classes, 9 and 6, proved by
counting upward from pitch class 9 to pitch class 6. However, the pitch-class set
[2,6,9] has a distance of only seven between its outer pitch classes, 2 and 9. Thus,
the pitch-class set [2,6,9] represents the normal (unordered) form of the ordered

w # ww
& # www # ww w

FIGURE 3.3 Root positions of triads have a smaller range between their outer
notes.
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 104

104 HIDDEN STRUCTURE

set [6,9,2]. A glance back at Figure 3.3 proves this yet further, because the root-
position D major triad represents pitch classes 2, 6, and 9, in that order, from the
lowest pitch upward.
Using a traditional twelve-hour analog clock face provides a much easier way to
visualize numerical pitch-class relationships (see Figure 3.4) and to compute the
normal form of sets. The internal x’s here denote the pitch-class set [6,9,2], called
“unordered” at this point because once placed on the clock face, the original order
of the pitches in the music can no longer be ascertained. The set [2,6,9] clearly rep-
resents the normal form because beginning and counting clockwise from pitch
class 6 produces an distance of eight between outer pitches, and beginning and
counting clockwise from pitch class 9 creates a distance of nine between outer
pitches, both of which are larger than the clockwise distance of seven between the
outer pitches of [2,6,9].
Figure 3.5 presents another set for normal form pitch-class analysis. The [1,6,t]
pitch-class set shown here does not produce the smallest outer range, because the
outer pitch classes produce a distance of 9, larger than the smaller range of 7 pro-
duced by [6,t,1]. Using clock faces to discover the various forms of a set is analo-
gous to using Venn diagrams in mathematical set theory. Although they are mainly
graphic devices, both Venn diagrams in mathematics and clock faces in music make
analysis by human beings far easier and often even inviting.
Up to this point we have been reading pitch-class sets clockwise. Reading sets
counterclockwise adds a powerful comparative tool for discovering more similari-
ties between apparently diverse pitch-class sets, particularly post-tonal pitch-class
sets. This process, called mirror inversion, means that all intervals appear pre-

0
11 1

10 2
x

9 x 3

4
8

7
x 5
6

FIGURE 3.4 A clock face arranged to allow clearer visualization of pitch-class

relationships.
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 105

REGISTER AND RANGE IN SET ANALYSIS 105

cisely inverted in musical notation. In Figure 3.5, two sets have the same distance
between outer pitches: [6,t,1] figured clockwise, and [1,t,6] figured counterclock-
wise. In situations such as this, the pitch-class set with the smallest internal inter-
vals packed toward the pitch of origin wins, Thus, pitch-class set [1,t,6] (read
counterclockwise from 1) succeeds because it has an internal interval of 3 (1 to t
read counterclockwise equals 3) rather than 4 (6 to t read clockwise). At this point,
we transpose this smallest form ([1,t,6]) to 0 by reading the intervals counterclock-
wise and beginning with “0” replacing “1,” creating what we call its “prime form,”
the set [0,3,7].
In like form, the pitch-class set [2,6,9] when figured as above on a clock face and
transposed becomes [0,3,7] in prime form. This procedure, while it might seem
somewhat artificial, closely resembles the process we use when analyzing for musi-
cal function in tonal music, where a pitch set in one register reduces to a V7 in C
major (e.g., [G,B,D,F]), and a pitch set in another register reduces to a V7 in D ma-
jor (e.g., [A,C#,E,G]). Because major and minor keys do not typically exist in post-
tonal music as they do in tonal music, [0,3,7] represents all major and minor
triads and their inversions. This apparent contradiction should not be particularly
bothersome because the reductive process is intended to reveal similarities, not
equivalencies.
Figure 3.6 presents a straightforward example of using set theory to discover
similarities in post-tonal music. The four chords in this example appear very differ-
ent both in pitch content and in register. To a discerning ear, however, they sound
similar in many ways. Each of these sets reduces to the same prime form

0
11 1
x
10 x 2

9 3

4
8

x 5
7
6

FIGURE 3.5 The [1,6,t] pitch-class set does not resolve to the pitch-class set
[0,5,9] (9 distance when rotated to 0) because of the smaller range of
[6,t,1] (7 distance and equating to [0,4,7] when rotated to 0).
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 106

106 HIDDEN STRUCTURE

w w # # www
& www # wwww b # www

? b ww bw w ww
w #w w

FIGURE 3.6 A straightforward example of the use of set theory to find similarities
in post-tonal music.

[0,1,3,6,8,9] following the processes just described. Figure 3.7 shows each chord in
Figure 3.6 represented on a clock face by the process used to discover the prime
form provided in the figure caption.
There are, then, typically three forms of pitch-class sets: ordered, normal, and
prime. The ordered form indicates that the order of pitches in the set matters (thus,
[0,1,3] and [1,3,0] represent different sets even though they contain the same pitch
classes). The normal form (unordered because it no longer reflects the order in the
music) accounts for the normal inversions of, for example, triads, and places sets
so that they cover the smallest overall range. The prime form then accounts for
mirror inversions of sets, finding the smallest outer range, packing pitch classes to-
ward the pitch class of origin, and transposing the result to begin on 0. These defi-
nitions generally follow those of Babbitt (1961) and Forte (1973, see esp. 4–5). In
summary, then, the following presents one instance of these three forms of the
same pitch-class set:
[6,9,2] ordered form
[2,6,9] normal form
[0,3,7] prime form
The above set, initially a major triad in first inversion (ordered form), appears as a
root-position normal form, and then as a minor triad in inverted and transposed-to-
zero form.
Unfortunately, the fact that all major and minor triads ultimately reduce to a mi-
nor triad in prime form means that an unambiguous major triad expressed in post-
tonal music will be lost during analysis. It is not that the prime-form minor triad is
objectionable, but that the normal-form major triad cannot be clearly discerned
from the minor triad in the prime form. Many analysts have reverted to a two-form
version called a representative form: Tn, read as normal form transposed, and TnI,
read as normal form transposed with inversion taken into account (see Rahn 1980,
75–76). Rahn also uses the term type, while still others use the terms pitch-structure
(Howe 1965) and chords (Regener 1974). Morris (1987, see esp. 79) offers these
and other classification systems of pitch-class sets. These extra forms are used to
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 107

REGISTER AND RANGE IN SET ANALYSIS 107

a)
a. 0 b.b ) 0
1 1
11 x 11 x
x
10 2 10 2
x x

9 3 9 x x 3

x 4 4
8 8
x x x
7 5 x 5
7
6 6

c)
c.
0
d)
d.
0
1 11 1
11 x
x x x
10 x 2 10 2

9 3 9 x x 3

x x 4
4
8 8
x
x 5 x 5
7 7
6 6

FIGURE 3.7 (a) [e,0,2,5,7,8] is [0,1,3,6,8,9] when rotated one number clockwise;
(b) [6,7,9,0,2,3] is [0,1,3,6,8,9] when rotated six numbers clockwise;
(c) [7,6,4,1,e,t] is [0,1,3,6,8,9] when rotated seven numbers counter-
clockwise; (d) [0,e,9,6,4,3] is [0,1,3,6,8,9] when rotated twelve num-
bers counterclockwise.

account for works in which the two forms of many sets, such as the major triad—
original and inversion—appear in important places in the music being analyzed and
thus should be reflected in the analysis of that music. Also, valuable characteristics
of same-prime-form sets can be revealed when comparing only normal forms (e.g.,
numbers of common tones, etc. [see Straus 2005, 71–72]).
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 108

108 HIDDEN STRUCTURE

One simple solution that many of my theory colleagues and I have adopted—
that of simply transposing the currently accepted normal form to zero—is called
the t-normal form (for transposed normal form). Thus, the ordered pitch-class set
[6,9,2] would translate to the following forms:
[2,6,9] normal form
[0,4,7] t-normal form
[0,3,7] prime form
Keeping track, then, of the manners in which ordered, normal, t-normal, and prime
forms interact with one another in music becomes much easier and more obvious.
The interplay of t-normal and prime forms expressed in this way becomes immedi-
ately apparent in works where composers juxtapose pitch-class sets of the same
prime form but different t-normal forms in their music.
From here on I will be using both this zero-based t-normal form and the more
traditional normal form. This will become increasingly important in the following
discussions in this chapter when the mathematical creation of sets helps us to un-
derstand how logical new sets—otherwise considered unrelated—result from pair-
ings of sets that appeared previously. Extending “normal order” to t-normal order
has no direct effect on computer applications of music analysis. However, as will be
seen in this and later chapters, the manner in which the programs accompanying
this book on CD-ROM use this “t-normal order” has a substantial effect on the abil-
ity to clearly describe a program’s operation and the principles that enable that
program to work effectively.
Translating pitch sets into t-normal pitch class sets can be accomplished in Lisp
quite easily, as the following code demonstrates:

(defun translate-to-t-normal-pitch-class-set (set)

(translate-to-pcs (sort-and-clean set)))
(defun sort-and-clean (set)
(my-sort #'< (remove-duplicates (modulo12 set))))
(defun modulo12 (set)
(if (null set)()
(cons (mod (first set) 12)
(modulo12 (rest set)))))
(defun translate-to-pcs (mod-set
&optional (first-set (first mod-set)))
(if (null mod-set)()
(cons (abs (- first-set (first mod-set)))
(translate-to-pcs (rest mod-set) first-set))))

The function translate-to-t-normal-pitch-class-set here acts as a top-

level operator of the two functions sort-and-clean and translate-to-pcs.
The function sort-and-clean simply maps modulo12 (a simple recursive func-
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 109

REGISTER AND RANGE IN SET ANALYSIS 109

tion that uses the Lisp primitive mod to reduce all elements of its set argument to
within the range 0–11) onto its argument, removes all duplicates, and then sorts
the result into ascending order. The function my-sort (available in many of the
code files on the CD-ROM accompanying this book) bypasses the side effects of
Lisp’s somewhat unpredictable standard sort function. The function translate-
to-pcs then transposes the set to begin on 0 by subtracting the first element of
the set from each of its members. The use of Lisp’s &optional keyword in the first
line of translate-to-pcs allows for optional arguments that users do not have
to use when calling the function and that here enable the variable first-set to
continue to represent the first element of mod-set even though mod-set is slowly
diminishing in size, owing to recursion. The arguments to &optional come in lists
in which the first element is the name of a variable and the second element the
default data contained in that variable. Running this code with varying arguments
produces:
? (translate-to-t-normal-pitch-class-set '(60 64 67))
(0 4 7)
? (translate-to-t-normal-pitch-class-set '(60 63 67))
(0 3 7)
? (translate-to-t-normal-pitch-class-set '(60 63 67 70))
(0 3 7 10)
? (translate-to-t-normal-pitch-class-set '(64 67 79 84))
(0 4 7)
Note that the last data converted here contains octave doubling and an inversion
that nonetheless convert to the same result as the first test.
Set-theory analysis provides many other tools for understanding post-tonal mu-
sic. For example, interval vectors (counts of all the intervals present in a set) pro-
vide interesting ways for analysts to relate prime forms of sets that may otherwise
seem unrelated. Vectors are represented by six-digit counts of the intervals of a mi-
nor second, major second, minor third, major third, perfect fourth, and augmented
fourth; the remaining intervals in the octave are considered mirror inversions of
these intervals. Thus, the set [0,1,3] has the vector 111000, because 0,1 is a minor
second, 0,3 is a minor third, and 1,3 is a major second (all of the possible intervals
contained in the set). As an example of vector relationship, consider the two sets
[0,1,3,7] and [0,1,4,6], both of which have the same vector, 111111. Other pitch-
class-set comparison techniques include similar but not equivalent prime forms of
pitch-class sets indicating variation techniques in use. Analyzing interval-class sets
as well as pitch-class sets can also help analysts discover interesting contrasts and
similarities in post-tonal music. Processes such as these represent but a few of the
ways that musical set theory can aid in our fundamental understanding of post-
tonal music.
Although musical set theory does not reveal function in the way that tonal analy-
sis does in tonal music, using musical set theory to decipher similarities between
complex and otherwise apparently unrelated groupings of pitches often provides
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 110

110 HIDDEN STRUCTURE

insights into music otherwise considered impenetrable. The manner in which com-
posers limit their compositions to just a few of the 208 possible prime forms of
chords between trichords (three-pitch groupings) and nonachords (nine-pitch
groupings) indicates that set theory provides a valuable tool for analyzing post-
tonal music. (The other chord forms are tetrachords, pentachords, hexachords,
septachords, and octachords, respectively.)
One of the biggest problems analysts face when using set theory to analyze
post-tonal music is how to group music into appropriate collections for revealing
the optimum number of logical set relationships. Computer programs can aid signif-
icantly in this process by using extraordinary accuracy and speed to remove the
drudgery from what can otherwise be an enormously time-consuming process of
trial and error. Even brute-force computer programs that simply compare sets of
ever-decreasing sizes to find the sets that appear most often can minimize the ef-
fort of what would otherwise take an analyst weeks or even months to do by hand.
Of course, such computer programs cannot make intuitive leaps or musical deci-
sions. They do, however, offer analysts a palette of possibilities from which they
can then choose the most promising alternative.
Before describing more innovative programs for analyzing post-tonal music, I will
first demonstrate a function (define-and-lexicon-all-patterns in the file
called database in the folder called sets database) that groups and analyzes music
for sets and—what is possibly more important—set variances, returning its best
guess for one or more sets as the basis of the music under study. This program
uses a straightforward previously discussed technique known as pattern matching
and includes several processes that are important for understanding this and sub-
sequent chapters. The primary focus of these processes involves separating group-
ing and pitch-class-matching programs into discrete tasks and comparing only
groupings that match. This focus will eliminate the need to match all possible
groupings, thus reducing the time necessary for analysis and increasing the poten-
tial for varying the parameters for subsequent analyses. Furthermore, we can col-
lect similar—in whatever ways we wish to define—pitch-class patterns and include
them as well by selecting other, related lexicons. Thus, by collecting patterns be-
fore comparing them and carefully distributing them into appropriately named lexi-
cons, we not only speed the process of pitch-class matching, but we also make a
wide range of pitch-class matching possible without having to redo the grouping
process. Pattern matching in this way reveals a wide variety of possibilities not
otherwise evident in pattern-matching systems that require grouping and matching
simultaneously.
As mentioned previously, equivalent but transposed pitch-class sets can be rep-
resented by Tn, where “T” represents the word transposition and the subscript “n”
indicates the distance between two fundamentally equivalent pitch-class sets. For
example, the normal-form sets [2,4,5,7] and [5,7,8,t] have three half steps separat-
ing them, the second set therefore having a T3 relationship to the first set. In this
book, we simply transpose both sets to the same t-normal form ([0,2,3,5]). In either
case, the process involves addition and/or subtraction of two pitch-class sets that
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 111

REGISTER AND RANGE IN SET ANALYSIS 111

are assumed to be equivalent. The notion of adding and subtracting the elements of
two non-equivalent sets to achieve a third set, however, is less intuitive. For exam-
ple, consider the same pitch-class set ([2,4,5,7]) added to the pitch-class set
[2,5,9,t]. The result ([4, 9, 14, 17]), when reduced modulo 12 and transformed to
t-normal form, produces a new pitch-class set of [0,2,3,7], similar in many ways, but
not equivalent to, the two sets that produced it. Although this creation of new
pitch-class sets from the addition or subtraction of two dissimilar pitch-class sets
may seem far-fetched for analysis, when three different pitch-class sets appear as
the primary sets of a work under analysis and two of these may be added or sub-
tracted to produce the third, the process seems more purposeful.
Multiplying and dividing pitch-class sets by each other may seem an even more
remote process than adding and subtracting such sets. However, these processes
become immediately less remote when we encounter them in music. As an exam-
ple, Figure 3.8 presents the pickup and opening four bars of Schoenberg’s op. 19,
no. 6 from Sechs kleine Klavierstücke.
The t-normal forms of the first two trichords here, [0,3,5] and [0,5,7], show little
relation to the third trichord’s t-normal form, [0,1,8] ([0,1,4] in prime form). How-
ever, when the [0,3,5] and [0,5,7] t-normal forms are multiplied together (i.e., when
multiplying the aligned pitch classes of each set modulo 12), the new set results in
the t-normal form [0,3,e], or [0,1,4] in the prime form. Thus, the three sets share a
common multiple. I have included a small program called SetMath on the CD-ROM
accompanying this book that accepts either normal forms or unordered pitch-class
sets and provides a series of results in the following form:
(run-sets '(0 3 5) '(0 5 7))
Added sets: (0 8)
Subtracted sets 2 from 1: (0 10) Subtracted sets 1 from 2: (0)
Multiply sets: (0 3 11)

# œœ ww ˙˙ # ˙˙ ww
# ˙>
˙
œ˙˙ # œ Œ # œœ
&c œ w ˙ ˙
Ó
w œ
π p
Piano Ø j
Ó #œ. œ œ #œ Œ Ó
&c Œ Ó Ó
n nn ˙˙˙ ˙˙
˙ Œ n nn ˙˙˙ .. ˙˙
˙

FIGURE 3.8 The opening 4 measures of Schoenberg, Six Little Piano Pieces, op. 19,
no. 6.
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 112

112 HIDDEN STRUCTURE

Note that the t-normal forms of the two input sets use parentheses instead of
brackets and lack commas, the results of using Lisp. Note also that multiplying and
otherwise mathematically combining sets can cause duplications, resulting in an
output of fewer pitch classes.
Before proceeding further, I remind readers here of my earlier comments about
composer intent. What composers do or do not intend to include in their music,
though of interest, should not deter analysts from revealing discovered relations,
no matter how unlikely these relations may seem to the perceived concept of the
work as we know it. I further remind readers that analysis informs listening as well
as vice versa. Whether a process that exists in music can be heard or not should
not deter us from appreciating its presence.
To this point, sets in this book have primarily been comprised of pitch classes.
Readers should be reminded, however, that sets may consist of representations of
any parameter of music. For example, rhythmic information such as sets of dura-
tions in thousandths of a second [1000,4000,2000,8000], channel settings [1,4,2,3],
dynamics [55,60,75,45], and so on can all produce useful relationships when per-
muted and compared. I have even found interesting results when creating formal
sets as in [a,a,b,b,c,a,b] for the purpose of comparing the forms of entire works.
Such formal collections remind us that sets may consist of many different kinds of
symbols other than numerical information, as in this case, where the alphabetical
letters represent phrases of a particular type of musical thematic or harmonic data.
Musical set theory involves far more than I have described here. In fact, the prin-
ciples expressed thus far in this chapter barely cover the fundamentals of what can
be a very complex and often personal approach to understanding post-tonal music.
However, given that this book attempts to cover a wide variety of approaches, I
leave it to individual readers to explore musical set theory further, using books of
their own choosing or those in the bibliography (particularly Forte 1973; Lewin
1987; Straus 2005).

REGISTER
Many post-tonal composers use register in important ways in their music. Unfortu-
nately, register often gets lost when analyzing this music using pitch-class sets. For
example, register does not accompany pitch-class sets when they are reduced to
their normal and prime forms. This is unfortunate, because equivalent prime-form
sets with significant registral differences can often sound as contrasting as different
prime-form sets that share registral arrangements (Bruner 1984; Gibson 1986 and
1988; Isaacson 1996). At the least, when viewed in their registral contexts, some
equivalent prime forms seem more equal than other equivalent prime forms (Kuusi
2003). I have found in both my analysis and my composition that retaining register
information during grouping and reduction allows me to compare pitch-class sets
more meaningfully.
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 113

REGISTER AND RANGE IN SET ANALYSIS 113

In the following section, I present a duodecimal notation for representing pitch-

class sets, a notation that—by virtue of its radix of 12—allows both register and
pitch class to share one simple notation. After demonstrating this notation, explain-
ing its mathematical manipulations, and presenting a series of simple musical
excerpts by which I hope to convince readers of the usefulness of this notation, I
then provide musical examples of my own and other composers’ music from both
analytical and compositional perspectives. I also discuss several register-related
analytical parameters, including range, interval mapping, interval summations, and
linear interval vectors, each of which provides further insight into the music under
analysis.
Tonal music offers many examples of the importance of register. For instance,
the beginning of Claude Debussy’s La cathédrale engloutie (1909), shown in Figure 3.9,
presents two related but varied musical ideas that are distinguished in part by
their registral separation. One idea appears in both hands at the onsets of mm. 1, 3,
and 5, with its wide separation of register, resultant open fourths and fifths, and
slow rhythmic progression. The second idea appears in the center register of the
keyboard with close-knit planing in both hands. Interestingly, the first chord of this
prelude is echoed in the second chord, which follows immediately after the first,
with no difference in their pitch-class content, only in their lowest pitches—the

Profondément calme (dans une brume doucement sonore)

√w . œœ √w . œœ
6 ww . œ œ œœ œœœ œ ww . œ œ œœ œœœ œ
& 4 Œ œœ œœ œ ∑
Œ œ œ
œ œ œ
π œ œœ œœ œ œœ
Œ œ œ œœ œ
Piano

? 46 Œ œœ œœ œ Œ
œ œ
ww . ww . œ ww .
w. w. œ œ -̇ .w.
√w . œ ˙- œ ˙-
4
ww . œ œœ œœ œœ œ ˙ œ ˙
& ∑ œœ œ œ œ
Œ œ œ
Œ Œ œ œ œœ œ œ
œ œ -̇ œ -̇
? œ œ
ww . œ ww .
w .. œ œ -̇ w.
FIGURE 3.9 The opening measures of Claude Debussy, La cathédrale engloutie
(The Sunken Cathedral), 1909.
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 114

114 HIDDEN STRUCTURE

second chord being a closed-position inversion of the initial dyad. Clearly these
two basically equivalent chords play extremely different roles, as is highlighted by
their registral variance.
Register matters in this work, as it does in most music. Even in Bach chorales,
register can determine the inversion, position (open or closed), voice-leading, and
certainly the resonance of a harmony or passage. Most tonal and post-tonal analyti-
cal techniques, however, exclude register from their reductive notations. I have
long sought a way to retain register in these analytical notations, especially in post-
tonal music, where register often plays such an important role. Although one could
make the same kind of case for rhythm, timbre, and other important elements in
post-tonal music, register seems so closely tied to the kinds of pitch analysis in-
volved in post-tonal pitch-class processes that it has always seemed to me para-
doxical not to include it—if indeed a logical way to do so can be found.
Duodecimal numbering offers a more inherently musical way to notate equal-
tempered pitch classes than does decimal numbering. The duodecimal system
contains twelve different numerals, rather than the ten numerals of the decimal
system, making duodecimal equivalent to the twelve pitch classes of the chromatic
scale. (Note here that the terms dodecimal and duodecimal mean roughly the same
thing. I have chosen to use the word duodecimal because mathematicians clearly
prefer duodecimal, and because I wish to clearly differentiate this term I describe
from the term dodecaphonic, occasionally used to describe twelve-tone or serial
music.)
Duodecimal counting resembles decimal counting in many ways. The primary
difference between the two systems occurs at the point in which the numbers ex-
tend beyond the limit indicated by the name of the numbering system. For exam-
ple, the ten decimal numerals range from 0 to 9 inclusive, for a total of ten, before
transcending to its “tens,” such as 10, 11, and so on. In contrast, the twelve numer-
als of the duodecimal system count from 0 through 9 and then on to “t” and “e”
inclusive, for a total of twelve (again, “t” and “e” represent ten and eleven, respec-
tively) before transcending to its “tens,” as in 10 and 11, with these latter numbers
equating to 12 and 13 in the decimal system. For non-musicians, counting in
duodecimal can be tedious and confusing. However, for musicians the match of
duodecimal numbers to the twelve-pitch equal-tempered scale makes a nice fit. For
every duodecimal number, the leftmost digit represents the octave, and the right-
most digit indicates the pitch class within that octave (with the single-digit num-
bers 0–e having an implied 0 register).
Figure 3.10 provides a straightforward conversion chart that relates decimal to
duodecimal notation. The left-hand number in each pair in a column represents
decimal notation, and the parallel right-hand number its duodecimal equivalent.
Figure 3.11 presents a possibly more musically meaningful way to think in terms of
duodecimal numbers, with each horizontal row representing a complete octave.
The duodecimal number 60 equates to middle C, with 70, 80, 90, and so on repre-
senting successive octaves above middle C. Interestingly, the term “octave” derives
from the diatonic scale, as do major and minor seconds, steps and half steps, and
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 115

REGISTER AND RANGE IN SET ANALYSIS 115

0 0 21 19 42 36 63 53 84 70
1 1 22 1t 43 37 64 54 85 71
2 2 23 1e 44 38 65 55 86 72
3 3 24 20 45 39 66 56 87 73
4 4 25 21 46 3t 67 57 88 74
5 5 26 22 47 3e 68 58 89 75
6 6 27 23 48 40 69 59 90 76
7 7 28 24 49 41 70 5t 91 77
8 8 29 25 50 42 71 5e 92 78
9 9 30 26 51 43 72 60 93 79
10 t 31 27 52 44 73 61 94 7t
11 e 32 28 53 45 74 62 95 7e
12 10 33 29 54 46 75 63 96 80
13 11 34 2t 55 47 76 64 97 81
14 12 35 2e 56 48 77 65 98 82
15 13 36 30 57 49 78 66 99 83
16 14 37 31 58 4t 79 67 100 84
17 15 38 32 59 4e 80 68 101 85
18 16 39 33 60 50 81 69 102 86
19 17 40 34 61 51 82 6t 103 87
20 18 41 35 62 52 83 6e 104 88

FIGURE 3.10 A conversion chart relating decimal to duodecimal notation.

60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 6t, 6e
70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 7t, 7e
80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 8t, 8e
90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 9t, 9e
. . . . and so on

FIGURE 3.11 A more understandable and musical way to think in terms of duodec-
imal notation.

so on. Although I certainly do not recommend here that we develop new terms to
more accurately represent these intervals, I do suggest—particularly in the case of
post-tonal analysis—that we adopt duodecimal notation to represent the twelve
pitch classes and their registers.
Figure 3.12 demonstrates how to convert decimal to duodecimal numbers and
vice versa. To convert decimal to duodecimal numbers, first divide the decimal
number by 12 to obtain the register digit, and then use the remainder as the pitch-
class digit. To convert duodecimal to decimal numbers, multiply the left-hand digit
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 116

116 HIDDEN STRUCTURE

(register) of the duodecimal notation by 12, and then add the right-hand digit
(pitch class) to the result.
Figure 3.13 shows duodecimal addition, subtraction, multiplication, and division
using arbitrary numbers. As one can see, these processes do not quickly reveal
their underlying principles, especially for those more familiar with calculating in
decimal numbers. Musicians, however, typically add and subtract using relatively
small numbers that do not pose particularly difficult problems. Indeed, octave
transposition—the most common calculation in music—is even easier in duodeci-
mal notation than in decimal notation; simply add or subtract the number of oc-
taves of the transposition from the left-hand number, leaving the right-hand number
intact.
For those readers familiar with MIDI pitch numbering, duodecimal numbers may
cause some initial confusion. For example, in duodecimal notation the number 75—
which in MIDI terms equates to the E-flat an octave and a minor third above middle
C—represents pitch class 5 (or F) in the seventh octave. However, distinguishing
these two approaches to numbering will become easier after one studies the analy-
ses presented in this chapter. The benefits of using duodecimal notation when ana-
lyzing examples of post-tonal music using wide register separations far outweigh
any initial difficulties encountered.
For those interested in how to pronounce duodecimal numbers with letters, the t
(“Tee”) and e (“Eee”) sound just as one might imagine, with combinations figured

58 decimal 4t duodecimal
divide by 12 for left digit multiply left digit by 12
right digit is remainder of above add right digit
= 4t duodecimal = 58 decimal

FIGURE 3.12 Conversion methods used for decimal to duodecimal numbers and
vice versa.

2E 2E E 18
+11 -20 * 2 /4
____ ____ ____ ____
40 E 1T 5

FIGURE 3.13 Subtraction, addition, multiplication, and division using arbitrary

numbers in duodecimal notation.
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 117

REGISTER AND RANGE IN SET ANALYSIS 117

as composites of these, as in “TenTee,” “TenEee,” “TwentyTee,” “TwentyEee,”

“ThirtyTee,” “ThirtyEee,” and so on. Knowing these pronunciations is invaluable
when discussing analyses using duodecimal numbers.
Clearly, reducing ordered pitches to ordered pitch classes causes registral infor-
mation to disappear. As previously described, this does not occur when using
duodecimal notation, where the left digit retains this information. Figuring the nor-
mal and prime forms of pitch-class sets in duodecimal notation follows the same
processes as those used in decimal notation, with the exception that the left-hand
digit of each number does not change during reduction. The left-hand digit remains
paired with the possibly changing numbers of the right-hand digit pitch classes,
as often required when figuring normal and prime forms of pitch-class sets. The
arrangements of these duodecimal numbers may also shift during reduction. Thus,
the ordered duodecimal set [52,60,73] ([2,0,3] in decimal notation) becomes
[60,52,73] ([0,2,3] in decimal notation) in unordered normal form, and [70,51,63]
([0,1,3] in decimal notation) in prime form. Although possibly confusing at first,
these processes become quite natural with continued use.
Figure 3.14 presents examples of decimal reductions. The analysis of the 2 two-
chord progressions in Figures 3.14a and 3.14b follow the standard musical set-
theory practice of first creating a t-normal form, and then a prime form of the sets.
Figures 3.15a and 3.15b then show these same reductions with registers intact as
duodecimal numbers. Again, the left-hand register digits remain with their right-
hand counterparts during this process. Other than that single rule, reducing pitch
sets to their normal and prime forms is as straightforward with registral indications
as it is without.

a.
#w b.
nw w
& c #w www b b ww b ww

? c #w ∑ ∑ ∑

ordered
pitch set: [1,3,6] [2,5,7] [3,4,6] [2,3,e]
transposed
normal form: [0,2,5] [0,3,5] [0,1,3] [0,3,4]
prime form: [0,2,5] [0,2,5] [0,1,3] [0,1,4]

FIGURE 3.14 Two two-set progressions showing pitch classes (first), t-normal form
(second), and prime (third).
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 118

118 HIDDEN STRUCTURE

a.
#w b.
nw w
& c #w www b b ww b ww

? c #w ∑ ∑ ∑

ordered
pitch set: [41,53,66] [52,55,57] [53,64,56] [52,63,5e]
transposed
normal form: [40,52,65] [50,53,55] [50,61,53] [50,63,54]
prime form: [40,52,65] [50,52,55] [50,61,53] [50,61,54]

FIGURE 3.15 Same reduction as Figure 3.14 but with registers intact in the form of
duodecimal notation.

Duodecimal notation can reveal differences between equivalent pitch-class sets

as well as similarities between different pitch-class sets. For example, in Figure
3.14a, the two sets have identical prime forms. However, these first two sets differ
significantly in register, as shown by the 4, 5, 6 and 5, 5, 5 registral indications in
Figure 3.15a, a duodecimal registration analysis of the same chord set as in Figure
3.14a. The second pair of chords, shown in Figures 3.12b and 3.15b, differs slightly
([0,1,3,] and [0,1,4]) in their prime forms, but they share identical ordered registra-
tion (5, 6, 5), demonstrating how differing sets can relate to one another by similar
registration.

RANGES AND VECTORS

Registral notation has drawbacks in that with C (or 0) as an arbitrary register shift
point, some sets appear registrally quite different when they are actually very simi-
lar or equivalent. For example, setting C to 0 forces a set such as [5e,60,5t] to look
more registrally diverse than it actually is. This set spans but two semitones, even
though the register shifts twice. On the other hand, the set [63,6e,61] covers a
range of ten semitones, even though the register remains the same for each mem-
ber of the set. To counteract this disparity, I use a range index R as in R10, where R
stands for range and the subscript 10 represents the interval distance between the
highest and lowest pitches of the set. Hence, the pitch class–register set [5e,60,5t]
has a range of R2 and the pitch class–register set [63,6E,61] has a range of R10, thus
,
distinguishing their differences in range even when their registers may suggest the
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 119

REGISTER AND RANGE IN SET ANALYSIS 119

[63,6e,61]r10

FIGURE 3.16 Range notation shown to the lower right.

opposite. This range notation then appears to the lower right of the duodecimal
pitch-class sets, as shown in Figure 3.16.
The stage of pitch-class set analysis prior to reduction to normal, t-normal, and
prime forms offers analysts an opportunity to compare interval contents of group-
ings under study. The duodecimal-numbered ordered set [50,64,56], for example,
has a decimal interval content of (16, -10). This interval content matches exactly
the (16, -10) decimal-numbered intervals of the [28,40,32] duodecimal-numbered or-
dered set. These observations would not be possible if one simply compared these
same sets in decimal t-normal form ([0,2,6]) or in unordered decimal prime form
([0,2,6]), or even in their duodecimal t-normal ([60,52,56], [40,32,28]) and prime
([50,62,56], [40,42,26]) forms, where set manipulations have altered the original
pitch-class orders.
To ensure that interval mapping follows principles consistent with the logic in
the music being analyzed, pitches must be sorted by time and—for simultaneously
sounding pitches—from lowest to highest. Although this latter process may seem
arbitrary or biased by tonal concepts, figuring chords from their lowest pitches up-
ward guarantees that the range covered and the order of the pitches for mapping
will remain consistent from one grouping to the next.
Pitch-class set reduction prior to determining normal, t-normal, and prime forms
also offers opportunities to calculate and compare interval summations of pitch-
class sets. These summations indicate the direction-sensitive intervallic differences
between the first and last pitches of groupings. Summations can prove quite valu-
able for comparing groups whose intervallic content is quite different, but whose
overall interval directions are consistent. For example, equality or inequality in the
values of these numbers, especially in non-vertical groupings, can indicate interest-
ing aspects of either the composer’s composition techniques and/or analysts’
grouping processes. Equalities often reveal interesting and possibly important de-
tails about a work’s motivic structure. Summations appear in decimal form, with
minus signs indicating downward motion.
Direction-insensitive linear (ordered) interval vectors provide further useful in-
formation for comparing the basic interval content of set motions. I use linear inter-
val vectors—each interval figured only in respect to each pitch’s immediately
following pitch—because although these linear vectors still fold intervals into the
compacted range of an augmented fourth or less, they nonetheless do not count in-
tervals from each pitch class to every other pitch class as do traditional vectors.
For example, a standard interval vector of the set [0,1,3] would be 111000, but the
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 120

120 HIDDEN STRUCTURE

linear interval vector would be 110000, the intervals between the neighboring ele-
ments of the set. Linear interval vectors often reveal interesting and important rela-
tionships between sets that would not be accessible from standard vector analysis.
For example, the ordered duodecimal sets [60,64,66,59] and [50,46,52,45] both pro-
duce the standard interval vector of 110121. However, these same sets produce
linear interval vectors of 010110 and 000111, respectively. Sets with similar shapes
have the same or similar interval vectors, while sets with different shapes that re-
duce to the same prime forms need not have the same or similar linear interval
vectors. Comparing linear interval vectors has proven quite valuable in my analysis
processes, as I will soon describe.
The program Register on the accompanying CD-ROM groups music examples al-
gorithmically. The program then extracts pitch classes from these groupings, simul-
taneously figures and retains register, and produces appropriate sets of duodecimal
numbers. Following this, the program then calculates the total range covered, as
well as the resultant prime-form set (ensuring that the register numbers neither
change nor separate from their pitch classes), along with an interval map of the or-
dered pitch set, an interval summation of the ordered pitch set, and the set’s linear
interval vector. Finally, the program rearranges and presents this information as a
single list of sublists, as in:
(((60 71 74) r13) ((7e 6t 72) (-13 4) -9) 100100))
The first sublist here—(60 71 74)—represents the prime form of the grouping in
duodecimal numbers. The r13 paired with this first sublist indicates the original or-
dered range of the grouping in numbers of half steps in absolute-value decimal
numbers. The sublist of (7e 6t 72) contains the original ordered pitch-class set in
duodecimal numbers. The (-13 4) sublist describes the interval map of the ordered
set in decimal numbers. The -9 that follows the interval map represents the summa-
tion in decimal numbers, with negative numbers reflecting downward direction.
The last entry in this list—100100—defines the linear interval vector of the grouping.
The groupings of music in the figures of this chapter represent my own or my
program’s choices and are not intended in any way to indicate optimum or even
preferable approaches. Obviously, changing the sizes or locations of the groupings
of this music can and often does produce different analytical results from those
shown. I welcome such alternate perspectives and the potential they offer for even
more insights into the music being analyzed.

COMPARISONS
Register’s lists of duodecimal pitch classes in their prime and ordered forms,
ranges, contour maps, summations, and ordered interval vectors present informa-
tion that analysts can then compare for similarity and developmental structure. Fig-
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 121

REGISTER AND RANGE IN SET ANALYSIS 121

ures 3.17 and 3.18 show a typical example of both a work in which register plays an
important role, and a computer analysis that provides useful information as to the
nature of that role. Many of the melodic intervals in Schoenberg’s Suite for Piano
op. 25, Gavotte (1923) in Figure 3.17 approach or exceed octaves, with some (e.g.,
the last measure in the figure) expanding to two octaves or more. Even a cursory
reading of this music gives the impression that the composer has definite ideas for
the deployment of register and range. The contrapuntal nature of the two register-
and hand-separated melodic lines, combined with held notes, helps group the mu-
sic here into naturally forming trichordal and tetrachordal collections for analysis.
The registral analysis of Schoenberg’s Gavotte, shown in Figure 3.18, reveals very
interesting similarities and relationships between particular groupings of this mu-
sic. I have placed markings (*, †, ‡, and °) after associated linear interval vectors to
point out equivalencies. Additionally, duplicate ranges and interval summations ap-
pear in boldface. Interestingly, the first four entries in Figure 3.18 match the second
four entries in terms of their prime forms (not including the registers) and their lin-
ear interval vectors—both are paired ABCD (*, †, ‡, and °) statements. In other
words, the first four prime-form sets match the second four prime-form sets, sub-
stantiating and reinforcing their relationships. The ranges have many duplicates

nœ
1. b œ.
œ . œ b ˙> œ b œ . œ b œ.
3.
5.
&C Ó n˙
≈ n œ. œ. œ. œ
bœ
Piano p 2.

?C nœ nœ nœ œ nœ 4.
j
& bœ
∑ ‰ nœ ?‰
J J nœ Œ ‰ j
b œ 6.
S p
9.

n bn œœœ
¬

4 >
&b ˙ bœ
7.

b œ œ b œ. œ. œ.n œ ≈ Ó . bœ ∑
b œ. . . fl
p
¬

nœ. nœ
œ> œ . n œ^ n œ
10.
?n œ . œ œ Ó nœ ? j Œ.
J b œ. œ. œ. œ. & # n œœ n œ> b œ
¬

nœ œ œ œ œ œ


8.
11.

FIGURE 3.17 Arnold Schoenberg, Suite for Piano, op. 25 (1923), Gavotte, mm. 1–6,
with groupings circled and numbered.
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 122

122 HIDDEN STRUCTURE

1. (((70 72 83 76) r15) ((84 75 77 71) (-11 2 -6) -15) 110001)*

2. (((60 41 62 53) r23) ((4e 60 59 6t) (13 -3 13) 23) 201000)†
3. (((60 71 74 86) r30) ((76 73 88 62) (-3 17 -30) -16) 001011)‡
4. (((60 41 53) r13) ((59 60 4e) (3 -13) -10) 101000)º
5. (((60 62 53 76) r15) ((5t 69 67 71) (11 0 0 -2 6) 15) 110001)*
6. (((50 61 42 53) r25) ((43 52 55 64) (11 0 3 11) 25) 201000)†
7. (((60 51 54 66) r10) ((58 5e 66 60) (3 0 7 0 0 -6) 4) 001011)‡
8. (((50 41 53) r14) ((55 52 43) (-3 0 -11) -14) 101000)º
9. (((70 72 77) r10) ((70 75 7t) (5 5) 10) 000020)
10. (((60 52 65) r14) ((59 66 6e) (9 5) 14) 001010)
11. (((60 61 72) r11) ((62 63 71) (1 10) 11) 110000)

FIGURE 3.18 Computer analysis of Schoenberg‘s Gavotte by tetrachords and tri-

chords as shown circled and numbered in figure 3.17

and similarities as well (shown in boldface), which in some cases (see lines 1 and 5,
for example) coincide with similarities in the music and at other times relate very
different sets to one another (see lines 7 and 9, for example). Interestingly, many
of the different-ranged groupings here closely approximate one another (see, e.g.,
ranges in lines 2 and 6, lines 4 and 8, lines 4 and 5, etc.).
This often-analyzed work (see, e.g., Straus 2005, 205–12) represents one of
Schoenberg’s first attempts at serialism. The row used here has only four distinct
forms—P4, Pt, I4, and It—integrating the pitch content of this music, which other-
wise relies on dynamics, articulation, and, of course, register and range for variety.
However, as the analysis in Figure 3.18 points out, even register and range conform
to somewhat regular patterns of equivalence and similarity that provide further
cohesion.
Set grouping usually follows harmonies, rhythms, articulations, or similar guides
in music. However, register and range can also provide excellent boundaries for
selecting or supporting pitch grouping for pitch-class set analysis. Webern’s Varia-
tions, op. 27 (1936), for piano, the first seven measures of which appear in Figure 3.19,
offers a good example of register and range support for alternate grouping possibili-
ties. The op. 27 Variations is a highly serialized work in which unison A-naturals (here
shown in m. 2, but also appearing elsewhere in this work and notable for their
striking contrast to the otherwise registrally diverse sets) act as a kind of portal
through which contrapuntally sounding versions of a twelve-tone row switch hands
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 123

REGISTER AND RANGE IN SET ANALYSIS 123

Sehr Schnell
j 5.
n œ n œ. rh 2.
rh 3.
q = 160
?2 #œ
rh 1.
n œ- J ‰ Œ
8.
j ? 9.
4 & n œ. ‰ ‰ n œ
# # œœ ‰ nœ rh 4.

f 1.
p 2.
f 3. p f > p nœ
# œ-
Piano
bœ n œ>
7. lh 4.

& 42 ‰ n œ. ‰ ? ‰ n œj n œ. Œ & n # œœ ‰ Œ
lh 1.
J J
4. lh 2. lh 3.
6.

6
n œ. 13.
? ‰ #œ ‰ & Œ #œ
j

f n œ. 11. ƒj
10. b œ bœ
J j
& ?
n œ ‰ n œ.
p 12.

FIGURE 3.19 Webern, Variations (1936), mm. 1–7.

and registers. The dyad analysis in Figure 3.20a of the first few measures of this
work indicates a natural grouping based on the beamed duplets. I have again
placed correlated signs (*, †, ‡, ^, and °) after certain lines to show linear interval
vector equivalencies. All but three of the ranges also have correlates, as shown by
the boldfaced “r” numbers here. The “nil” initiating each list in this analysis indi-
cate the dyad is an interval rather than a set.
From the dyad analysis in Figure 3.20, Webern’s Variations clearly demonstrates
consistency in terms of register and range. Also, only the unison in m. 2 resists vec-
toral similarity to other sets. However, grouping trichords by voice in this work—as
shown in the analysis in Figure 3.20b—provides further insight into Webern’s intri-
cate formalisms. Even the unison A-natural dyad contributes to the coherent struc-
ture of the Variations in this trichordal analysis. The parallel-numbered right-hand/
left-hand tetrachords here often occur in different registers but share equivalent
linear interval vectors. The symbols following each line of this analysis (#, @, +, and
~) again indicate the connections between sets with identical vectors. Boldface
ranges relate equivalencies. Clearly, this trichordal analysis reveals as much of the
integration in this work as does the dyad analysis.
Comparing non-equivalent pitch-class sets often proves as useful as comparing
equivalent sets. For example, the ranges and registers of differing pitch-class sets
can reveal important information. The first nine measures from Boulez’s Structures
(1952), shown in Figure 3.21, serve as a good example. The tetrachordal analysis in
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 124

124 HIDDEN STRUCTURE

(a)
1. ((nil r26) ((7t 58) (-26) -26) 010000)*
2. ((nil r0) ((69) (0) 0) 000000)
3. ((nil r16) ((61 75) (16) 16) 000100)‡
4. ((nil r3) ((4e 52) (3) 3) 001000)^
5. ((nil r3) ((87 84) (-3) -3) 001000)^
6. ((nil r6) ((66 70) (6) 6) 000001)º
7. ((nil r16) ((75 61) (-16) -16) 000100)‡
8. ((nil r6) ((66 70) (6) 6) 000001)º
9. ((nil r38) ((84 52) (-38) -38) 010000)*
10. ((nil r26) ((58 7t) (26) 26) 010000)*
11. ((nil r44) ((87 4e) (-44) -44) 000100)‡
12. ((nil r13) ((63 52) (-13) -13) 100000)†
13. ((nil r13) ((73 84) (13) 13) 100000)†

(b)
right hand
1. (((60 71 64) r21) ((7t 69 61) (-13 -8) -21) 100100)#
2. ((nil r3) ((87 84) (-3) -3) 001000)@
3. (((70 61 66) r11) ((61 66 70) (5 6) 11) 000011)+
4. (((50 81 56) r41) ((52 58 87) (6 35) 41) 100001)~

left hand
1. (((50 61 65) r13) ((58 69 61) (13 -8) 5) 100100)#
2. ((nil r3) ((4e 52) (3) 3) 001000)@
3. (((60 71 76) r11) ((66 70 75) (6 5) 11) 000011)+
4. (((70 41 86) r41) ((84 7t 4e) (-6 -35) -41) 100001)~

FIGURE 3.20 Webern analysis by dyads (a) and trichords by hand (b) as shown
circled—dyads by solid line and trichords by dotted line—in Figure
3.19.
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 125

REGISTER AND RANGE IN SET ANALYSIS 125

√
bœ œ
1. 2.

2
&8
RÔ ≈ . ‰ ? ® r r ® ‰ j r ® ‰ 38
bœ nœ .
nœ . œ.
Piano 1 P legato
? 28 ∑ ® nœ. œ œ ≈. ≈. Kr j ≈. Kr j 38
R J RÔ
nœ œ bœ œ
Modéré, presque vif ◊ √ ˘
bœ
n ˘œ
? 28 r ‰ & R ∑ ? ® KKr ‰ . ≈ r‰ ‰. R 38
#œ &
fl nœ nœ
Piano 2 f 4. 5. fl fl 6.
Ø
? 28 r ≈ r ≈ ® n œr Kr ‰ ≈. rK 38
#œ ‰ ‰ r≈ ∑
#œ nœ fl nœ fl nœ
fl fl fl fl
#œ œ.
6 nœ 2 œ œ 3.
? 38 ‰ ® Kr Kr ® ≈ R 8 J RÔ ≈ . ‰ 5 ≈ . & RÔ J
16
nœ œ
? 38 j r rK ‰ 28 ∑ ® n œR . œ
J 5 œR .
16 ® ‰.
œ œ. #œ
√˘
# ˘œ n ˘œ nœ
˘ ˘
nœ
& 38 Π?
r ® Kr 28 ‰ n œR ≈ & ≈ . RÔ R ≈ 16 5 ‰ R ≈ R

#œ bœ √ ˘
7. fl fl 8. bœ 9.

? 38 28 ˘
n œ r ® ‰ & 16 5 ≈ . RÔ ‰ .
r ® rK ‰ ® rK ≈ ∑
RÔ
nœ nœ
#œ fl bœ
fl fl
fl

FIGURE 3.21 Nine measures from Boulez, Structures (1952).

04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 126

126 HIDDEN STRUCTURE

Figure 3.22, in contrast to the Schoenberg and Webern analyses just presented,
yields very few matches between any aspects of pitch-class sets, registers, ranges,
interval mappings, summations, or linear interval vectors. The boldfaced selections
here point out the two vector and range equivalencies and, conversely, the paucity
of otherwise obvious relationships.
What appears as inconsistencies, or at least difficult-to-relate sets, however, can
reveal interesting patterns. For example, the ranges of 85, 17, and 65 in the right
hand and of 62, 13, 81, 13, 44, and 18 in the left hand demonstrate alternations of
large and small intervals. Although these intervals—with the exception of the
interval 13, which occurs twice in the left hand—vary in exact size, the dramatic
contrasts of these alternating numbers present similar patterns. The summations in
the right hand (-71, -3, and 65) move upward, while the summations in the left hand
(-14, 3, -10, -11, 25, and 18) simultaneously move upward, downward, and then up-
ward again. This directional pivoting provides a kind of precarious balance for
this passage, which the numbers indicate by their differences rather than their sim-
ilarities. The linear interval vectors—aside from the two that equate—have several
similarities, indicated here in italics and underlining. Clearly, however, this analysis
demonstrates more differences than similarities.
The registral and other types of analyses of each passage in Figures 3.17, 3.19,
and 3.21 provide important information not otherwise available in strict pitch-class

right hand
1. (((30 91 22 57) r85) ((9t 29 54 3e) (-85 31 -17) -71) 100020)
2. (((40 31 43 46) r17) ((48 33 42 45) (-17 11 3) -3) 101010)
3. (((30 61 56 87) r65) ((31 57 60 86) (30 5 30) 65) 000012)

left hand
4. (((80 32 46 38) r62) ((48 82 30 36) (42 -62 6) -14) 010002)
5. (((40 31 43) r13) ((45 37 48) (-10 13) 3) 110000)
6. (((30 91 32 34) r81) ((3e 9t 39 31) (71 -73 -8) -10) 200100)
7. (((30 41 32) r13) ((42 31 33) (-13 2 0) -11) 110000)
8. (((50 71 35) r44) ((54 39 75) (0 -19 44) 25) 000110)
9. (((100 81 92 86) r18) ((86 8e 9t t0 (5 11 2) 18) 110010)

FIGURE 3.22 Boulez analysis by tetrachords by hand as shown circled in Figure 3.21.
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 127

REGISTER AND RANGE IN SET ANALYSIS 127

set analyses. The often widely spaced intervals, though noticeable to the eye and
ear, do not lend themselves to quick mental analysis. Thus, providing the computer-
generated duodecimal registers, ranges, interval maps, summations, and linear in-
terval vectors enhances analysts’ abilities to access these important contributors
to this music. Revealing this information during pitch-class set analysis helps with
grouping, relating similar and dissimilar sets, and finding logic when comparing
sets that otherwise seem disparate and possibly even illogical. Analyzing longer mu-
sical passages produces further benefits, as the following analyses will now attest.

MUSICAL EXAMPLES
Anton Webern’s Concerto for Nine Instruments (op. 24) provides an excellent exam-
ple of how register can play an important role in the analysis of a work. As many
analysts have demonstrated (Bailey 1991; Straus 2005; Williams 1997), Webern’s
row for this highly formalized music consists of four trichords, each of which re-
duces to a [0,1,4] unregistrated prime form, as clearly shown in mm. 1–3 of the first
movement in Figure 3.23, a two-keyboard arrangement of this work. The equivalent
prime-form pitch-class sets in these measures occur in various registers and sound
different in many respects. However, the ranges of these trichords are precisely
the same in each case—13. This equivalence remains true in the second statement
of the row in mm. 4 and 5. In the third presentation of the row in m. 6 and the first
half-beat of m. 7, however, the ranges for the trichords shift to 15, 13, 13, and 15,
adding a subtle nuance to the otherwise static pitch-class sets. This range variation
extends further in the fourth row statement—from beat 2 of m. 7 through m. 8—
with each pitch-class set covering the interval of 11.
The full iteration of the row in Figure 3.23 appearing in mm. 9 and 10 (less the
F-natural appearing in piano one on the last eighth note of m. 10) presents a strik-
ing contrast to the entirely single-sounding pitches thus far encountered in this
work. However, in each case, these dyads and trichords all reiterate the opening
range interval of 13. This consistency gives this passage a continuity that it might
otherwise lack owing to its changes in texture. The line numbering in Figure 3.24
follows the numbering that identifies each of the groupings in Figure 3.23.
The ordered register sets of Webern’s row demonstrate interesting permutations.
Mm. 1–3, for example, use the ordered registers of (7, 6, 7), (8, 8, 7), (6, 6, 7), and (7,
8, 7) for its trichords. Mm. 4 and 5, however, have (7, 6, 7), (7, 8, 8), (7, 6, 6), and
(7, 8, 7) as ordered registers; each of the trichord register collections are retro-
grades of the associated originals. The third statement of the row in m. 6 through
the first beat of m. 7 has registrations of (8, 7, 6), (6, 7, 7), (8, 7, 7), and (6, 7, 7) for
its trichords, each a different arrangement of one or more of the original sets. This
kind of register-motivic play continues throughout the work.
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 128

128 HIDDEN STRUCTURE

2.
b ˘œ n ˘œ 4.
# œ-
# ˘œ ‰ n œ- 3
2
&4 Œ n œ- Œ Œ
Piano 1 f 3
3
p
1. nœ 3.
nœ Œ
& 42 ‰ bœ nœ ≈ ‰ #œ nœ ∑
f
& 42 ∑ ∑ ∑

Piano 2

& 42 ∑ ∑ ∑

& ∑ ∑

& ∑ ∑
6.
nœ bœ 8.

#œ ‰
nœ #œ
& Œ ‰ ‰ nœ ≈
5.
3
f p 3
7.

˘ ˘
3
n ˘œ n œ.
& n œJ ‰ bœ
J
‰ J ‰
n œ. # œ. ‰
f
FIGURE 3.23 Webern, Concerto for 9 Instruments, op. 24, 1st movement, mm. 1–7.
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 129

REGISTER AND RANGE IN SET ANALYSIS 129

#œ nœ n œ. nœ
b œ. n œ. nœ
6 15.
#œ ‰
11.

&≈ bœ ≈ ∑
9.
f ƒ
# œ. n œ. ‰
3
# œ- b œ-
10. 16.
&‰ ≈ n œ. ∑ Œ
n œ-
f nœ nœ
f
12. nœ #œ
&Œ ‰ ≈ bœ nœ ‰ ‰ ≈‰ Œ
f ƒ 14.

nœ bœ
& ∑ Œ nœ ≈ ∑
13.

9
∑ Œ ‰
#œ nœ Œ
& nœ
f # œ-
& ∑ ∑ ‰ J nœ
n œ-
f
19. b œ
17. n # œœ #œ
& ‰ ‰ nœ ‰ ‰ ‰ Œ ∑
3 3 3
Í π 20.

nœ ‰ n œ- 3

&Œ
18.
# œ Œ n b œœ ‰ Œ ? ‰ j
n œ b œ-
3 Í
FIGURE 3.23 continued
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 130

130 HIDDEN STRUCTURE

1. (((60 71 74) r13) ((7e 6t 72) (-13 4) -9) 100100)

2. (((80 71 84) r13) ((83 87 76) (4 -13) -9) 100100)
3. (((60 71 64) r13) ((68 64 75) (-4 13) 9) 100100)
4. (((80 71 74) r13) ((70 81 79) (13 -4) 9) 100100)
5. (((60 71 74) r13) ((72 6t 7e) (-4 13) 9) 100100)
6. (((80 71 84) r13) ((76 87 83) (13 -4) 9) 100100)
7. (((60 71 64) r13) ((75 64 68) (-13 4) -9) 100100)
8. (((80 71 74) r13) ((79 81 70) (4 -13) -9) 100100)
9. (((70 61 84) r15) ((81 79 6t) (-4 -11) -15) 100100)
10. (((70 61 74) r13) ((65 76 72) (13 -4) 9) 100100)
11. (((70 81 74) r13) ((84 73 77) (-13 4) -9) 100100)
12. (((70 71 64) r15) ((68 70 7e) (4 11) 15) 100100)
13. (((70 71 74) r11) ((70 7e 73) (11 -8) 3) 100100)
14. (((70 81 84) r11) ((84 78 87) (-8 11) 3) 100100)
15. (((80 71 74) r11) ((79 85 76) (8 -11) -3) 100100)
16. (((60 71 64) r11) ((71 62 6t) (-11 8) -3) 100100)
17. (((80 71 74) r13) ((70 79 81) (9 4) 13) 001100)
18. (((60 71 64) r13) ((64 68 75) (4 9) 13) 001100)
19. (((80 71 84) r13) ((76 83 87) (9 4) 13) 001100)
20. (((60 71 74) r13) ((6t 72 7e) (4 9) 13) 001100)

FIGURE 3.24 (a) Calculations of registral information in Webern, Concerto for 9

Instruments, op. 24, 1st movement, mm. 1–3.

Comparing interval maps and summations of ordered pitch-class sets also re-
veals a good deal about the formalistic nature of Webern’s concerto. For instance,
the first eight of the twenty trichordal analyses given in Figure 3.24—calculations of
registral and other information from mm. 1–10 in Figure 3.23—show interval-
mapping combinations of the positive and negative intervals of 13 and 4 only, and
interval summations of the positive and negative interval of 9. The final four tri-
chords in Figure 3.23 and resultant ordered interval-map analyses in Figure 3.24
contain only the intervals of positive 4 and 9, along with their summations of 13.
The twenty trichords here contain but five different absolute intervals (4, 8, 9, 11,
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 131

REGISTER AND RANGE IN SET ANALYSIS 131

and 13) in their maps, with all but the interval 9 appearing in both positive and neg-
ative directions. The interval summations contain only four absolute intervals (3, 9,
13, and 15), with 9, 3, and 15 appearing in both positive and negative direction, and
with 15 and -15 occurring but once each.
In the final measure of the first movement of this concerto (shown in Figure 3.25),
the ranges and summations unify to 11 owing to the chordal nature of the tri-
chords, which helps to produce a strong cadential resolution as well as a contrast
to the somewhat more varied approach to range in the opening twenty trichords.
Interestingly, the final chord of this movement of the concerto—the first time more
than three pitches have sounded simultaneously—creates polytonal B major and
E-flat minor triads, reminding us that throughout his work, Webern has alluded to
similar triadic relationships, which seem slightly beyond reach due to intervening
notes. Figure 3.26 presents a computer analysis of the excerpt in Figure 3.25, with
the music grouped by trichord. Note the consistent ranges, the varied-in-direction
but nonetheless equally consistent summations, and the different vectors used in the
final four trichords. As with the opening passage of this work, shown in Figure 3.23,
Webern demonstrates his tight control over all of the musical elements described
in this chapter.
Range and register in Webern’s concerto deserve as much attention as do its
pitch, rhythm, articulation, timbre, and other components. By condensing, expand-
ing, and varying the ranges and registers of his trichords, Webern produces a sub-
tle tension that adds significantly to the effectiveness of this music. Seeing register,
range, and other elements of this music represented in analysis notation reveals
some of the important roles these parameters play in the deployment of pitch
classes in this work.
The second movement of Boulez’s Piano Sonata no. 2 (1950) presents another
example of interesting uses of register and range in post-tonal music. The first six-
plus measures of this movement appear in Figure 3.27. The initial four tetrachords
of this music, distinctly separated into four measures, consist of two different alter-
nating prime-form decimal sets ([0,1,2,3] and [0,1,2,7]). Boulez enhances the pairing
of these sets by using similar ranges (in the case of decimal set [0,1,2,3], R38 and
R37, respectively) and identical ranges (in the case of decimal set [0,1,2,7], both
with R13 ), as shown in the analysis in Figure 3.28.
The ensuing four tetrachords in this movement pair as two differently duodeci-
mal versions of the decimal set [0,1,2,3] followed by the decimal sets [0,2,6,8] and
[0,2,5,8]. These latter four tetrachord groupings follow articulation and beaming
rather than temporal order. The ranges of R33 and R35, followed by the ranges of R44
and R27, enhance these pairings. Although in Boulez’s sonata range and register do
not have the exactness as in Webern’s concerto, they nonetheless contribute to the
sonata’s sense of logic in the tetrachordal groupings and their various interrelation-
ships.
Comparisons of interval maps and interval summations of linear pitch-class sets
also need not be as exact as in Webern’s concerto to inform us of important as-
pects of musical structure. Boulez, for example, does not seem to formalize these
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 132

132 HIDDEN STRUCTURE

n ˘œ 3.
n ˘œ nœ nœ
2 ∑
# ˘œ ‰ ∑
#œ Œ
&4
ƒ ƒ 3 7.

b ˘œ ˘ n ˘œ
Piano 1
1. nœ bœ
5.
2
&4 ‰ nœ ‰ ‰ ∑ ‰ nœ ≈ ‰ ∑
ƒ 3 ƒ
nœ 4.
œ-
& 42 ∑ Œ #œ bœ ≈ ∑ Œ b œ-
Piano 2 2. 3 ƒ ƒ3 8.
bœ bœ
& 42 Œ ‰
nœ nœ ‰ ‰ Œ Œ nœ nœ ∑
ƒ 3
ƒfl 6. fl fl
5 nœ nœ 16.
b ˘œ
3

∑ ‰ ≈
#œ ‰ Œ ‰ n œœ ‰
n
& J
ƒ 12.
ç
nœ bœ
‰ b œj ‰
3
& Œ ‰ ≈ nœ ‰ Œ Œ
bœ
15. n œ
ƒ 10.
fl
nœ b nn œœœ ç
13.
- 3
& #œ Œ Œ ≈ bœ #œ Œ Œ
ƒ
çfl
9. 11.

bœ n n œœ
& Œ ≈ nœ
nœ ∑
#œ
ƒ fl 14.

FIGURE 3.25 Webern, Concerto for 9 Instruments, op. 24, 1st movement, last 7
measures.

aspects of his sonata in the ways in which Webern does, as seen in Figures 3.23 and
3.25 and their associated analyses in Figures 3.24 and 3.26. However, the interval
summations in Boulez’s sonata nonetheless provide interesting information about
the nature of the various sets. In the opening four tetrachords, for example, Boulez
uses 23, -5, -27, and 7, alternating large and small interval summations, and in his
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 133

REGISTER AND RANGE IN SET ANALYSIS 133

1. (((70 71 74) r11) ((78 70 7e) (-8 11) 3) 100100)

2. (((70 61 64) r11) ((64 73 67) (11 -8) 3) 100100)
3. (((70 81 84) r11) ((85 76 82) (-11 8) -3) 100100
4. (((70 61 74) r11) ((71 79 6t) (8 -11) -3) 100100)
5. (((70 71 74) r11) ((7e 70 78) (-11 8) -3) 100100)
6. (((70 61 64) r11) ((67 73 64) (8 -11) -3) 100100)
7. (((70 81 84) r11) ((82 76 85) (-8 11) 3) 100100)
8. (((70 61 74) r11) ((6t 79 71) (11 -8) 3) 100100)
9. (((70 61 64) r11) ((67 73 64) (8 -11) -3) 100100)
10. (((70 71 74) r11) ((7e 70 78) (-11 8) -3) 100100)
11. (((70 61 74) r11) ((6t 79 71) (11 -8) 3) 100100)
12. (((70 81 84) r11) ((82 76 85) (-8 11) 3) 100100)
13. (((80 71 74) r11) ((75 78 84) (3 8) 11) 001100)
14. (((60 71 64) r11) ((61 69 70) (8 3) 11) 001100)
15. (((50 61 64) r11) ((57 63 66) (8 3) 11) 001100)
16. (((70 61 74) r11) ((6e 72 7t) (3 8) 11) 001100)

FIGURE 3.26 Calculations of registral information from Figure 3.25.

second grouping of four tetrachords (following beams and legato marks) we find
similar summations of -22, 10, -14, and 27, two relatively large and two relatively
small variations of his original summations. With the exception of the final map
here, each of his ordered interval maps contains either two positive and one nega-
tive interval, or two negative and one positive interval, also providing consistency
for this passage.
I use register and range as an integral part of my composing processes. In the
opening of my Triplum for flute and piano (1975), for example, pointillistic sets of
various sizes act as an introduction to the piece. As shown in Figure 3.29, this intro-
duction begins with an [80,71,62] prime form trichord that spans an interval range
of 14 (see m. 1 of Figure 3.29), though the decimal prime form of the set ([0,1,2])
covers only a major second. This opening group precedes a similar chromatic
duodecimal tetrachord ([60,71,62,63]) encompassing the same range of 14 (see m. 2
through the first two beats of m. 4). Like the first grouping, the prime form of the
set again consists of consecutive minor seconds. This tetrachord then expands to a
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 134

134 HIDDEN STRUCTURE

2.
Lent e = 80 √ >
j bœ œ
1.
b œ¯
3. 4.
nœ
& Œ J ‰ n œ- . ‰ ‰ ‰ ‰ ‰ n œ¯ Œ Œ ?≈ r
J n n œœ Œ &
P ß n n œpœ-
3
P - n œ- . ß fl
Piano

? ‰ j ‰ ‰ n ˙- J Œ ?
#˙ p 3
Œ & ‰ ‰ j‰ ‰ j ‰
# œ< b œ< nœ
bœ œ
absolument sans pedale (observer rigoureusement les silences de chaque contrepoint)

#œ. n œ¯ 7.
nœ. œ.
5
‰ Œ J ‰ # œ- ?‰
3
j nœ Ó.
& nœ. nœ ‰ & ‰ J
#œ. <
P nœ. p
? Œ.
5.
nœ œ. n œ¯ #œ. ‰
≈ ‰ ‰ ‰ Ó
#œ. n œ-
bœ. nœ.
6. sans pedale 8.
3

FIGURE 3.27 Boulez, Second Sonata for Piano (1948), 2nd movement, beginning.

1. (((70 51 42 63) r38) ((46 78 57 65) (38 -25 10) 23) 120000)
2. (((80 91 82 87) r13) ((89 93 82 84) (6 -13 2) -5) 110001)
3. (((60 51 62 33) r37) ((61 50 6e 3t) (-13 23 -37) -27) 300000)
4. (((40 31 42 47) r13) ((42 38 47 49) (-6 11 2) 7) 110001)

(as notes are beamed and by legato marks)

5. (((50 61 72 83) r33) ((74 83 65 56) (11 -22 -11) -22) 210000)
6. (((50 31 62 43) r35) ((5e 3t 48 69) (-25 10 25) 10) 210000)
7. (((80 72 76 48) r44) ((87 71 4e 75) (-18 -26 30) -14) 010002)
8. (((40 62 35 58) r27) ((39 42 56 60) (5 16 6) 27) 000111)

FIGURE 3.28 Calculations of registral information in mm. 1–6 in Boulez, Second

Sonata for Piano (1948), 2nd movement.
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 135

REGISTER AND RANGE IN SET ANALYSIS 135

pentachord grouping, which extends the range to 16, as shown in the second half of
mm. 4–6. Measure 7 through the first half of m. 8 produces the [60,71,64] ([0,1,4]
decimal) trichord, providing a significant break from the more compressed prime-
form sets heard thus far in this work. However, the interval range 11 of this group-
ing continues the somewhat large registral separations of these pitch classes. This
short grouping is then followed by a [50,61,72,64,75] set beginning in the second
half of m. 8 and continuing through m. 9, and spanning the range of 22, the widest
interval yet covered. Measure 10 and the first half of m. 11 then return to a range
similar to that used in the work’s beginning (13) and the same prime-form trichord
([0,1,2] decimal), but with the ordered registers shifted slightly—from [7,8,6] to
[6,5,4].
In the second half of m. 11 and extending through m. 12, the initial prime-form
trichord of this work implodes to the packed range of a major second, providing a
dynamic contrast to the wide ranges thus far presented. Although other aspects of
the music—rhythm, dynamics, and so on—have changed here, this sudden range
compaction provides the primary impetus for a turn in the music’s direction. This
compact version of the [0,1,2] decimal prime-form set continues through another it-
eration in m. 13 and the first half of m. 14. A return to the wider range of m. 13 then
follows in the remainder of mm. 14 and 15.
Register and range provide the principal driving force for the music in this open-
ing section. The simplistic descending chromatic scale—the primary pitch source
for this section of Triplum—produces a rather static melodic and harmonic base
that otherwise creates little tension. Without the accordioning range here, this
opening passage would doubtless have little forward momentum and leave Triplum
with little motivation for its ensuing sections.
The computer analysis for this passage of Triplum, shown in Figure 3.30, reveals
interesting interval maps and summations. Even though the sets here have varying
numbers of pitch classes (from three to five), the interval maps contain many simi-
lar intervals and repeated motions indicative of the motivic imitations in the music.
Only the trichords lack this feature. Although the summations repeat only one in-
terval (-1, stated three times), all but the final summation show a series of down-
ward or static motions (-1, -3, -14; 0, -8, -13; and -1, -1, 11), clearly the result of the
descending chromatic-scale foundation of this music, though not its necessary
consequence.
The seven measures of Triplum shown in Figure 3.31 provide another example in
which range and register contribute to this music’s momentum. The piano part of
this passage consists of several groups, all but one of which cycles through six as-
cending major sevenths (the last group has seven). Because the number of pitches
in these collections does not divide the chromatic scale evenly, and because I con-
nect the end of each set with the beginning of the following set via the next pitch
class of the ongoing descending chromatic scale, each grouping of ordered pitch
classes is different. The rigid range of each contributing interval and this cycling
through the downward chromatic scale while moving upward in register creates a
static canvas on which the flute—playing a modal Navajo melody—tonicizes C.
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 136

136 HIDDEN STRUCTURE

q = 42 # œ .. 1.‡ œ œ. [no accent]

2.
#˙ 3
3.

& 45 n ˙ .
3
œ Ó ‰ bœ ˙ bœ ˙. œ ‰Œ nœ
Flute
R
∏ ç p ƒ P f π ƒ
w œ w
œ œ‡
(l.h.)

& 45 n w
(r.h.) (r.h.)
∑
ç ‰. R ç
Piano
[use 2 fingers]
ç
≈ œr ‰ ‰ œr ≈ ‰ ∑
3
Ó.
(l.h.)
? 45 ∑ ∑ &
. . nœ
F P ç
5 3
‰ Œ ‰
3
b˙ œ ‰. Ó ˙
3
œœ œ
Fl. &˙ œ b œ œ. n œ>b œ œ n œ. b œ- J
p ƒ subito π F ƒ F
(r.h.)
& ∑ ∑
bw œ
Pn. ç (l.h. take over
from r.h.)

&w œ w Œ ∑

8
4.
j
3>
# œ œ œ> œ .. ‡ 5. 6.
‡
7.

& œ . œ. ® Œ œ œ bœ ˙.
bend
Œ‰ j ≈Œ
R nœ ˙. œ .. n ˙
Fl.
f >
(r.h.)
ƒ F ∏ ƒ
3
‡
Ó.
(r.h.)

& œ . œb œ Œ ˙ . #˙.
? Ó ‰
w œ &
#œ œ
Ï
.ç F Ï
œ>.
Pn.
n œ. œ̆ œ.
(l.h.)

& ∑ ? Ó ≈ ® RÔ ‰ Œ ‰ . Ó ‰ . œj ˙ œj ‰ Œ Œ Œ ≈J
ƒ F
> Ï
FIGURE 3.29 The author, Triplum for flute and piano (1975), beginning.
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 137

REGISTER AND RANGE IN SET ANALYSIS 137

˘ ˘
˙.
12 8. 9.
bend
# œ # œ K
r # bend

œ ˙.
bend
&˙ ‰ Œ nw Œ ‰ nœ ≈ ≈ nœ ‰ ® nœ ≈
œ.
Fl.
F . P p
π f ƒ F
r >
& œ .. ≈ Ó ≈b œ. b œ. ≈≈ b œ. ‰ b œ. ‰ .b ˙ œ Œ Ó ≈ œj . w
R >
Pn.
ƒ f F ç
?˙ œ 3
JŒÓ Ó.
3
j
Œ b œ œ œ .. ≈ Ó . ∑
&
>
F
FIGURE 3.29 continued

1. (((80 71 62) r14) ((70 81 6e) (13 -14) -1) 110000)

2. (((60 71 62 63) r14) ((6t 79 68 67) (11 -13 -1 0) -3) 300000)
3. (((70 61 62 63 64) r16) ((76 65 63 62 64) (-13 -1 -1 -1 2 -1 -1 2) -14) 620000)
4. (((60 71 64) r11) ((69 70 61) (11 -3 3 -11) 0) 202000)
5. (((50 61 72 64 75) r22) ((67 76 79 6t 5e) (11 3 -11 -11 0 0) -8) 301000)
6. (((50 61 42) r13) ((60 51 4e) (-11 -2) -13) 110000)
7. (((60 61 52) r2) ((60 61 5e) (1 -2) -1) 110000)
8. (((60 61 62) r2) ((69 6t 68) (1 -2 2 -2 2 -2) -1) 150000)
9. (((60 71 62) r13) ((67 65 76) (-2 13 -13 13 -13 13) 11) 510000)

FIGURE 3.30 Calculations of registral information in the author’s Triplum for flute
and piano (1975), mm. 1–7.

In m. 46 of Triplum—an extended thirty-four-second proportionally notated sec-

tion of music, shown in Figure 3.32—I use register to help establish a nine pitch-
class set. In this section, pitch classes retain their initial registers, which are
broadcast over a two-plus octave range. The vertical dotted lines in Figure 3.31 pro-
vide the timing in seconds, with pitches occurring proportionate to their horizontal
visual locations with respect to these dotted lines. Were each pitch class here to
appear in various registers rather than just one, the conglomerate—the normal or-
dered [60,81,72,63,74,66,78,69,7t] duodecimal nonachord set—would be much less
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 138

138 HIDDEN STRUCTURE

5 ∑ ∑
3
Flute &4 œ- œ œ ˙ Ó
p -

3
n œ . √n œ
#œ
& 45 ∑ Ó nœ J ≈ J Œ J ∑
3
Piano

≈ j ‰ n œ ® # œÔR œ .
3
Ó. Ó.
3
? 45 ∑ ‰ #œ
bœ . œ
bœ.
°
F F
101 3

& Ó ‰ ‰ œ r . Ó.
Fl.
œ- œ- - œ- œ- œ œ ‰

√ #œ
bœ nœ n œ .. ≈ J ≈
& nœ nœ ® Œ J Ó. Ó. nœ
3
Pn.
? ∑ Œ #œ
nœ Œ J ® Ó
bœ. 3

103 3
Fl. & Œ œ œ- œ œ œ œ œ œ œ . œ œ
j ‰ ∑
π- - -
√ # œ ..
n œ√ 3

Œ Ó. Ó
j b œ n Jœ ® J
& ‰ nœ
Pn.
bœ
? Ó.
3
∑ j ‰ nœ
bœ nœ
◊

FIGURE 3.31 The author, Triplum for flute and piano (1975), mm. 98–104.
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 139

REGISTER AND RANGE IN SET ANALYSIS 139

perceivable. The music would also lose its wind-chime effect, so important to this
particular moment in the piece.
The register-static trichord that appears twice separately in the flute and piano
([65,67,6e] or F-natural, G-natural, and B-natural) gains prominence from its deploy-
ment of otherwise unused pitch classes and from its complementary relationship
with the backdrop of the register-frozen wind-chime keyboard music, producing a
complete twelve-pitch combination. Register thus plays an extremely important
role in this passage, even though no actual register shifting takes place.
Obviously, grouping this excerpt into several smaller collections of pitch classes
could produce fragments of the nonachord and thus create a false analysis of the sta-
sis in this passage. Such an analysis would also mask the polarity of the nonachord-
trichord relationship. Again, the register-frozen pitch classes enhance both the aural
and visual impression that no actual change—apart from the twice-occurring com-
plementary trichord—takes place. Clearly, this music depends on register as much
as it depends on pitch class and other factors.
As just demonstrated, aspects of register and range can prove useful even in the
analysis of large groupings such as phrases and sections. For example, I have found
that with certain composers and works, range and register maps yield information
about phrase length, location of entrances of new material, or positions of impend-
ing cadences not easily revealed by analysis of pitch class and other elements of
the music alone. Of course, such maps can occasionally be deceiving, as when their
extremes define merely the range limits of instruments rather than important char-
acteristics of the music itself. Although studying rhythm, timbre, and other salient
features can significantly impact our understanding of a work, analyzing for regis-
ter, range, interval maps, summations, and linear interval vectors, as shown by the
analyses presented here, can also provide valuable insights.

PROGRAM DESCRIPTION
The software called Register (Macintosh and PC compatible), available on the CD-
ROM accompanying this book, takes a series of ordered note-events and returns
the prime form in duodecimal numbers, its registration in decimal numbers, the or-
dered pitch-class set in duodecimal numbers, the ordered interval map in decimal
numbers, the interval summation in decimal numbers, and the linear interval vec-
tor. As shown previously in the text and figures, these returned sets take the form
(((80 71 62) r-14) ((70 81 6e) (13 -14) -1) 110000).
I have also placed the music of the figures in this chapter as note-events on
the CD-ROM accompanying this book, both as examples for readers to use and for
verifying the findings presented in the figures here. Shared linear interval vectors
obviously play a significant role in my interpretation of these analyses. I encourage
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 140

140 HIDDEN STRUCTURE

Flute & ∑

#œ
& nœ #œ #œ nœ #œ #œ nœ nœ #œ #œ nœ nœ #œ #œ #œ

Piano
π
& #œ nœ #œ nœ nœ #œ #œ #œ nœ #œ #œ nœ #œ
nœ nœ nœ
una corda

Fl. & n˙. n˙ n˙ Œ

p
#œ #œ #œ nœ #œ #œ #œ
& nœ nœ#œ #œ #œ #œ

Pn.

& #œ #œ nœ #œ nœ nœ #œ # œ n œ # œ n œ> n œ nœ nœ
#œ nœ #œ >n œ # œ
f π
>π
f fπ
Fl. & n˙. n˙
p
#œ #œ #œ #œ #œ nœ #œ #œ
nœ nœ #œ #œ nœ#œ #œ nœ nœ
&
Pn.

& #œ nœ #œ #œ nœ #œ #œ n œ n œ># œ
nœ nœ #œ
nœ
nœ
>π
fπ f
FIGURE 3.32 Another passage from the author’s Triplum for flute and piano (1975),
m. 43.
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 141

REGISTER AND RANGE IN SET ANALYSIS 141

Fl. & n˙

#œ
& nœ #œ #œ #œnœ nœ nœ

Pn.
>

& nœ #œ #œ #œ nœ #œ
nœ nœ
f π
FIGURE 3.32 continued

readers to compare and cross-reference the other parameters to obtain further

insights into the music represented by this data.
Maintaining registral information in duodecimal notation when reducing pitches
to pitch-class t-normal and prime forms, and using this information to calculate
range, contour maps, interval summations, and linear interval vectors, can signifi-
cantly enhance the analysis of many post-tonal works. Although such information
exists in the music itself, it can easily be lost amid the various machinations that
ordinary pitch-class set analysis requires. Combining register with pitch class in a
relatively simple duodecimal notation enables analysts to more fully account for
the roles register and range play in many post-tonal works and hence contributes
to a fuller understanding of that music.
I have also included a program called Set Multiples that produces all the possi-
ble combinations of prime-form sets that multiply to an individual prime form set—
more or less a reverse process of that described in SetMath. The top-level function
find-all-source-multiples-of, provided a single argument, produces a list of
all possible combinations of sets that multiply to that set. Although this function
and process may seem more of use to composers than to analysts, find-all-
source-multiples-of can provide a long series of possible source sets for analy-
sis when simple neighboring sets do not multiply to successive sets. For example, in
Figure 3.32, m. 7, the new set [0,2,6] occurs (E–D–G  ). This set seems somewhat
removed from the sets that appeared earlier in this work. However, running find-
all-source-multiples-of on this new set produces many possible groups of
sets, including [0,2,5] and [0,1,4], the sets appearing in the first few measures of the
piece, another rationale for including these sets at this point in the piece.
Figure 3.33 presets the ninety-seven possible origin sets for the decimal set
[0,2,6]. The parentheses here are organizational and relate to the programming
language rather than to any preferred notation. The lack of commas results from the
same source. I have left these in the output in this figure to make sure that readers
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 142

142 HIDDEN STRUCTURE

who use the software provided with this book are prepared for the difference be-
tween most of this book’s notation and the output they will experience. Also, the
two sets to the left (aggregated by super parentheses) are the sources, and the final
set—the same in each case—represents the result of the multiplication of the first
two sets. Note further that as the list progresses, the multiplication of much larger
sets than the trichord output appear as multiples. The smaller set output results
from the duplications that reduce out from the mathematics involved.
I do not intend to suggest here that such additions, subtractions, and multiplica-
tions occur commonly in post-tonal music, just that simple mathematical processes
such as those described here do occur and that analysts should be aware of this
fact. Sets that seemingly arrive from nowhere in music often relate to earlier- or later-
appearing sets by way of these processes, and including them in analysis can
enhance one’s understanding of how differing sets can relate to one another. Com-
puter programs make such machinations more quickly and accurately acquirable,
as a few paper computations should easily prove.
Mapping one multiplier across a single set can also produce interesting results
(see Morris 1987, 42), particularly for composers. In fact, Morris’s book and that of
John Rahn (1980) contain many extraordinary machinations of pitch-class sets that,
given the proper computer program, could make very interesting tools for post-
tonal analysis. Rather than produce such programs here, I point readers to these
and other fine sources (especially Lewin 1987) for further research.
Stravinsky’s Three Pieces for String Quartet (1914) produces some interesting in-
stances of set computations. For example, the first two tetrachordal sets occurring
at the beginning of his quartet as beats 2 and 3 produce the decimal prime form
[0,1,4,7] when multiplied together, the same set that concludes the passage in m. 7.
This kind of interrelationship between sets helps to unify what could otherwise
seem to be dislocated, even arbitrary, set progressions. The voice-leading, though
certainly the connecting principle for the repeating sets, is therefore not the only glue
that holds them together as a unified whole. Schoenberg’s comment that “chords
are formed merely as accidents of voice leading, and they have no structural signifi-
cance since responsibility for the harmony is borne by the melodic line” (Schoen-
berg 1983, 312) notwithstanding, Stravinsky’s chordal sets represent important
correspondences in this music.
Because I will be using sets more abstractly for most of the remainder of this book,
I will not compel readers to continue reading the duodecimal registral notation as
presented in this chapter. I have omitted the registral information not because of
difficulties that might be encountered with the notation, but from the lack of rele-
vance this added information has to the topics being discussed. I will, however, re-
turn to duodecimal pitch-class sets in Chapter 7. The various processes described
in this and upcoming chapters will coalesce into Muse, a computer program de-
signed to analyze post-tonal music from the many different perspectives provided
here.
04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 143

REGISTER AND RANGE IN SET ANALYSIS 143

((((0 1 2) (0 2 4)) (0 2 6)) (((0 1 2 4 6) (0 2 4 6 8)) (0 2 6))

(((0 1 4) (0 2 5)) (0 2 6)) (((0 1 2 5 6) (0 2 3 6 8)) (0 2 6))
(((0 2 4) (0 1 5)) (0 2 6)) (((0 2 3 4 6) (0 1 2 3 7)) (0 2 6))
(((0 1 5) (0 2 6)) (0 2 6)) (((0 2 3 4 6) (0 1 2 6 7)) (0 2 6))
(((0 2 5) (0 1 6)) (0 2 6)) (((0 2 3 4 6) (0 1 2 6 8)) (0 2 6))
(((0 2 5) (0 2 6)) (0 2 6)) (((0 2 3 4 6) (0 1 4 5 8)) (0 2 6))
(((0 1 2 3) (0 2 3 6)) (0 2 6)) (((0 1 2 6 7) (0 2 4 6 8)) (0 2 6))
(((0 1 3 4 7) (0 2 4 5 8)) (0 2 6))
(((0 1 2 3) (0 2 4 8)) (0 2 6))
(((0 1 3 4 7) (0 2 4 6 8)) (0 2 6))
(((0 1 2 4) (0 2 3 6)) (0 2 6)) (((0 1 3 6 7) (0 2 4 6 8)) (0 2 6))
(((0 1 2 4) (0 2 4 6)) (0 2 6)) (((0 2 3 4 7) (0 1 4 5 8)) (0 2 6))
(((0 1 2 4) (0 2 4 8)) (0 2 6)) (((0 2 3 4 7) (0 1 4 6 8)) (0 2 6))
(((0 1 2 4) (0 2 6 8)) (0 2 6)) (((0 2 3 4 7) (0 3 4 5 8)) (0 2 6))
(((0 1 2 5) (0 2 3 6)) (0 2 6)) (((0 1 2 4 8) (0 2 4 6 9)) (0 2 6))
(((0 1 4 5) (0 2 3 6)) (0 2 6)) (((0 1 2 6 8) (0 2 4 6 9)) (0 2 6))
(((0 2 3 5) (0 1 2 6)) (0 2 6)) (((0 1 3 5 8) (0 2 4 6 9)) (0 2 6))
(((0 2 3 5) (0 1 4 6)) (0 2 6)) (((0 1 3 7 8) (0 2 4 6 9)) (0 2 6))
(((0 2 3 5) (0 2 4 6)) (0 2 6)) (((0 2 3 5 8) (0 2 4 6 8)) (0 2 6))
(((0 1 2 6) (0 2 3 6)) (0 2 6)) (((0 2 3 5 8) (0 1 4 6 9)) (0 2 6))
(((0 1 2 6) (0 2 4 6)) (0 2 6)) (((0 2 3 5 8) (0 2 4 6 9)) (0 2 6))
(((0 1 2 6) (0 2 3 7)) (0 2 6)) (((0 2 3 6 8) (0 2 4 5 8)) (0 2 6))
(((0 1 2 6) (0 2 4 8)) (0 2 6)) (((0 2 4 5 8) (0 2 4 6 8)) (0 2 6))
(((0 1 5 6) (0 2 4 6)) (0 2 6)) (((0 2 4 5 8) (0 1 3 6 9)) (0 2 6))
(((0 2 4 5 8) (0 2 4 6 9)) (0 2 6))
(((0 2 4 6) (0 1 3 7)) (0 2 6)) (((0 2 4 5 8) (0 1 2 4 7)) (0 2 6))
(((0 2 4 6) (0 1 6 7)) (0 2 6)) (((0 2 4 6 8) (0 3 4 5 8)) (0 2 6))
(((0 2 4 6) (0 2 3 7)) (0 2 6)) (((0 3 4 5 8) (0 2 4 6 9)) (0 2 6))
(((0 2 4 6) (0 2 4 7)) (0 2 6)) (((0 1 2 3 4 6) (0 2 3 4 6 8)) (0 2 6))
(((0 2 4 6) (0 3 4 7)) (0 2 6)) (((0 1 2 3 4 6) (0 2 3 4 6 9)) (0 2 6))
(((0 2 4 6) (0 1 5 8)) (0 2 6)) (((0 1 2 3 5 6) (0 2 3 4 6 8)) (0 2 6))
(((0 1 5 7) (0 2 4 8)) (0 2 6)) (((0 1 2 3 5 6) (0 2 3 4 6 9)) (0 2 6))
(((0 1 6 7) (0 2 4 8)) (0 2 6)) (((0 1 2 6 7 8) (0 2 4 6 8 10)) (0 2 6))
(((0 1 6 7) (0 2 6 8)) (0 2 6)) (((0 1 2 6 7 8) (0 2 3 4 6 9)) (0 2 6))
(((0 2 3 7) (0 1 4 8)) (0 2 6)) (((0 2 3 4 5 7) (0 2 4 6 8 10)) (0 2 6))
(((0 2 4 7) (0 1 5 8)) (0 2 6)) (((0 1 2 4 7 8) (0 2 4 6 8 10)) (0 2 6))
(((0 2 4 7) (0 3 5 8)) (0 2 6)) (((0 1 3 4 7 8) (0 2 4 6 8 10)) (0 2 6))
(((0 2 5 7) (0 1 4 8)) (0 2 6)) (((0 1 4 5 8 9) (0 2 3 4 6 8)) (0 2 6))
(((0 3 4 7) (0 2 5 8)) (0 2 6)) (((0 2 3 4 6 8) (0 1 2 3 6 9)) (0 2 6))
(((0 1 2 4 6 8) (0 2 4 6 8 10)) (0 2 6))
(((0 3 4 7) (0 2 6 8)) (0 2 6))
(((0 1 3 4 6 8) (0 2 4 6 8 10)) (0 2 6))
(((0 2 5 8) (0 2 6 8)) (0 2 6)) (((0 1 3 4 6 9) (0 2 4 6 8 10)) (0 2 6))
(((0 2 6 8) (0 3 5 8)) (0 2 6)) (((0 1 3 5 6 9) (0 2 4 6 8 10)) (0 2 6))
(((0 1 2 3 4) (0 2 3 4 6)) (0 2 6)) (((0 2 4 6 8 10) (0 2 3 4 6 9)) (0 2 6))
(((0 1 2 3 5) (0 2 3 4 6)) (0 2 6)) (((0 1 2 3 4 8) (0 2 3 4 6 9)) (0 2 6))
(((0 1 2 3 6) (0 2 3 4 6)) (0 2 6)) (((0 1 2 3 7 8) (0 2 3 4 6 9)) (0 2 6))
(((0 1 2 3 6) (0 2 3 4 7)) (0 2 6)) (((0 2 3 4 5 8) (0 1 2 3 6 9)) (0 2 6))
(((0 1 2 3 6) (0 2 3 6 8)) (0 2 6)) (((0 1 2 3 5 8) (0 2 3 4 6 9)) (0 2 6))
(((0 1 2 4 6) (0 2 3 6 8)) (0 2 6)) (((0 1 2 3 6 8) (0 2 3 4 6 9)) (0 2 6))
(((0 1 2 4 6) (0 2 4 5 8)) (0 2 6)) (((0 2 3 4 5 6 8) (0 1 2 3 6 7 9))(0 2 6))
(((0 1 2 3 4 6 8) (0 2 3 4 6 7 9))(0 2 6))
(((0 2 3 4 6 7 9) (0 1 2 3 5 6 8))(0 2 6)))

FIGURE 3.33 The 97 possible origin sets for [0,2,6].

04_DAS_Chap3_pp99-144 1/29/09 1:34 PM Page 144

144 HIDDEN STRUCTURE

CONCLUSIONS
This chapter has explored a number of important roles that both range and register
play in post-tonal music, and how analysis of these properties can lead to impor-
tant revelations. The simple duodecimal notation proposed will certainly enhance
set-theoretical analysis. At the outset of this chapter I noted that its modest goals
would allow “both those familiar with mathematics but unfamiliar with music, and
those familiar with music but unfamiliar with mathematics, to find common ground
before the book takes more adventuresome directions in Chapter 4 and beyond.”
Chapter 4, then, takes the somewhat unusual view that because musical scales
have served music analysts well for centuries, they should continue to serve as a
tool for analyzing post-tonal music. Computers can in fact provide a perfect vehicle
for extracting such scales from apparently non-scalar chromatic music. Thus, Chap-
ter 4 lays a foundation for Chapter 5’s provocative notion that scales can provide a
foundation for revealing function in post-tonal music.
05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 145

FOUR
Computer Analysis of
Scales in Post-Tonal
Music

Discussing pitch centricity in post-tonal music is more complicated than identify-

ing the tonic of a tonal piece. In post-tonal music, we can talk about an entire
spectrum of centric effects. At one extreme, represented by much twelve-tone mu-
sic, there is little or no sense of centricity. Even so, of course, the pitch classes
are not treated identically, and it is important to be sensitive to any kind of spe-
cial treatment accorded to pitch classes or pitch-class sets. At the other extreme,
many post-tonal pieces are deeply preoccupied by questions of centricity.

Joseph Straus, Introduction to Post-Tonal Theory (2005), 133

For over a millennium, musical scales have provided a fundamental palette upon
which music analysts could build superstructures for determining modes and
keys—primary ingredients in music theory. With the advent of highly chromatic
and serial music in the early twentieth century, scales lost favor as a basis for de-
scribing such notions as key and function. In this chapter, I suggest that scales may
yet have a place in the analysis of post-tonal music, and that computers offer an
extraordinary opportunity for analysts to rediscover the importance of scales in
all music. After all, scales define the most frequently occurring pitches in a work of
music, and—after nearly a century of discussion and dissension—most analysts
now agree that no work portrays its various pitch classes as truly equal. This chap-
ter, then, centers on Principle 2 as described in Chapter 1 of this book.

MATHEMATICAL SEQUENCES
A mathematical sequence, of which a musical scale is an example, is an ordered list
of objects (typically numbers) of finite or infinite length. The order of the elements
of a sequence matters, unlike the order of mathematical sets, discussed in Chapter
3. For example, the same elements of sequences may return, and their position and
order in that sequence is important. Notated sequences that extend indefinitely
usually terminate in ellipses, such as 1, 2, 3, 4, 5 . . . , to indicate their infinite con-
tinuation. Sequences often result from the application of algorithms such as

145
05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 146

146 HIDDEN STRUCTURE

[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12 . . .],

expressible as Un = n + 1, where, to each number n in a sequence U1, U2, . . . Un,
there corresponds a number Un. Known as an ascending incremental sequence of
numbers in decimal notation, a duodecimal version of this sequence produces the
chromatic scale in music. A sequence of odd integers such as
[1, 3, 5, 7, 9, 11, . . .]
could be notated as
Un = 2n - 1
and
[1, 3, 6, 10, 15, 21, . . .]
as
Un = n(n + 1)/2,
and so on.
More interesting and useful sequences include the famed Fibonacci sequence,
created in the thirteenth century by Leonardo of Pisa (also known as Fibonacci).
The Fibonacci sequence consists of numbers created by the addition of two previ-
ous numbers:
[0, 1, 1, 2, 3, 5, 8, 13, 21, 34, 55 . . .]
It is usually seeded with a given 0, 1. As will be further discussed in Chapter 7, this
sequence produces what is often called the golden mean or golden section,
resulting from the division of a given number in the sequence by its predecessor. In
terms of the notation just used, the Fibonacci sequence could be expressed as Un =
Un-1 + Un-2. Many other methods for figuring the golden mean exist mathematically
as well.
There are virtually an infinite number of sequences, all of which—by using mod-
ulo 12—can be reduced to a particular musical scale. For example, the Fibonacci
sequence produces the scale [0,1,2,3,4,5,7,8,9,t,e]—all members of the chromatic
scale except its midpoint, 6—from the following code:
(defun create-sequence-scale (sequence)
“Produces a scale out of the sequence in its arg.”
(my-sort #'< (remove-duplicates
(loop for element in sequence
collect (mod element 12)))))
(defun fibonacci (limit n1 n2)
“Creates a fibonacci sequence up to first arg.”
(if (> n2 limit)()
(cons (+ n1 n2)
(fibonacci limit n2 (+ n1 n2)))))
05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 147

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 147

(defun my-sort (function lists)

“Non-destructive sort.”
(loop for sorted-item in
(sort (loop for item in lists
collect (list item)) function :key #’car)
collect (first sorted-item)))
The use of the Lisp macro loop—in the function create-sequence-scale—
requires some explanation here. The loop macro allows many of its arguments to
appear without parentheses, thus making it both more readable and, unfortunately,
less Lisp-like. In this function, loop states that the variable element will represent
in turn each member of its sequence argument, thus applying mod 12 to each in
turn and returning it in a list supplied by the use of collect.
Applying the function create-sequence-scale to a variety of number se-
quences provides quite interesting results. For example, the following results from
using this function on all of the primes up to 100
[1,2,3,5,7,e]
and the
(defun additive-sequence
(limit &optional (start 0)(increment 1))
(if (> start limit)()
(cons start
(additive-sequence limit (+ start increment)(1+
increment)))))
function produces output in the form of
(0 1 3 6 10 15 21 28 36 45 55 66 78 91 105 120 136 153 171
190 210 231 253 276 300 325 351 378 406 435 465 496 528 561
595 630 666 703 741 780 820 861 903 946 990 1035 1081 1128
1176 1225 1275 1326 1378 1431 1485 1540 1596 1653 1711 1770
1830 1891 1953 2016 2080 2145 2211 2278 2346 2415 2485 2556
2628 2701 2775 2850 2926 3003 3081 3160 3240 3321 3403 3486
3570 3655 3741 3828 3916 4005 4095 4186 4278 4371 4465 4560
4656 4753 4851 4950)
producing the octatonic scale, when reduced with create-sequence-scale,
[0,1,3,4,6,7,9,t]
and so on. The number of sequences and resulting scales produced is dependent
only on the imaginations of those inventing and using this kind of code.
The following polynomial formula creates a scale with quite interesting dimensions:
ƒn = (a * 3n)+ (b * 2n) + (c * n) + d
producing the pitch-class scale of
05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 148

148 HIDDEN STRUCTURE

[0,2,3,5,6,9,e]
when A equals 5, and B, C, and D each equal 2. The Lisp function that creates the
sequence for the above output appears below.
(defun polynomial-sequence
“Creates a sequence within limit based on polynomial
equation.”
(modulo limit &optional (start 0)(n 1)(a 5)(b 2)(c 2)
(d 2))
(if (> start limit)()
(cons (mod start modulo )
(polynomial-sequence modulo
limit
(+ (* a (* n n n))
(* b (* n n))
(* c n) d)
(+ n 1)))))
Obviously many other sequences possess interesting potentials for musical
use. Whether such sequences originate from algorithmic or mathematical sources
hardly matters, as long as the resulting scales prove useful for analyzing music. The
following discussion assumes that scales are examples of both finite sequences
and sets, though in the latter case differences appear as I will soon describe.

SCALES
Musical scales represent an important resource for composers and for the music
analysts who analyze the music of these composers. Scale analysis has not, how-
ever, found as well recognized a place in the canon of analytical processes available
for post-tonal music as it has for tonal music. This section of this chapter presents
a computer approach for quickly and efficiently discovering scales in post-tonal
music. After initially defining an algorithm that produces all possible one-to-twelve-
pitch equal-tempered scales for reference, I describe a system of linear interval vec-
tor analysis that abstracts these scales into collections of similar scales based on
shared traits. I then explain a series of analytical techniques for revealing scales in
post-tonal music. These techniques are intended to enhance, not replace, other ap-
proaches to analysis such as pitch-class set theory (see Chapter 3). Finally, several
examples of scales used in post-tonal composition demonstrate the effectiveness of
using such an approach to better understand at least some post-tonal music.
For many hundreds of years, scales have played an important role in defining
Western classical modal and tonal music (Hindemith 1942; Schoenberg 1983).
05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 149

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 149

Scales provide important information for interpreting harmonic function, chromati-

cism, and modulation (Carey and Clampitt 1989; Gamer 1967). Even in twentieth-
and twenty-first-century musical styles, scales—octatonic, whole tone, chromatic,
and so on—have contributed to the interpretation and appreciation of many quasi-
tonal works (Agmon 1989; Browne 1981). For reasons that are not fully clear, how-
ever, scale analysis has not played a leading role in the analysis of most post-tonal
music, but has taken a back seat to grouping approaches such as pitch-class set
theory, which reveal, I believe, incomplete (though significant) insights into musi-
cal structure (Clough 1980; Morris 1987; Perle 1996; Rahn 1980; Straus 1990). Defin-
ing and mapping scales in post-tonal music can provide important information,
ultimately contributing to a more complete understanding of this music, regardless
of its style or the formal compositional techniques used (Lindley and Turner-Smith
1993).
In order for readers to fully understand my arguments regarding non-tonal scales
and their use, I review here a few of the fundamental ways in which traditional
Western classical composers use tonal scales. I realize that for most readers, my
comments here may seem painfully obvious. However, not making these points will
seriously undermine my ability to persuade these same readers of my contention
that scales participate significantly in post-tonal music.
The passage from the first movement of Beethoven’s Piano Sonata op. 53 shown
in Figure 4.1 begins and ends in C major, even though many pitches foreign to C ma-
jor eventually appear. In fact, this passage could represent the chromatic scale, be-
cause all twelve pitches appear here in one form or another. Even musical amateurs
can generally find this chromaticism by eye. However, actually counting the pitches
produces a more accurate result. Figure 4.2 gives the exact number of each pitch
class in the 298 pitches present in this example.
Not surprisingly, the C major scale pitches far outnumber the non–C major scale
pitches in Figure 4.1—by 237 to 61. Interestingly, however, some individual non-
scale pitches outnumber scale pitches. For example, B-flat outnumbers D-natural,
A-natural, and E-natural, and A-flat outnumbers A-natural in these rankings. If one
were not aware of the important roles keys play in tonal music, the scale here
might be construed as C–D–E–F–G–A-flat–B-flat–C, rather than as C major. Including
durations when figuring these weightings does little to counter this analysis. My
first point, then—that simply counting pitches does not necessarily lead one to an
accurate conclusion regarding the scale in use—suggests that other important fac-
tors must be considered. Clearly, nonharmonic tones, secondary harmonies, modu-
lations to other keys or regions, and meter, among other elements, all play crucial
roles in establishing a particular key or scale in tonal music. The lone C-sharp that
begins in m. 4, for example, has but one iteration, indicating its lesser value. The
two F-sharps at the end of m. 2, occurring one after another as the third of a re-
peated secondary dominant-of-the-dominant harmony, may actually further estab-
lish C major as the key in use, rather than the contrary. The B-flat, A-flat, and E-flat
here occur in a B-flat major region beginning in m. 5 and not ending until the feint
05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 150

150 HIDDEN STRUCTURE

? c ‰ œœ œœ œœ œœ œœ œœ œœ œœ œœ œœ œœ œœ œœ # œœ œœ œœ .
Allegro con brio œœœ
J ‰ Œ &

Piano π
?c
œ œœ œœ œœ œœ œœ œœ œœ œœ œœ œœ œœ œœ œœ œœ œœ œœ œœ œœ œœ œœ œœ œœ œœ

#œ
j œ œ nœ œ œ œ
4

& Π? .
b œœ œœ œœ œœ œœ œœ œœ œœ b œœ œœ œœ œœ œœ œœ œœ œœ œœ .
œœœ ‰ Œ
J &
π
?
œœœ œ œœœœ œœœœœœœœ œœœœœœœœ œœœœ œœœœ
œœœ œ œ œ œ œ bœ œ œ œ œ œ œ œ bœ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ

8 œ
j œ œ bœ bœ œ œ œ
j ˙. œ bœ œ œ nœ œ œ bœ œ œ œ œ œ œ œ œ œ œ œ œ
& Œ
cresc.

?
œœœ œ œœœœ œœœœœœ œ œ œ œ œ œ œ œ œ œ
bœ œ œ œ œœœœ œ œ œ œ œ œ bœ œ œ œ bœ œ œ œ œ œ
11 œ œ œbœ œ œ œ œ œbœ œ œbœ
& œ œ œ œ ‰ œj ‰ j ‰ j ‰ U
J bœ œ œœœœœœœœœœœœœœœœ
f S w
j dim. p π
? ˙ œ œ œ œœ ‰ œ ‰ b œj ‰ œj ‰ U œ œœœœœœœœœœœœœœ
œ ˙ œ J œ bœ œ w
œ S w œ

FIGURE 4.1 Beethoven, Piano Sonata in C Major, op. 53, beginning.

05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 151

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 151

C 48
C 1
D 26
E 11
E 17
F 57
F 2
G 52
A 16
A 13
B 31
B 24

FIGURE 4.2 The numbers of iterations per pitch of the 298 pitches present in
Beethoven, Piano Sonata in C Major, op. 53, beginning (C, 48; C-sharp,
1; D, 26; E-flat, 11; E, 17; F, 57; F-sharp, 2; G, 52; A-flat, 16; A 13; B-flat,
31; B, 24).

to C minor that occurs in mm. 11 and 12. I will not describe the possible reasons
Beethoven chose to allude to B-flat major here, because his rationale has little rele-
vance to this discussion of scales. The important point I wish to make is that tonal
music—and, one might therefore assume, post-tonal music—includes many pitches
foreign to the scales it represents.
The two most notable uses of chromaticism in Figure 4.1 (secondary and modu-
latory or regional) must also be allowed—in principle, if not in exact form—in any
scalar analysis of post-tonal music. Not allowing post-tonal composers the same
kinds of access to temporary and transpositional chromaticism would subject the
analysis of post-tonal music to the same misreading that deriving a twelve-pitch
scale from the Beethoven excerpt would create for tonal music.
Traditional major and minor scales, as well as more recent alternatives such as
octatonic and other nontraditional scales, have dominated traditional music analy-
sis for decades. However, thousands of other scales exist that, for one reason or
another, have not received similar attention. Examining the full panoply of these
scales will help clarify the role they can play in post-tonal music (Lewin 1987;
Slonimsky 1947). I shall limit my studies here to twelve-tone equal temperament to
avoid the incredible combinatorial possibilities engendered by including other tun-
ing systems.
To compute all possible arrangements of from one to twelve different pitch
classes of equal temperament, I use twelve-digit binary (base 2) numbers that indi-
cate the on-off states of each pitch class in the chromatic scale (e.g., 110110110110,
with [1] equaling the presence and [0] equaling the non-presence of each pitch
class). Because the scales in this collection of scale classes contain the pitch of
05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 152

152 HIDDEN STRUCTURE

origin—in this case, C, or pitch class 0—the number of possible scales computes to
2,048, or 211, where the number 2 represents the possible presence or non-presence
of a pitch class in a scale and the exponent 11 indicates the number of available
pitches above C or pitch class 0. A list of these scales appears on the CD-ROM ac-
companying this book. This list begins with the scale [0] and proceeds through
scales of increasing numbers of pitch classes in ascending order. The list ends with
the chromatic scale [0,1,2,3,4,5,6,7,8,9,t,e]. In binary notation, these scales range
from 000000000001 to 111111111111, with only eleven of these twelve digits—the
leftmost eleven—switching states from 0 to 1, and vice versa.
As expected, many of the scales on this list closely resemble one another. For ex-
ample, the major scale [0,2,4,5,7,9,e] has much in common with the Dorian mode
[0,2,3,5,7,9,t]. Both of these scales contain two half steps and five whole steps. In
order to more easily discover these kinds of similarities, I use what I call linear in-
terval vectors. Linear interval vectors (introduced in Chapter 3) result from figuring
each interval of a scale in ascending order from scale degree to scale degree up to
and including the interval produced by the penultimate scale member and the pro-
jected octave above pitch class 0 (12). In other words, for a scale consisting of
eight ascending pitch classes including the octave above the first pitch class, there
would be a total of seven intervals. These intervals are then collected and notated
following the standard musical-vector practice of listing minor seconds first, major
seconds to the right, and so on up through the augmented fourth, with intervals be-
yond the augmented fourth mirror-inverted to their smaller sizes (i.e., perfect fifths
become perfect fourths, minor sixths become major thirds, and so on). Thus, a
scale with two minor seconds and five major seconds produces a linear interval
vector of 250000.
As mentioned earlier, the term linear interval here differentiates this type of vec-
tor from a standard musical vector, which contains every possible interval figured
from all but the last pitch class of a grouping of pitch classes. Standard musical
vectors used for scales with many pitches have such high numbers of certain inter-
vals as to make these vectors less useful for scales. For example, the scale
[0,1,2,3,4,6,8,t] would have a standard musical vector of 474643 and a linear inter-
val vector of 440000, the latter revealing more immediately accessible scale infor-
mation. Linear interval vectors also indicate similarities between scales that
otherwise may not be clearly evident.
Linear interval vectors do not account for the internal order of the intervals they
represent. For example, the major scale and all the related church modes have the
same vector of 250000. The linear interval vector of 250000 allows for many other
arrangements of these intervals as well. For example, the 250000 vector also in-
cludes the ascending melodic minor ([0,2,3,5,7,9,e]) scale. Several other interesting
scales, such as [0,1,2,4,6,8,t] and [0,1,3,4,6,8,t], share this vector as well. Although
these different arrangements that have the same vector may seem unrevealing
when analyzing post-tonal music, they provide very helpful information regarding
scale similarity.
05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 153

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 153

The harmonic minor [0,2,3,5,7,8,e] scale resolves to the 340000 linear interval vec-
tor. Whole-tone ([0,2,4,6,8,t] 060000), octatonic ([0,1,3,4,6,7,9,t] or [0,2,3,5,6,8,9,e]
both as 440000), pentatonic ([0,2,4,7,9] 032000), and chromatic ([0,1,2,3,4,5,6,7,8,9,t,e]
w00000, with “w” representing the number 12) provide examples of other often
used scales and their linear interval vectors.
Because standard pitch-class sets can contain up to twelve pitch classes and
theoretically extend over any period of musical time, distinguishing scale sets from
pitch-class sets can occasionally prove difficult. Therefore, I limit pitch-class sets in
this book to those groups of pitches that appear in relatively close proximity to one
another and that occur over a period of musical time of less than a phrase. Scale
sets, on the other hand, typically result from analyzing one or more phrases of
music. Scale sets may also have implied scale membership or, conversely, include
pitch classes foreign to the scale set, representing other distinctions between scale
sets and pitch-class sets. Although these distinctions may seem somewhat vague,
they will become clearer when analyzing scales in specific musical examples later
in this chapter.

VECTOR CLASSES AND METACLASSES

As previously discussed, scales with equivalent linear interval vectors bear strong
resemblance to one another and thus belong to the same linear interval vector
class. Collecting all possible scales into linear interval vector classes produces
77 distinct scale types, reduced from the 2,048 possible scales. Figure 4.3 lists all 77
linear interval vector classes from the smallest number of pitch classes (the single-
interval or octave scale) to the largest number of pitch classes (the continuous
half-step or chromatic scale). Every possible equal-tempered scale belongs to one
of these 77 classes and shares one or more characteristics with other members of
that same class. Six of these 77 vector classes have unique single members: those
representing scales consisting of one mirrored interval, as in 002000, where the two
minor thirds identify a minor third and its mirror, the major sixth. The other num-
bers of differing intervals in linear interval vector classes range from two different
intervals (as in 210000) to four different intervals (as in 121100).
Viewing a list such as that shown in Figure 4.3 proves much more productive
than consulting the entire list of 2,048 scales as presented on the CD-ROM accom-
panying this book. Each line in this figure contains two items, a version of the scale
in pitch-class numbers and the linear interval vector of that same scale. Interest-
ingly, even this more compact representation contains several rather straight-
forward scale classes that I find less useful when distinguishing between analyzing
or composing music. Therefore, I use a special metaclass notation indicating only
the number of half steps each linear interval vector class contains. This shorthand
notates the number of half steps in the scale followed by “⍀” (Greek letter omega),
05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 154

154 HIDDEN STRUCTURE

[0] 000000 [0,1,2,3,5,7] 320010

[0,1] 200000 [0,1,2,3,5,8] 311100
[0,2] 020000 [0,1,2,3,6,9] 303000
[0,3] 002000 [0,1,2,4,6,8] 230100
[0,4] 000200 [0,1,2,4,6,9] 222000
[0,5] 000020 [0,1,3,5,7,9] 141000
[0,6] 000002 [0,2,4,6,8,t] 060000
[0,1,2] 210000 [0,1,2,3,4,5,6] 600001
[0,1,3] 111000 [0,1,2,3,4,5,7] 510010
[0,1,4] 101100 [0,1,2,3,4,5,8] 501100
[0,1,5] 100110 [0,1,2,3,4,6,8] 420100
[0,1,6] 100011 [0,1,2,3,4,6,9] 412000
[0,2,4] 020100 [0,1,2,3,5,7,9] 331000
[0,2,5] 011010 [0,1,2,4,6,8,t] 250000
[0,2,6] 010101 [0,1,2,3,4,5,6,7] 700010
[0,2,7] 010020 [0,1,2,3,4,5,6,8] 610100
[0,3,6] 002001 [0,1,2,3,4,5,6,9] 602000
[0,3,7] 001110 [0,1,2,3,4,5,7,9] 521000
[0,4,8] 000300 [0,1,2,3,4,6,8,t] 440000
[0,1,2,3] 301000 [0,1,2,3,4,5,6,7,8] 800100
[0,1,2,4] 210100 [0,1,2,3,4,5,6,7,9] 711000
[0,1,2,5] 201010 [0,1,2,3,4,5,6,8,t] 630000
[0,1,2,6] 200101 [0,1,2,3,4,5,6,7,8,9] 901000
[0,1,2,7] 200020 [0,1,2,3,4,5,6,7,8,t] 820000
[0,1,3,5] 120010 [0,1,2,3,4,5,6,7,8,9,t] t10000
[0,1,3,6] 111001 [0,1,2,3,4,5,6,7,8,9,t,e] w00000
[0,1,3,7] 110110
[0,1,4,7] 102010
[0,1,4,8] 101200
[0,2,4,6] 030001
[0,2,4,7] 021010
[0,2,4,8] 020200
[0,2,5,8] 012100
[0,3,6,9] 004000
[0,1,2,3,4] 400100
[0,1,2,3,5] 310010
[0,1,2,3,6] 301001
[0,1,2,3,7] 300110
[0,1,2,4,6] 220001
[0,1,2,4,7] 211010
[0,1,2,4,8] 210200
[0,1,2,5,8] 202100
[0,1,3,5,7] 130010
[0,1,3,5,8] 121100
[0,1,3,6,9] 113000
[0,2,4,6,8] 040100
[0,2,4,6,9] 032000
[0,1,2,3,4,5] 500010
[0,1,2,3,4,6] 410001
[0,1,2,3,4,7] 401010
[0,1,2,3,4,8] 400200

FIGURE 4.3 A list of all 77 linear interval vector classes.

05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 155

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 155

signifying “to the end of all of the possible numbers.” Therefore, 0⍀ represents all
scales containing no half steps, 1⍀ indicates all scales containing one half step, and
so on. The representation 4+⍀ signifies all scales containing four or more half
steps. Thus, one can quickly scan a scale for the number of half steps it contains
and then denote its metaclass with a number and a symbol. The whole-tone scale
therefore belongs to the 0⍀ metaclass, the major scale and the church modes are
included in the 2⍀ metaclass, the harmonic minor scale belongs to the 3⍀ meta-
class, and the octatonic scale appears in the 4+⍀ metaclass.
As one might suspect, the 0⍀ metaclass contains a large number of scale classes
(21). On the other hand, the 3⍀ scale metaclass has few representatives (8), as do
the 2⍀ (13) and 1⍀ (13) metaclasses. Although the 4+⍀ metaclass contains the
most scale classes, its total number of classes (22) closely approximates the num-
ber of classes in the 0⍀ metaclass (21).
Scales with vectors that do not match may also relate in other ways. For exam-
ple, the scales [0,2,5,6,7,t] and [0,3,6,9] produce the quite different vectors of
222000 and 004000 respectively, not matching in class or metaclass. As Figure 4.4a
shows, however, both of these scales are symmetrical, the first consisting of mir-
rored major second/minor third/minor second combinations, and the second out-
lining a diminished seventh chord. Symmetry binds otherwise quite differently
constructed scales together in special ways. Even inversionally symmetrical scales
as diverse as those shown in Figure 4.4b (of 222000 and 250000 vectors) deserve
such recognition because they both belong to the same metaclass.
Scales consisting of one interval—even though that interval may be different—
can also relate apparently quite diversely vectored scales. As example, major-third
(000300) and minor-third (004000) scales share this notion; furthermore, both are
symmetrical (see Figure 4.4c).
Patterned scales—scales consisting of a single two- or three-pitch pattern—also
relate, as demonstrated in Figure 4.4d, where scales with 440000 and 020200 vec-
tors share their consistent use of repeating patterns. Although the analyses that fol-
low do not particularly dwell on these possibilities, because doing so might
confuse the points being made, readers are nonetheless encouraged to discover
further such relationships themselves.
Scale use in diverse music requires different analytical techniques. For example,
in order that I not confuse the actual meanings of secondary functions and modula-
tion in tonality with similar concepts in post-tonality, I will substitute the terms
transient chromaticism, transpositional chromaticism, rotational chromaticism, and
transformational chromaticism. Transient chromaticism means that the chromatic
pitches are temporary and do not belong to the scale in current use. Transposi-
tional chromaticism, in contrast, indicates that a transposition of a current scale has
replaced the current scale in the music under analysis. In Figure 4.1, the F-sharp in
m. 2 represents a form of transient chromaticism, and the B-flat in m. 5 and beyond
is an example of transpositional chromaticism. Unlike tonal music, however, not
only may post-tonal music modulate to a transposed version of a current scale, but
05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 156

156 HIDDEN STRUCTURE

a.
⵩ ∩ ⵩
& w∩w ⵩ w # w w b w w ⵩ ⵩ b w⵩ w w
w bw

b.
∩ ∩∩
& w∩w b w⵩# w⵩w b w w ∩ ∩ ∩w w w w
w bw bw w

c.
⵩ ⵩
& w 䉭 w䉭 # w
䉭 w ⵩ ⵩ w w
w bw bw

d.
∩ ∩
& ∩w b w∩w b w∩b w n w w w 䉭 w∩# w䉭#w w
w w
䉭 = major third
⵩ = minor third
∩ = major second
= minor second

FIGURE 4.4 Four sets of scales related by symmetry (a, b, and c) and a repeating
pattern (d).

may also modulate to a new scale entirely (similar in concept to major–to-minor key
modulation in tonal music). If a different new scale in use belongs to the same vec-
tor class as the former scale, I use the term rotational chromaticism, because the
various intervals in the scale in use have rotated to different positions. If a different
new scale belongs to another vector class than the former scale, I use the term
transformational chromaticism, meaning that the scale has actually transformed
into combinations of new pitch classes and/or orders and types of intervals. Unlike
tonal modulation between different modes (i.e., major to minor or vice versa),
transformational chromaticism involves fundamental shifts between truly different
scales, including scales containing different numbers of pitches.
The following more formal definitions of these terms should help clarify any re-
maining confusions over their meanings:
Transient chromaticism includes pitches foreign to a current scale that appear
briefly, typically resolve to their associated scale degree, and/or appear in weak
metric, rhythmic, and/or agogic circumstances;
05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 157

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 157

Transpositional chromaticism includes pitches foreign to a current scale that re-

peat or otherwise become consistent over time, alter their native scale versions,
and/or that appear in strong metric, rhythmic, and/or agogic circumstances, and
which create a transposed version of the current scale in use;
Rotational chromaticism includes pitches foreign to a current scale that repeat or
otherwise become consistent over time, alter their native scale versions, and/or
that appear in strong metric, rhythmic, and/or agogic circumstances, and which
create a different scale, but a scale of the same vector class;
Transformational chromaticism includes pitches foreign to a current scale that
create—through repetition and in strong metric, rhythmic, and/or agogic circumstances
—a demonstrably new scale of a different vector class.
Of course, defining what these terms mean and discovering their presence and
placement in post-tonal music pose quite different problems. Also, having the
above-described terms replace the terms secondary and modulatory chromaticism
as it occurs in tonal music does not really provide much information on how to dis-
tinguish them in post-tonal music. Therefore, I will—after a brief discussion of im-
plied pitches and the computational methods I use to detect scales—provide a
series of brief analyses demonstrating scale analysis in post-tonal music.
As mentioned earlier, scales may appear even if certain scale members are ab-
sent. For example, Figure 4.5 shows an example of implied scale membership—
eleven measures from Beethoven’s Piano Sonata in E-flat, op. 7 (third movement).
The 181 pitches here clearly define the E-flat minor scale. Notable about this set of
pitches, however, is that the submediant scale degree of the key—C-flat—does not
appear anywhere in these measures. In fact, the resulting scale might not even be
analyzed as minor by a novice, but possibly as some form of a six-pitch scale. How-
ever, as evidenced by the tonal nature of this music, this passage is in the key of
E-flat minor, with an implied C-flat, and with the D-flat and D-natural as subtonic
and leading-tone pitches in the melodic minor version of this scale.
In order that readers not suspect that this music represents a rarity, I note that
the initial C major statement of the theme of the Beethoven Piano Sonata no. 21,
op. 53, in Figure 4.1, mm. 1–4, does not contain an F-natural, only an F-sharp (m. 2).
Clearly, however, the first four measures of this sonata, by key signature and repeti-
tion of the tonic chord, define C major as the key. Even though implied member-
ships such as this F-natural and the previously mentioned C-flat can encourage
creative analyses, they must be allowed in scales in both tonal and post-tonal mu-
sic. Without these implied memberships, even tonal music would be reduced to
scales of fewer than seven pitches, or non-tonal scales.
Implied pitches are therefore as important to tonal analysis as the inclusion of
pitches foreign to a key. Any realistic analytical process applied to tonal music must
take implied pitches into consideration or risk seriously misconstruing the diatonic
nature of the music. I believe that the same considerations must also be afforded to
post-tonal music and will endeavor to prove this as this chapter progresses.
Aside from counting pitch classes, the program Scale on the CD-ROM accompa-
nying this book adds the durations in the third position of each note-event for all
05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 158

158 HIDDEN STRUCTURE

bb 3 3
& b b bb 4 Ó œ œ œ œ œ œ œ œ
œ œœ œœ œœ œœ œœ œœœ œœ œœ œœ
Piano
3
? b b b b 43
bb ∑ œ œ œ œ œ œ œ œ œ nœ œ œ œ œ œ œ œ œ
œ œ œ œ nœ œ œ œ œ

5
b
& b bbbb œ œ œ œ œ œ œ œ œ œ œ œ
œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ bœ œ œ œ œ

? bb b b œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ n œ œ œ b œ œ œ œ œ œ
bb œ œ œ œ œ œ œ nœ œ œ œ œ

9
b
& b bbbb œ œ œ œ œ œ œ œ œ
œœ œœ œœ œœ œœ œœ œ œ œ œ œ œ
? b b b œœ œœ œœ œœ œœ œœ
bbb œ œ œ œ œ œ œ œ œ
œœ œœ œœ

FIGURE 4.5 Beethoven, Piano Sonata op. 7, mm. 97–105.

pitch classes in the indicated work, or some designated section of this work. These
duration additions are then paired with their pitch classes, such as in “(2 15238),”
where 2 indicates the pitch class, and 15238 represents the total duration in thou-
sandths of a second of that pitch class. This list of pitch classes and their total
durations, presented in order of longest duration to shortest, then acts as a model
for creating parallel lists of possible scales, with each successive scale having one
additional class—the one of the briefest overall duration—removed. Each scale
possibility is presented in three ways: pitch classes transposed to begin on 0 in as-
cending order, as a linear interval vector, and by original pitch classes in ascending
order. Figure 4.6 presents an example of this representation. From these lists users
can select the pitch-class series that they feel—after listening to and otherwise ana-
lyzing the music under study—best represents the underlying scale in use. When
an analyst chooses a scale with removed pitches, these removed pitches become
transient chromaticism to an analyzed scale.
05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 159

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 159

[0,1,2,3,4,5,6,7,8,9,t,e] w00000,[0,1,2,3,4,5,6,7,8,9,t,e]
[0,1,2,4,5,6,7,8,9,t,e] t10000,[0,1,2,4,5,6,7,8,9,t,e]
[0,1,4,5,6,7,8,9,t,e] 901000,[0,1,4,5,6,7,8,9,t,e]
[0,4,5,6,7,8,9,t,e] 800100,[0,4,5,6,7,8,9,t,e]
[0,1,2,3,4,5,6,7] 700010,[4,5,6,7,8,9,t,e]
[0,1,2,3,4,5,6] 600001,[5,6,7,8,9,t,e]
[0,1,2,3,4,6] 410001,[5,6,7,8,9,e]
[0,1,2,3,5] 310010,[6,7,8,9,e]
[0,1,3,5] 120010,[6,7,9,e]
[0,3,5] 011010,[6,9,e]
[0,3] 002000,[6,9]
[0] 000000,[6]

FIGURE 4.6 A series of possible scales, with each successive scale possibility lack-
ing one of the previous scale’s notes, the one with the shortest overall
duration. Each line here represents the scale’s transposed pitch-class
set beginning on 0, its linear interval vector, and the original pitch-
class set in ascending order.

These lists of scale possibilities can also be compared with other scale analyses
representing phrases, sections, works, and so on, for their equivalent scale content
or linear interval vector similarity. This comparison may then reveal scales that
support even longer sections of music or that demonstrate other kinds of similarity
(e.g., symmetry, as discussed previously). What would require hours of exhaustive
processing by hand takes but a few seconds when using a computer to produce the
results.
The Scale program lacks the ability to make best guesses for implied scale mem-
bers. I have resisted adding code to the program to make these guesses algorithmi-
cally, because postulating such membership in a scale has provided my most
interesting personal challenges. I rarely add more than one or two implied pitches—
if any—and then only to scales that otherwise consist of five or six pitches or fewer.
Implied pitches typically occur between awkward leaps in otherwise stepwise
scales, where they produce relatively unnatural motions between pitch classes
(e.g., an octatonic scale minus one member). A logical addition to Scale would be
to have the program relate scale class sets to the full 2,048 possibilities, returning
those sets with the greatest likelihood of possessing the missing members.
Although I could easily have figured dynamics, meter, and other considerations
into the weighting process used by Scale, I have refrained from doing so for a vari-
ety of reasons. First, such considerations do not consistently map to all post-tonal
music. In fact, when composing my own music—as I believe I can usefully argue—
I avoid allowing these elements to take active roles in establishing scales. I use
05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 160

160 HIDDEN STRUCTURE

meter, for instance, to keep performers together, not to accent or otherwise influ-
ence pitch-class prominence. I also use softer dynamics as often as louder dynam-
ics to articulate important pitches in a scale. Finding many of my own predilections
in other post-tonal music, I have opted to avoid the apparent inconsistencies of
weighting these factors by including pitch-class duration additions only (see Ler-
dahl 1989 for a somewhat different view of this).
Many composers, particularly polytonal composers, use more than one scale si-
multaneously. Although analyzing post-tonal music for two or more simultaneous
scales is possible by grouping pitches by voice or other principle, I have not in-
cluded such analysis processes in Scale to avoid confusing the basic principles dis-
cussed in this chapter. For example, non-scale and implied pitches become much
less plausible in polyscalar environments. Likewise, the argument against scales
serving as a basis for post-tonal music analysis on the basis of the difficulty of
hearing such scales becomes even stronger if one asserts that more than one scale
exists at the same time.
The choice of phrase lengths and placements in the following examples results
from my own interpretations of the music shown. Obviously, many analyses of
phrasing are possible; those given here represent just one set of decisions. Other
choices for phrase length may produce other scales or may even more strongly
support the case for the scales that I have chosen. To lessen bias in my approach
to determining scales in this music, I made all decisions regarding phrasing prior to
determining scale content.

VARÈSE’S DENSITY 21.5

Varèse’s Density 21.5 (1936) for solo flute, shown in its entirety in Figure 4.7, serves
as a good work for scale determination, because the single line and post-tonal chro-
maticism lend themselves logically to scale analysis (Bernard 1977; Perle 1990).
The fairly clear-cut phrase delineations provided by the composer also help to cap-
ture scale material in relatively unambiguous ways. Figure 4.8 presents a phrase-by-
phrase computer analysis of Varèse’s Density for reference during the following
discussion.
The first easily differentiated grouping of pitches in Density, ending on the third
beat of m. 3, provides a distinct five-pitch scale [1,4,5,6,7] with the resulting linear
interval vector of 301001. Varèse has strengthened the sense of scale here by
adding no new pitch classes in the following two-plus measures. The next three
measures (from m. 6 up to and including the C-natural that ends m. 8), however, flesh
out more details for a potential scale by creating a more complex [0,1,4,5,6,7,9,t]
eight-pitch series, yielding a 521000 linear interval vector when combined with ma-
terial from mm. 1–5. The music following m. 8 provides pitch classes 2, 3, and 8 in
05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 161

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 161

q = 72 3
3 3
& c œ œ #œ ˙. œ #œ. #œ œ ˙ ˙ Œ nœ #œ
œ œ œ œ # œj œ # œ œ œ
F f F p f F
5
œ œ b œ œj b œ . ˙
3 3
j 3
& œ œ . œ # œ œ ‰ # œj œ- œ- œ b œ ˙ bœ œ œ ˙ œ ≈œ
f p F p subito f ƒ F
subito

œ bœ. nœ. ˙ œ
œ œ œ n˙. œ # œJ œ 3# œj œ . # œ œj # œ .
10

& b œ œ œJ œ œ #œ œ J ‰ Œ
3 3
Ï 3
f 3
ƒ
> w>
œ # œ n œ œ œ œ œ. ‰ œ œ 3œ œ # œ
3
15 >
& J # œ # œ> ˙ œ # œj œ œ œ n œ œ # œ ‰ œ
J
p f p 3p ƒ p subito p
20 #˙ #˙ n˙ œ 3 #œ #œ œ> ≈ œ. œ.
&˙ J ‰ œ # œ œ n œ . # œj œ n œ # œ œ ∑
-
f ƒ p subito p f p3 subito

+ + +++
sharply articulated
+ + ++ + + + +>
24
-œ œ .œ j b œ
& ‰ œ # œ Œ Œ œ œ œ Œ # œj œ ‰ ‰ J R ‰ . Œ ‰ ‰ # œj J ‰ Œ # œ œ # œ ‰ Œ 4 ‰ # œ Œ œ ‰ J Œ 4
3 3 3
5 œ 3
3
. . . >. >p- p >
P P> >> p P P P
q = 60
> >
œ #œ > œ #œ œ> œ
29
#œ. œ #œ ‰ c #œ ≈ œ> œ # œ œj # œ œ 3 #œ œ
&43 J nœ œ
#œ œ #œ nœ >b œ .
ƒ ƒ >
3

œ . q√= 72 q = 60
32 œ
# œ . œ œ œ # œ ˙ . # œ . œ œ œ œ œ . œ œ
> œ œ œ >>
œ # œ œ œ œ> ˙
&J œ œ œ œ œ œ J œ œ œ≈
Ï 3 3 3 3
3 3 3
Ï
FIGURE 4.7 Varèse, Density 21.5 for solo flute.
05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 162

162 HIDDEN STRUCTURE

œ> > > >

b ˙>. œ œ b œ>. œ ≈ œ- œ- b œ- Œ b œ 3œ b œ> œ ‰ œ n œ> œ œ ≈ b œ b œ. œ n œ. ‰
37

& J J J
ƒ p f p
subito
3
f 3p f p
41 q = 72 3
3
œ>b œ œ # ˙ .
& # œ- # œ n œ ˙ . œ Œ Œ ‰ #œ #œ œ #œ #œ œ œ œ œ œ b˙ nœ
p >
p
√
3> > œ œ > œ> œ> œ3 3
> - q = 60
œ > œ œ œ œ œ œ œ œ œ œ œ ≈ œ œ . œ . œ œ ten. j
> 3
46 œ œ ≈œ
&‰ J J J bœ œ œ
- - -
ƒ 3 Ï
51 3 œ # œ>.. œ œ # œ>.. œ œ n œ> œ œ
#œ œ 3
j
& œ nœ œ n ˙> œ # œ œ œ> œ œ # œ œ œ œ .
pf p Ç Ç p3 ƒ > > > p π
> ˙ ˙
œ > # >œ # œ . # œ # ˙
œ>
56 cresc. molto 3

& b œ œj œ œ ˙ œ b œ Jœ Ó
œ> œ> œ œ> œ
> œ> œ> Ï
3
ƒ
FIGURE 4.7 continued

fairly quick order, with pitch class 11 not appearing until m. 18, thus completing the
full chromatic scale in the music from the beginning of the work to this point. It
would be easy here to analyze this opening passage as freely chromatic or, at the
least, to reduce it to a twelve-pitch scale were it not for the fact that Varèse has so
carefully defined important pitch classes by repetition and hyperextended dura-
tion, as well as by grouping notes as motives and variations of those motives
throughout the first section of this piece.
Analyzing post-tonal music for nonharmonic tones is not a new concept (see par-
ticularly Solomon 2003 and 2005; Straus 2005). For example, George Perle (1990,
11–12) argues for a diminished seventh scale (i.e., C-sharp, E-natural, G-natural,
B-flat) scale (004000) for the first nine measures of this work, even though these
same pitches plus intervening pitches—D-sharp, F-sharp, A-natural, and C-natural,
both actual and implied—could provide the basis for analyzing this music originat-
ing from a major octatonic scale (minus E-flat).
05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 163

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 163

Section 1 [0,1,2,3,4,6] 410001 [5,6,7,8,9,e]

[0,1,2,3,5] 310010 [6,7,8,9,e]
mm. 1 - 8 [0,1,3,5] 120010 [6,7,9,e]
[0,3,5] 011010 [6,9,e]
[0,1,4,5,6,7,9,t] 521000 [0,1,4,5,6,7,9,t] [0,3] 002000 [6,9]
[0,1,4,6,7,9,t] 331000 [0,1,4,6,7,9,t] [0] 000000 [6]
[0,1,4,6,7,t] 222000 [0,1,4,6,7,t]
[0,3,5,6,9] 113000 [1,4,6,7,t] rest of m. 36-m. 40
[0,5,6,9] 102010 [1,6,7,t]
[0,1,4] 101100 [6,7,t] [0,1,2,3,e] 400100 [0,1,2,3,e]
[0,1] 200000 [6,7] [0,1,2,3] 301000 [0,1,2,3]
[0] 000000 [7] [0,2,3] 111000 [0,2,3]
[0,3] 002000 [0,3]
mm. 9-14 [0] 000000 [3]

[0,1,2,3,4,8,9,t] 610100 [0,1,2,3,4,8,9,t] Section 3

[0,1,2,3,4,9,t] 510010 [0,1,2,3,4,9,t]
[0,1,2,3,4,9] 401010 [0,1,2,3,4,9] mm. 41-first note of 50
[0,1,2,3,4] 400100 [0,1,2,3,4]
[0,1,2,4] 210100 [0,1,2,4] [0,1,3,4,5,6,7,8,9,t] 820000 [1,2,4,5,6,7,8,9,t,e]
[0,1,3] 111000 [1,2,4] [0,1,4,5,6,7,8,9,t] 711000 [1,2,5,6,7,8,9,t,e]
[0,1] 200000 [1,2] [0,1,4,5,6,7,9,t] 521000 [1,2,5,6,7,8,t,e]
[0] 000000 [1] [0,1,4,5,6,7,t] 412000 [1,2,5,6,7,8,e]
[0,1,5,6,7,t] 311100 [1,2,6,7,8,e]
mm. 15-23 [0,1,6,7,t] 211010 [1,2,7,8,e]
[0,1,6,t] 110110 [1,2,7,e]
[0,1,2,3,4,5,6,7,8,9,t,e] w00000 [0,1,2,3,4,5,6,7,8,9,t,e] [0,5,9] 001110 [2,7,e]
[0,2,3,4,5,6,7,8,9,t,e] t10000 [0,2,3,4,5,6,7,8,9,t,e] [0,5] 000020 [2,7]
[0,1,2,3,4,5,6,7,8,9] 901000 [2,3,4,5,6,7,8,9,t,e] [0] 000000 [2]
[0,1,2,3,4,5,7,8,9] 711000 [2,3,4,5,6,7,9,t,e]
[0,1,3,4,5,7,8,9] 521000 [2,3,5,6,7,9,t,e] rest of m. 50 to first beat of m. 53
[0,3,4,5,7,8,9] 412000 [2,5,6,7,9,t,e]
[0,3,5,7,8,9] 222000 [2,5,7,9,t,e] [0,5,6,7,8,9] 401010 [0,5,6,7,8,9]
[0,3,5,8,9] 113000 [2,5,7,t,e] [0,5,6,7,8] 300110 [0,5,6,7,8]
[0,2,5,6] 111001 [5,7,t,e] [0,5,6,8] 110110 [0,5,6,8]
[0,2,6] 010101 [5,7,e] [0,1,3] 111000 [5,6,8]
[0,4] 000200 [7,e] [0,2] 020000 [6,8]
[0] 000000 [e] [0] 000000 [8]

Section 2 last 3 beats of m. 53 to end of piece

mm. 24-28 [0,1,2,3,4,5,6,7,9,t,e] t10000 [0,1,2,3,4,5,6,7,9,t,e]
[0,1,2,4,5,6,7,9,t,e] 820000 [0,1,2,4,5,6,7,9,t,e]
[0,1,2,3,7] 300110 [1,2,3,4,8] [0,1,2,4,5,7,9,t,e] 630000 [0,1,2,4,5,7,9,t,e]
[0,1,3,7] 110110 [1,2,4,8] [0,1,2,4,5,7,t,e] 521000 [0,1,2,4,5,7,t,e]
[0,1,3] 111000 [1,2,4] [0,1,2,4,7,t,e] 412000 [0,1,2,4,7,t,e]
[0,1] 200000 [1,2] [0,1,3,6,9,t] 222000 [1,2,4,7,t,e]
[0] 000000 [2] [0,1,3,9,t] 220001 [1,2,4,t,e]
[0,1,3,t] 120010 [1,2,4,e]
mm. 29-beat 1 of m. 36 [0,2,9] 11010 [2,4,e]
[0,2] 020000 [2,4]
[0,1,2,3,4,5,6,7] 700010 [4,5,6,7,8,9,t,e] [0] 000000 [4]
[0,1,2,3,4,5,6] 600001 [5,6,7,8,9,t,e]

FIGURE 4.8 Computer scale analysis of the phrases of the three sections of Varèse,
Density 21.5 (see Figure 4.7), with author’s determination of best scale
possibility in boldface.
05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 164

164 HIDDEN STRUCTURE

Post-tonal music also contains scalar patterns that can help delineate a scale
in use. For example, in Varèse’s Density 21.5 in Figure 4.7, the second pitch of m. 6,
G-natural, up to the D-natural beginning m. 11 reveal the possible scale of G, A,
B-flat, C, D-flat, and D-natural.
Even though the music does not place these pitches one after another in typical
scale order, the fact that the music in these measures contains only these pitches
clearly suggests the quasi-tonal scale pattern. Without the D-flat (possibly a chro-
matic passing tone), this pattern could be construed as a G minor scale, even
though the surrounding context might belie that analysis. Although such scale pat-
terns do not appear as commonly in post-tonal music as in tonal music, they defi-
nitely require special observation when they do occur.
Developing a single scale from mm. 1–18, or, more likely, m. 23, the end of the
first section of this piece, thus appears to have limitations. Furthermore, mm. 9–14
of Density constitute such a distinct possibility for a separate phrase that the likeli-
hood of another scale in use should not be overlooked. This second phrase re-
duces to a scale of [0,1,2,3,4,8,9,t] having a linear interval vector of 610100.
Although this vector differs from the vector of the first eight measures of this piece
(521000), changing but one pitch class of the scale (4 to 5) produces a linear inter-
val vector of 521000, which demonstrates this scale’s close proximity to the scale
analyzed in mm. 1–8. Also, the interval vector of 610100 for the unaltered scale of
mm. 9–14 belongs to the same vector metaclass (4+⍀) as 521000, thus further em-
phasizing this passage’s similarity to the scale of the first eight measures. Ulti-
mately, deciding whether a section contains one or more separate scales is more a
matter of individual preference than of applying a particular principle. In this book,
I do not analyze regions separately from overall scales, so that the basic points of
my arguments remain clear.
Measures 15–23 of Density contain all twelve pitches of the chromatic scale.
However, viewing smaller and smaller numbers of scale pitch classes—discarding
the ones with less overall durational value—produces the scale of [2,3,5,6,7,9]
yielding the 521000 linear interval vector, the identical vector class that resulted
from the analysis of mm. 1–8 of this work. Identifying this 521000 scale for mm. 15–
23 requires the elimination of four pitch classes from the collection of twelve pitch
classes that occur during the passage. As the computer analysis in Figure 4.8
shows, however, these four pitch classes (0, 1, 4, 8, or C-natural, C-sharp, E-natural,
and G-sharp) all have less durational value than the eight included in this analysis,
and thus make sense as transient chromaticism.
Looking at the entire first section of Density by phrase then generates the follow-
ing three linear interval vectors: 521000 (mm. 1–8), 610100 (mm. 9–14), and 521000
(mm. 15–23). Only the third vector requires note deletion, with the first two
phrases using all of their pitches to build their relevant scales (shown in boldface
in Figure 4.8). I should note here that the alignment of scale boundaries with
phrase and section boundaries requires assumptions that may not necessarily be
valid. For example, in tonal music, many phrases contain modulations, and less often
05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 165

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 165

do new phrases suddenly begin in different keys. Even with this in mind, however,
it makes more sense at this point to analyze by phrase and section rather than by
some other technique, simply because in Western music phrases and sections of-
ten fall within single keys.
Note that my decision to suppress low-duration pitches here is based not on a
fixed threshold or on any innate sense of these pitches as foreign to the music be-
ing analyzed, but on a desire to find identical or closely related scale structures in
different phrases that I feel tend to unify the music. Clearly, then, my search for
scale similarities as described here does not result from a neutral point of view, but
rather from a focused desire to discover them.
The second section of Density 21.5 contains three distinct phrases—mm. 24–28,
m. 29 through the first pitch of m. 36, and the second pitch of m. 36 through m. 40,
as indicated in Figure 4.8. Using the Scale program to analyze the note-events of
these phrases produces the identical 310010 class in each case. In phrase 1, the
chosen linear interval vector does not require any transient chromatic elimina-
tions. Contrastingly, in the second phrase, three pitches must be removed to reveal
this vector. The third phrase requires that only one pitch be omitted.
The third section of Density also consists of three phrases—m. 41 to the first
note of m. 50, the rest of m. 50 to the first beat of m. 53, and the final three beats
of m. 53 to the end of the piece, as shown in Figure 4.8. The first and last phrases
here produce 521000 linear interval vector classes, with the second phrase of
401010 linear interval vector sandwiched in between. The phrase-by-phrase analy-
sis of this section resembles the phrase-by-phrase analysis of section 1, both pro-
ducing A–B–A scale structures at this micro-level, with the scales of 521000 linear
interval vector acting as the A in both sections.
Analyzing the three sections of this piece as entire sections, rather than as sepa-
rated phrases within these sections, produces a nine-pitch linear interval vector
of 630000 in each case, as shown in Figure 4.9. In each section, the three weakest
overall-duration pitches have been removed in order to create the scale vector.
Given the twelve-pitch content of each of these sections, this nine-pitch vector pro-
vides a chromatic framework without the necessity of resorting to the full chro-
matic scale. Since there are three unique classes containing nine pitches—800100,
711000, and 630000—the appearance of the shared 630000 linear interval vector
here does not seem coincidental. Interestingly, the only phrase analyzed in Figure
4.8 that by itself contains a 630000 linear interval vector as a possibility is the final
phrase of the work (with two extracted transient chromatic pitches), supporting
the case for the 630000 vector analysis of the entire piece.
Although the three scales shown in boldface in Figure 4.9 differ somewhat in ex-
act pitch-class content (compare the right-hand boldface non-transposed scales),
the underlying linear interval vector of 630000 unifies them. The 630000 scales of
sections 1 and 2 are transpositionally related (five neighboring minor seconds, with
the sixth minor second removed by one major second), while the scales of sections
2 and 3 have a permutational relationship (the scale for the last section having four
05_DAS_Chap4_pp145-188 1/29/09 1:34 PM Page 166

166 HIDDEN STRUCTURE

Section 1. mm 1-23

[0,1,2,3,4,5,6,7,8,9,10,e] w00000 [0,1,2,3,4,5,6,7,8,9,t,e]

[0,1,2,3,4,5,6,7,9,10,e] t10000 [0,1,2,3,4,5,6,7,9,t,e]
[0,1,2,4,5,6,7,9,10,e] 820000 [0,1,2,4,5,6,7,9,t,e]
[0,1,2,4,6,7,9,10,e] 630000 [0,1,2,4,6,7,9,10,e]
[0,1,3,5,6,8,9,10] 440000 [1,2,4,6,7,9,t,e]
[0,1,3,5,6,9,10] 331000 [1,2,4,6,7,t,e]
[0,3,5,6,9,10] 222000 [1,4,6,7,t,e] *
[0,3,6,9,10] 113000 [1,4,7,t,e]
[0,6,9,10] 111001 [1,7,t,e]
[0,3,4] 101100 [7,t,e]
[0,3] 002000 [7,t]
[0] 000000 [7]

Section 2. mm. 24-40

[0,1,2,3,4,5,6,7,8,9,t,e] w00000 [0,1,2,3,4,5,6,7,8,9,t,e]

[0,1,2,3,4,5,6,7,8,9,e] t10000 [0,1,2,3,4,5,6,7,8,9,e]
[0,1,2,3,4,6,7,8,9,e] 820000 [0,1,2,3,4,6,7,8,9,e]
[0,1,2,3,4,6,8,9,e] 630000 [0,1,2,3,4,6,8,9,e]
[0,1,2,3,6,8,9,e] 521000 [0,1,2,3,6,8,9,e] †
[0,1,2,3,6,9,e] 412000 [0,1,2,3,6,9,e]
[0,1,2,5,8,t] 222000 [1,2,3,6,9,e] *
[0,1,4,7,9] 113000 [2,3,6,9,e]
[0,3,6,8] 012100 [3,6,9,e]
[0,3,6] 002001 [3,6,9]
[0,3] 002000 [3,6]
[0] 000000 [6]

Section 3. mm. 41-61

[0,1,2,3,4,5,6,7,8,9,t,e] w00000 [0,1,2,3,4,5,6,7,8,9,t,e]

[0,1,2,4,5,6,7,8,9,t,e] t10000 [0,1,2,4,5,6,7,8,9,t,e]
[0,1,3,4,5,6,7,8,9,t] 820000 [1,2,4,5,6,7,8,9,t,e]
[0,1,3,4,5,6,7,9,t] 630000 [1,2,4,5,6,7,8,t,e]
[0,1,3,4,5,6,7,t] 521000 [1,2,4,5,6,7,8,e] †
[0,1,3,4,5,6,t] 420100 [1,2,4,5,6,7,e]
[0,1,3,5,6,t] 230100 [1,2,4,6,7,e]
[0,1,3,6,t] 121100 [1,2,4,7,e]
[0,1,6,t] 110110 [1,2,7,e]
[0,5,9] 001110 [2,7,e]
[0,9] 002000 [2,e]
[0] 000000 [2]

FIGURE 4.9 Computer analysis of Varèse, Density 21.5, by complete section with
author’s determination of best scale possibility in boldface and sec-
ondary possibilities marked by “*” and “†.”
05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 167

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 167

half steps grouped, with the other two half steps separated by single whole steps).
The computer scale analysis of complete sections of Density 21.5 in Figure 4.9 also
provides alternative scale possibilities marked by “*” and “†.” Note the 521000 lin-
ear interval vector presence in sections 2 and 3 in Figure 4.9, interesting because
this linear interval vector played such a strong role in the phrase-by-phrase analy-
sis shown in Figure 4.8.
Figure 4.10 presents musically notated versions of the ascending original pitch-
class scales of the three sections of Density as defined in Figure 4.9. Connecting
these pitch classes in Figure 4.10 to their section-dependent pitches—the right-
hand group of pitch-class numbers in the boldface lists in Figure 4.8—and to their
actual pitches in Figure 4.7 provides a sense of how the single overarching 630000-
vectored scale materializes musically. Note how the whole steps here (easier to
count than the half steps, because there are fewer of them) bunch into groups of
two and one in the first and second sections (2–4, 4–6, and 7–9; 4–6, 6–8 and 9–e,
as transpositionally related), while in section 3 the whole steps appear separately
as 2–4, 8–t, and e–1, permutationally relating the first two sections.
Figure 4.11 shows the first section of Density with non-scale tones—according to
the analysis in Figure 4.9—circled (again easier, because there are usually fewer of
these tones than there are scale tones). Note that these non-scale tones often ap-
pear in collections, just as regions do for different keys in tonal music. Most of the
non-scale collections found in Density represent transformational chromaticism in
that the new scales formed in these regions diverge from the current linear interval
vector. Somewhat atypically, mm. 11 and 12 in Figure 4.11 contain only two pitches

a.

& w #w w w #w w w #w w

b.

& w #w w #w w #w #w w w

C.

&#w w w w #w w #w #w w

FIGURE 4.10 Unordered but nontransposed scale pitch classes for each section in
Figure 4.7.
05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 168

168 HIDDEN STRUCTURE

q = 72 3
3 3
& c œ œ #œ ˙. œ #œ. #œ œ ˙ ˙ Œ nœ #œ
œ œ œ œ # œj œ # œ œ œ
F f F p f F
5 3
3 3
j
& œj œ . œ #œ œ ‰ j œ œ bœ
# œ œ- -
˙ œ œ bœ œ bœ. ˙
f p F p subito f
œ ≈ œ bœ œ œ œ œ œ œ œ n˙. œ # œJ œ 3# œj œ . # œ œj # œ .
9
bœ œ ˙
& œ J #œ
ƒ F 3
subito 3
Ï
3
f 3

13 œ bœ. nœ. ˙ œ
œ #œ nœ œ
>
œ œ œ. ‰ œ œ 3œ œ # œ
& œ J ‰ Œ
J # œ # œ>
ƒ> p f p 3p ƒ
w #˙
17 > #˙
œ # œj œ œ œ n œ œ # œ
3
& ˙ ‰ œ ˙
p subito pJ f
21 n˙ œ 3
j #œ #œ œ> ≈ n œ. œ.
& J ‰ œ #œ œ nœ. #œ œ nœ #œ œ Ó
-
ƒ p subito p f p 3
subito

FIGURE 4.11 The first section of Varèse, Density 21.5, with non-scale tones circled.

belonging to the current scale (D-natural and A-natural). This short passage may be
analyzed separately as a 200200 linear-interval-vectored scale, as in [2,3,8,9].
One particularly notable exception to the grouping of non-scale tones in
Figure 4.11 appears in m. 1, on the first note of the piece. One might argue that be-
ginning a work on a non-scale tone tends to discourage hearing a scale in use. I
would argue that this note—which initiates both the first and second phrases—has
a very brief duration and acts more as a propellant to initiate the motives here than
as something that defines an important scale pitch. Ending a phrase or section on a
non-scale tone, on the other hand, poses a significantly different problem, and I
avoid eliminating such pitches from scale analysis whenever possible.
05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 169

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 169

Without entering into a polemic more properly belonging to music perception

and cognition (Avdeev and Ivanov 1993), note that I, at least, hear the three ver-
sions of the 630000-vectored scale in use in Density. My hearing these versions re-
sults from aurally determining the conglomerates of six half steps and three whole
steps, rather than directly perceiving the scales themselves. In short, I hear the
vectoral relationships, rather than the precise versions of the three 630000 scales.
In many ways, this parallels the perception of scales in tonal music, where determi-
nation of actual keys, functions, and even scale degrees is often elusive, though not
for experienced musicians. Discovering the presence of a certain scale informs me
of one more fundamental structural element of post-tonal music and ultimately
leads to a better understanding of this work. For me, the scales underpinning post-
tonal music unify the music beyond the motives present, regardless of one’s ability
to clearly discern these scales.
The scale analysis presented here does not, of course, begin to reveal the other
structural details of this work. However, such analyses are available elsewhere (see
Cope 2000). And as I mentioned earlier in this chapter, scale determination should
be used to enhance, rather than supplant, other types of analysis.

SCHOENBERG’S SIX LITTLE PIANO PIECES, OP. 19,

NO. 1 (1911)
The first of Schoenberg’s Six Little Piano Pieces, op. 19 (1911), shown in Figure 4.12,
provides several opportunities to observe scalar transposition and transforma-
tion at work. Measures 1–4 of this seemingly straightforward nine-measure work
establish—though pitch repetition and melody above harmony—a relatively clear
scale of 521000 linear interval vector. All of the pitches of this passage fit into this
scale with no transient or implied pitches, as seen in Figure 4.13. Measure 5 and the
first half of m. 6 also produce a scale of 521000 linear interval vector, though this
analysis requires excluding E-natural (F-flat) and B-flat appearing on beat 1 of m. 5.
This 521000 linear interval vector scale represents a rotational variation of the scale
in mm. 1–4, because the major seconds and minor third in the scale are in slightly dif-
ferent relationship to one another ([0,1,3,6,7,8,9,e] compared to [0,2,3,6,7,8,9,e]).
The ensuing measures of Schoenberg’s music prove difficult to parse into phrases.
As with Density 21.5, each different phrase determination can cause different scale
interpretations and relationships. Here, for example, analyzing the music from beat
3 of m. 6 through m. 7 produces a 440000 linear interval vector. This phrase inter-
pretation could, in turn, break into two distinct groups or scales, with the former as
the held chord (320010) on the second half of m. 7, and the latter as the whole-tone
transformational scale (060000, m. 7, without its first eighth note, and when includ-
ing the first beat of m. 8 with an implied pitch class 1, which does occur on the sec-
ond half of beat 3 of that measure). However, viewing chords alone and parts of
05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 170

170 HIDDEN STRUCTURE

Langsam n n œœ
&c ∑ Ó Œ #œ #œ. j
nœ nœ bœ
Piano F p espress.

j j j j j j
& c Œ n n œœ. œœ. œœ. ‰ ‰ œœ. Œ œœ œœ œœ ‰ ‰ œœ
. . . .
Œ n n œœ œœ œœ ‰ ‰ œœ
. . .
π
>
ggg b bb œœœ
g R ≈ ‰ ? # # œœ. ‰ Ó
4

& ∑ J & Ó
etwas gedehnt
n n ˙˙
b œ- j b œœ- n œ j j ? # n ˙˙
& œœ œœ. œ œœ ‰ œœ Œ J nœ œœ Œ
.
œœ œœ j
. . b n œœ n # œœ n b œœ # # ˙˙
U
7 gut im Takt
j j j # b bn ˙˙˙˙
& œ n n œœ. Œ ‰ œœ œœ ‰
. .
Œ œœ ‰ Œ
.
œœ œœ Œ
. .
‰ n n œœ ˙˙
œ J
j
π poco rit.

? œœ ‰ Œ n n œœ. ‰ Œ b œœ. ‰ Œ .
‰ b œœ Œ ‰ œœ. Œ
U
Ó
œœ J J J J
π
FIGURE 4.12 Schoenberg, Six Little Piano Pieces, op. 19, no. 1 (1911).

measures as complete phrases seems a bit myopic. Interestingly, the final measure
of this work by itself analyzes to another 521000 linear interval vector scale, thus
wrapping the entire piece—when analyzed as four phrases—in bookends, the re-
sulting 521000–521000–440000–521000 creating a kind of A–A–B–A scale structure.
However, as Figure 4.13 shows, the analysis data from Scale when provided
note-events from Schoenberg’s op. 19, no. 1, demonstrate that the 521000 linear in-
terval vector pervades this work when viewed in three phrases. I have highlighted
the 521000 linear interval vectors and related scales. Note that the next levels
upward—the nine-pitch scales—in both sections 2 and 3 analyze as 711000, con-
trasting the nine-pitch 630000 of Varèse’s Density 21.5. Note here as well that the ac-
tual scales of the three 521000-vectored sections of Schoenberg’s work represent
rotational variations, with all three sections having different arrangements of the
five half steps, two whole steps, and one minor third present in their scales.
05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 171

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 171

Phrase 1; mm. 1-4

[0,2,3,6,7,8,9,t] 521000 [0,2,3,6,7,8,9,t]

[0,2,3,6,7,8,t] 412000 [0,2,3,6,7,8,t]
[0,2,3,7,8,t] 311100 [0,2,3,7,8,t]
[0,3,7,8,t] 202100 [0,3,7,8,t]
[0,3,7,t] 101200 [0,3,7,t]
[0,4,8] 000300 [3,7,t]
[0,4] 000200 [7,t]
[0] 000000 [t]

Phrase 2; mm. 5-6.5

[0,1,3,4,6,7,8,9,e,t] 820000 [0,1,3,4,6,7,8,9,e,t]

[0,1,3,6,7,8,9,e,t] 711000 [0,1,3,6,7,8,9,e,t]
[0,1,3,6,7,8,9,t] 521000 [0,1,3,6,7,8,9,t]
[0,1,3,7,8,9,t] 420100 [0,1,3,7,8,9,t]
[0,1,3,7,9,t] 230100 [0,1,3,7,9,t]
[0,1,3,7,t] 210200 [0,1,3,7,t]
[0,3,7,t] 101200 [0,3,7,t]
[0,7,t] 100110 [0,7,t]
[0,4] 000200 [7,t]
[0] 000000 [7]

Phrase 3; mm. 6.5-9

[0,1,2,3,4,5,6,7,9,e,t] t10000 [0,1,2,3,4,5,6,7,9,e,t]

[0,1,2,3,4,5,6,7,e,t] 901000 [0,1,2,3,4,5,6,7,e,t]
[0,1,2,3,5,6,7,e,t] 711000 [0,1,2,3,5,6,7,e,t]
[0,2,3,5,6,7,e,t] 521000 [0,2,3,5,6,7,e,t]
[0,2,3,5,6,7,t] 420100 [0,2,3,5,6,7,t]
[0,1,3,4,5,9] 311100 [2,3,5,6,7,t]
[0,1,4,5,9] 202100 [2,3,6,7,t]
[0,1,4,9] 102010 [2,3,6,t]
[0,1,9] 101100 [2,3,t]
[0,8] 000200 [3,t]
[0] 000000 [t]

FIGURE 4.13 Computer readouts from a three-phrase analysis of Schoenberg, Six

Little Piano Pieces op. 19, no. 1.
05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 172

172 HIDDEN STRUCTURE

Since most works—tonal as well as post-tonal—contain all twelve pitches, it is

no surprise that Schoenberg’s short work, when viewed as a whole, does as well.
Figure 4.14 presents just such a single-scale analysis of op. 19, no. 1. Interestingly,
the unordered scale here (left-hand scale of the highlighted line) duplicates the
scale of phrase 3 of the phrase-by-phrase analysis shown in Figure 4.13. Also of in-
terest, the nine-pitch scale vector appears as 630000 in Figure 4.14, the same vector
that the scale analysis of Varèse’s Density 21.5 produced. Plotting the actual pitch
names of the right-hand scales proceeding downward toward the 521000-vectored
scale indicates the necessary transient chromatic pitch classes for this analysis.

OTHER MUSICAL EXAMPLES

As an example of implied scale pitches, I present a brief passage from one of my
own works, where I can verify the accuracy of the analysis. The Concerto for Cello
and Orchestra (1994) conforms to a post-tonal style while occasionally including
triadic and otherwise quasi-tonal or modal passages. The cello and instrumental
parts that appear in Figure 4.15 represent typical examples of this kind of synthe-
sis. From my perspective, the music in this example connects related scales over
time, with the phrases clearly delineated by extended-duration cadences. The
analysis that follows reflects my tendency to include implied pitches when I feel it
makes coherent musical sense.
The first phrase in Figure 4.15—m. 79 through the first two beats of m. 82—
presents a straightforward eight-note scale of 610100 linear interval vector, as

[0,1,2,3,4,5,6,7,8,9,t,e] w00000 [0,1,2,3,4,5,6,7,8,9,t,e]

[0,1,2,3,5,6,7,8,9,t,e] t10000 [0,1,2,3,5,6,7,8,9,t,e]
[0,2,3,5,6,7,8,9,t,e] 820000 [0,2,3,5,6,7,8,9,t,e]
[0,2,3,5,6,7,9,t,e] 630000 [0,2,3,5,6,7,9,t,e]
[0,2,3,5,6,7,t,e] 521000 [0,2,3,5,6,7,t,e]
[0,2,3,5,6,7,e] 420100 [0,2,3,5,6,7,e]
[0,2,3,6,7,e] 311100 [0,2,3,6,7,e]
[0,3,6,7,e] 202100 [0,3,6,7,e]
[0,3,7,e] 101200 [0,3,7,e]
[0,4,8] 000300 [3,7,e]
[0,4] 000200 [7,e]
[0] 000000 [e]

FIGURE 4.14 Computer analysis of the entirety of Schoenberg, Six Little Piano
Pieces op. 19, no. 1.
05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 173

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 173

Flute
À & 43
79
∑ ∑ ∑

Clarinet & 43 ∑ ∑ ∑

Viola
à & 43 ∑ ∑ ∑
≥
œ>. œ œ ≤ 7 ≥ œ≤ œ ≤ 7
add vibrato
œ œ 3# œ- œ œ
À & 43 œ œ bœ œ œ ≥ œ J
Violoncello œ? bœ œ œ œœœœ
f
Percussion
à ? 43 ∑ ∑ ∑

Fl.
À&
82
∑ ∑ ∑ ∑

Cl. & ∑ ∑ ∑ ∑

Va.
Ã& ∑ ∑ ∑ ∑
7
≥ ≤ ≥ ≥ ≤ 7≥
À >
œ b œ . œ3 œ b œ . œ # œ œ ≤ œ ≤ bend

? œ œ # œ œ œ # ˙ œ œ3
Vc. ‰® œœ œœ œœ œœ JŒŒ

Perc.
Ã? ∑ ∑ ∑ ∑

FIGURE 4.15 From the author’s Concerto for Cello and Orchestra (1994).
05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 174

174 HIDDEN STRUCTURE

Fl.
À& 86
∑ ∑ ∑ ∑

Cl. & ∑ ∑ ∑ ∑

Va.
Ã& ∑ ∑ ∑ ∑
fingernail pizz. fingernail

œ> œ>
thumb pizz. strum gliss. thumb strum
# œ . 3 œ ~~~ b œ
3
œ œ
Vc.
À? Œ J Œ
œ
3

‰ b b œœ ‰ œ œ
œ
≈ J ‰ ‰ J Œ
#œ œ œ
≈ œœ œœ
f J ß ç œ œ
f
F
Perc.
Ã? ∑ ∑ ∑ ∑

Fl.
À& 90
∑ ∑ ∑

Cl. & ∑ ∑ ∑

Va.
Ã& ∑ ∑ ∑

≥ œ≤
slow strum

#œ œ œ œ
3
fingernail pizz. arco 12

À? Œ œ # œ ‰ œJ ≈ # œ œ œ
vibr.
≤
# œ œ œ #œ
#œ œ n œ # œ œ
≈
5
≥ ≥
Vc.
œ œ
cresc.
ƒ œ œ œ

Perc.
Ã? ∑ ∑ ∑

FIGURE 4.15 continued

05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 175

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 175

Fl.
À&
93
∑ ∑ ∑ ∑

Cl. & ∑ ∑ ∑ ∑

Va.
Ã& ∑ ∑ ∑ ∑

≥ b ˘œ n œ> œ-
3
bœ
À ? ≤ œ œ œ .. Œ
œ œ œ ≈ ‰ #œ . œ
3

œ œ b œ># œ # œ œ œ . œ œ b œ #œ #œ
Vc.
F
3
ƒ
?
Perc.
à ∑ ∑ ∑ ∑

Fl.
À&
97
∑ ∑ ∑ ∑ Ó ‰ œj
π
& ∑ ∑ ∑ ∑ ≈œ œ ˙
π
Cl.

Va.
Ã& ∑ ∑ ∑ ∑ ∑

œ œ.
À œ. œ œ œ œ
3 3 3o o
? J ≈ ‰ J ≈ r j
& œ .. # œ ˙ œ œ œ ..
r
Vc.
œ œ œ œ dim. œ> œ
p π P
p
Perc.
Ã? ∑ ∑ ∑ ∑ ∑

FIGURE 4.15 continued

05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 176

176 HIDDEN STRUCTURE

Fl.
À& ˙.
102

˙. ˙. ˙.
j
œ œ‰ Œ

j3
Cl. &˙ œŒ ∑ ∑ ∑ ∑

π
Va.
à &Ó
œ ˙. ˙. ˙. œ ..
≈Œ

n.v. sensitively
# œ- œ .. œ- œ . b œ- œ
sul tasto

À
3
> 3 >
? œ. œ bœ œ J R J
˙.
Vc. &œœ J b ˙- œ
> p
timpani
? b˙. b˙ b˙. b˙. b˙
Perc.
à ∑
π . b˙. . bœæ bœ Œ Œ

À& Ó
107
3
‰ Œ ∑
˙. ˙. ˙.
Fl.
#œ œ œ
π
3
Cl. & ∑ ≈ j ‰ Œ ∑
œ. ˙ ˙. ˙. œ œ
π
Va.
Ã& ∑ ∑ ∑ ∑ ∑ ∑

>
À ? œ .# œ œ œœ œ œ œ- . -̇
≈J
˙. œ.
J ≈Œ Œ
J œ. œ œ ˙
.
Vc.

f p
3
P 5 P
Perc.
Ã? ∑ Ó œæ
π
˙. ˙
. ˙. ˙
. ˙. ˙.
∑

FIGURE 4.15 continued

05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 177

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 177

shown in Figure 4.16. The cluster of half steps bordered by F-sharp and B-natural
(pitch classes 6–e) provides the distinguishing characteristic of this scale. Follow-
ing this phrase—from the last beat of m. 82 through the first half of m. 85—the mu-
sic transforms to a distinctly different scale of 311100 linear interval vector. However,
the addition of pitch classes 0 and 3 to the topmost possibility for phrase 2’s non-
transposed pitch-class scale, as shown in boldface below the phrase analysis in Fig-
ure 4.16, produces a 610100 linear interval vector, just like that of the first phrase.
The implied pitch class 3 (E-flat) occurs in both the previous and the following—
mm. 86–88—phrase scales. Pitch class 0 (C-natural) appears as the result of a glis-
sando in m. 35. I believe these neighboring and glissando appearances help solidify
the analyzed scale and provide the music here with an overall logic, which was my
reason for using these scales in the first place. The third phrase then reduces to a
linear interval vector of 610100 without any eliminated chromatic-to-the-scale
pitches or added implied pitches.
Measures 89–91 (phrase 4) in Figure 4.15 yield several possible scales. As shown
in Figure 4.16, maintaining the eight-note configuration of the previous three phrases
suggests 440000 as one possibility. The boldfaced 630000 linear interval vector also
makes sense for this passage, with the scales of ten or more pitches approximating
a chromatic scale too closely for consideration. Interestingly, the analysis of
phrase 5, which does not list the linear interval vector of 610100 of the first three
phrases as a possibility, does not show linear interval vectors of 440000 or 630000
either. However, adding pitch classes 2 and 5 as implied pitches to the analysis of
phrase 5 produces a linear interval vector of 630000, a nine-pitch scale, a version of
which also appears in phrase 4. Again, both the preceding and the subsequent
phrases have these pitch classes present, which tends to substantiate these impli-
cations. Otherwise, the scales of phrases 4 and 5 are sufficiently different as to
cause doubt that they relate.
The next two phrases—phrases 6 and 7—both return to the scale vector of the
first three phrases, the 610100 linear interval vector, as shown in Figure 4.16.
Phrase 8, however, contains only five pitches, not enough to generate the eight-
pitch scales in five of the last seven phrases, nor does this phrase contain enough
pitches to produce the 630000 linear interval vector of phrases 4 and 5. However—
and this represents the biggest reach of the scale analyses thus far—by adding
three pitch classes (3, 4, and t) to the analyzed scale of phrase 8, as shown in
Figure 4.16 in boldface, the scale here resolves to a 610100 linear interval vector,
creating an ABA three-section grouping of the eight phrases of this passage
(610100–630000–610100 scale vector class types). Note that the three implied pitch
classes in this phrase appear in the preceding phrase (7).
As mentioned previously, using implied pitches to complete scales may lead
to rather creative analyses, where analysts can seem to arbitrarily fill out scales to
make them compare favorably to whatever point about the music they wish to
make. However, discreet inclusion of implied pitches, where logic and neighboring
scales appear to enhance and encourage their presence, can make as much sense
05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 178

178 HIDDEN STRUCTURE

Phrase 1. mm. 79-first half of 82 [0,9] 002000 [0,9]

[0,1,3,4,5,6,7,8] 610100 [3,4,6,7,8,9,t,e] [0] 000000 [0]
[0,1,3,4,6,7,8] 420100 [3,4,6,7,9,t,e]
[0,1,3,4,6,7] 320010 [3,4,6,7,9,t] [0,2,3,5,6,8,9,t,e] 630000 [0,2,3,5,6,8,9,t,e]
[0,2,3,5,6] 220001 [4,6,7,9,t]
[0,1,3,4] 210100 [6,7,9,t] Phrase 6. mm. 96-first half of m. 99
[0,2,3] 111000 [7,9,t] [0,1,2,3,4,5,6,7,8,9] 901000 [1,2,3,4,5,6,7,8,9,t]
[0,3] 002000 [7,t] [0,1,2,3,4,5,6,7,8] 800100 [1,2,3,4,5,6,7,8,9]
[0] 000000 [7] [0,1,2,3,4,5,6,8] 610100 [1,2,3,4,5,6,7,9]
[0,1,2,3,4,6,8] 420100 [1,2,3,4,5,7,9]^
Phrase 2. last beat of m. 82-first half of 85 [0,1,3,4,6,8] 230100 [1,2,4,5,7,9]
[0,1,3,4,8,9] 311100 [1,2,4,5,9,t] [0,2,3,5,7] 130010 [2,4,5,7,9]
[0,1,4,8,9] 202100 [1,2,5,9,t] [0,2,3,5] 120010 [2,4,5,7]
[0,4,8,9] 101200 [1,5,9,t] [0,1,3] 111000 [4,5,7]
[0,4,9] 001110 [1,5,t] [0,1] 200000 [4,5]
[0,9] 002000 [1,t] [0] 000000 [4]
[0] 000000 [1]
Phrase 7. mm. 101-106
[0,1,2,3,4,5,9,t] 610100 [0,1,2,3,4,5,9,t] [0,1,3,4,5,6,7,8] 610100 [3,4,6,7,8,9,t,e]
[0,3,4,5,6,7,8] 501100 [3,6,7,8,9,t,e]
Phrase 3. second half of m. 85-88 [0,4,5,6,7,8] 400200 [3,7,8,9,t,e]
[0,1,3,4,5,6,7,8] 610100 [3,4,6,7,8,9,t,e] [0,4,5,6,8] 210200 [3,7,8,9,e]
[0,1,3,4,5,6,8] 420100 [3,4,6,7,8,9,e] [0,4,5,8] 101200 [3,7,8,e]
[0,2,3,4,5,7] 320010 [4,6,7,8,9,e] [0,1,4] 101100 [7,8,e]
[0,2,3,4,5] 310010 [4,6,7,8,9] [0,4] 000200 [7,e]
[0,2,3,5] 120010 [4,6,7,9] [0] 000000 [7]
[0,1,3] 111000 [6,7,9]
[0,2] 020000 [7,9] Phrase 8. mm. 107-112
[0] 000000 [7] [0,1,2,5,9] 202100 [0,1,2,5,9]
[0,1,4,8] 101200 [1,2,5,9]
Phrase 4. mm. 89-91 [0,1,4] 101100 [1,2,5]
[0,1,2,3,4,5,6,7,8,9,t,e] w00000 [0,1,2,3,4,5,6,7,8,9,t,e] [0,3] 002000 [2,5]
[0,1,2,3,4,5,6,7,8,9,t] t10000 [0,1,2,3,4,5,6,7,8,9,t] [0] 000000 [2]
[0,1,2,3,4,5,6,7,9,t] 820000 [0,1,2,3,4,5,6,7,9,t]
[0,1,2,3,4,5,7,9,t] 630000 [0,1,2,3,4,5,7,9,t] [0,1,2,3,4,5,9,t] 610100 [0,1,2,3,4,5,9,t]
[0,1,2,3,5,7,9,t] 440000 [0,1,2,3,5,7,9,t]
[0,1,2,4,6,8,9] 331000 [1,2,3,5,7,9,t]
[0,1,2,4,6,8] 230100 [1,2,3,5,7,9]
[0,1,2,4,8] 210200 [1,2,3,5,9]
[0,1,4,8] 101200 [1,2,5,9]
[0,1,4] 101100 [1,2,5]
[0,1] 200000 [1,2]
[0] 000000 [1]

Phrase 5. mm. 92-95

[0,3,6,8,9,t,e] 412000 [0,3,6,8,9,t,e]
[0,3,6,8,9,e] 222000 [0,3,6,8,9,e]
[0,3,6,9,e] 113000 [0,3,6,9,e]
[0,6,9,e] 111001 [0,6,9,e]
[0,9,e] 111000 [0,9,e]

FIGURE 4.16 The author, Concerto for Cello and Orchestra (1994), computer
analysis by phrase
05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 179

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 179

in post-tonal music as it does in tonal music, where implied pitches must occur, or
scales other than major and the three minors would necessarily prevail.
Figure 4.17 presents a computer analysis by section of the concerto excerpt in
Figure 4.15. As shown by the highlighted scales, sections 1, 2, and 4 of this music
have the 222000 linear interval vector in common. Since these sections contain all
twelve pitches, each of these reductions requires the selective elimination of sev-
eral transient chromatic pitches. Section 3 shares only one linear interval vector
with the other sections (111000 in section 2). Of the scale possibilities in section 3,
the 230100 linear interval vector seems most likely, having six pitch classes in its
scale, as do the scale choices for the other three sections. Note that several other
vectors—marked with the special symbols *, ‡, and ø—match in these sections.
However, none of these matches accounts for more than two sections. Allowing ten
pitches per scale produces an 820000 linear interval vector in both the first and
last sections. Again, however, I tend to avoid scales with such large numbers of
pitches because they so closely approximate the chromatic scale. For similar rea-
sons, I have also ruled out the eleven-pitch scale with a linear interval vector of
t10000, also possible in the first and last sections. Interestingly, the vector I have ul-
timately chosen for each section contains a different collection of transient pitches,
ensuring that the actual pitch content of each scale prior to a transposition to 0 is
unique.
Figure 4.18 shows a computer analysis of the entire concerto passage shown in
Figure 4.15, taken as one section of music. Note that the 521000 and the 630000
linear interval vectors are the same as those used in Schoenberg’s Six Little Piano
Pieces op. 19, no. 1, and in Varèse’s Density 21.5, respectively. The 820000 vector
occurs here too, as it did in the analysis of sections 1 and 4 of this passage of the
concerto. Of the three possible scales, I prefer the 630000-vectored scale, even
though it appears only in the section 4 analysis (see Figure 4.17). Each of these
perspectives—phrase, section, and entire passage—provide useful insights into my
music, and I find it helpful to compare the similarities and differences between my
intended scales and the scales that computer analysis provides. In this case, the
sectional analysis most closely approximates my intent for this music.
Every scale has a complementary scale that contains all of those pitches that re-
main when that scale is compared with the full chromatic scale. Although this may
seem obvious, it has some interesting benefits. For example, some scales (e.g., the
whole-tone scale) have a transposed version of themselves as their complements.
Other scales have different versions of themselves as complements (e.g., the octa-
tonic scale, whose minor version is the complement to its major version). Still
other scales have seemingly unrelated scales as their complements (e.g., the major
scale). Complements do not mean much unless—as they occasionally do—the com-
plements appear in the same movement or work together. For example, in my Con-
certo for Cello and Orchestra the scale used in the third movement has a related
complement used much as the dominant key might be used in a tonal concerto.
Figure 4.19 demonstrates this relationship.
05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 180

180 HIDDEN STRUCTURE

Section 1: mm. 79-91

[0,1,2,3,4,5,6,7,8,9,t,e] w00000 [0,1,2,3,4,5,6,7,8,9,t,e]
[0,1,2,3,4,5,6,7,8,9,t] t10000 [1,2,3,4,5,6,7,8,9,t,e] ‡
[0,1,2,3,4,5,6,8,9,t] 820000 [1,2,3,4,5,6,7,9,t,e] ø
[0,1,2,3,4,5,6,8,9] 711000 [1,2,3,4,5,6,7,9,t]
[0,1,3,4,5,6,8,9] 521000 [1,2,4,5,6,7,9,t]
[0,3,4,5,6,8,9] 412000 [1,4,5,6,7,9,t] *
[0,3,5,6,8,9] 222000 [1,4,6,7,9,t]
[0,5,6,8,9] 211010 [1,6,7,9,t]
[0,6,8,9] 111001 [1,7,9,t]
[0,6,8] 010101 [1,7,9]
[0,6] 000002 [1,7]
[0] 000000 [1]

Section 2: mm. 92-95

[0,3,6,8,9,t,e] 412000 [0,3,6,8,9,t,e] *

[0,3,6,8,9,e] 222000 [0,3,6,8,9,e]
[0,3,6,9,e] 113000 [0,3,6,9,e]
[0,6,9,e] 111001 [0,6,9,e]
[0,9,e] 111000 [0,9,e]
[0,9] 002000 [0,9]
[0] 000000 [0]

Section 3: mm. 96-all but last sixteenth-note of 99

[0,1,2,3,4,5,6,7,8,9] 901000 [1,2,3,4,5,6,7,8,9,t]
[0,1,2,3,4,5,6,7,8] 800100 [1,2,3,4,5,6,7,8,9]
[0,1,2,3,4,5,6,8] 610100 [1,2,3,4,5,6,7,9]
[0,1,2,3,4,6,8] 420100 [1,2,3,4,5,7,9]
[0,1,3,4,6,8] 230100 [1,2,4,5,7,9]
[0,2,3,5,7] 130010 [2,4,5,7,9]
[0,2,3,5] 120010 [2,4,5,7]
[0,1,3] 111000 [4,5,7]
[0,1] 200000 [4,5]
[0] 000000 [4]

Section 4: mm. 101-112

[0,1,2,3,4,5,6,7,8,9,t,e] w00000 [0,1,2,3,4,5,6,7,8,9,t,e]

[0,1,2,3,5,6,7,8,9,t,e] t10000 [0,1,2,3,5,6,7,8,9,t,e] ‡
[0,1,2,3,5,7,8,9,t,e] 820000 [0,1,2,3,5,7,8,9,t,e] ø
[0,1,2,3,5,7,8,9,e] 630000 [0,1,2,3,5,7,8,9,e]
[0,1,2,4,6,7,8,t] 440000 [1,2,3,5,7,8,9,e]
[0,1,4,6,7,8,t] 331000 [1,2,5,7,8,9,e]
[0,3,5,6,7,9] 222000 [2,5,7,8,9,e] *
[0,5,6,7,9] 211010 [2,7,8,9,e]
[0,1,2,4] 210100 [7,8,9,e]
[0,2,4] 020100 [7,9,e]
[0,2] 020000 [7,9]
[0] 000000 [7]

FIGURE 4.17 Computer analysis of the excerpt from the author’s concerto in Fig-
ure 4.15, by section.
05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 181

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 181

[0,1,2,3,4,5,6,7,8,9,t,e] w00000 [0,1,2,3,4,5,6,7,8,9,t,e]

[0,1,2,3,4,5,7,8,9,t,e] t10000 [0,1,2,3,4,5,7,8,9,t,e]
[0,1,2,3,5,7,8,9,t,e] 820000 [0,1,2,3,5,7,8,9,t,e]
[0,1,2,3,5,7,8,9,e] 630000 [0,1,2,3,5,7,8,9,e]
[0,1,2,5,7,8,9,e] 521000 [0,1,2,5,7,8,9,e]
[0,1,4,6,7,8,t] 331000 [1,2,5,7,8,9,e]
[0,1,4,6,8,t] 141000 [1,2,5,7,9,e]
[0,4,6,8,t] 040100 [1,5,7,9,e]
[0,6,8,t] 030001 [1,7,9,e]
[0,6,8] 010101 [1,7,9]
[0,2] 020000 [7,9]
[0] 000000 [7]

FIGURE 4.18 Computer analysis of the entirety of the excerpt from the author’s
concerto shown in Figure 4.15.

œ œ œ bœ bœ bœ œ œ bœ bœ
? c œ. J œœ ≈ œ . œ # 3
œ 6 J J
bœ œ. #˙ œ œ œ #œ J bœ 8 bœ.

FIGURE 4.19 Selected measures of the solo cello part of the author’s concerto.
The scale used in the third movement has a related complement
used much as the dominant key might be used in a tonal concerto.

The beginning passage from the third movement of Stravinsky’s Three Pieces for
String Quartet (1914) contains all twelve tones. However, a scale analysis (see
Figure 4.20) by the Scale program on the CD-ROM of this book indicates that a
more likely explanation would be the major/minor set [0,2,3,4,5,7,8,9,t,e] with the
linear interval vector of 820000. This set contains both versions of the mediant,
submediant, and leading-tone/subtonic pitches, as one would expect of such a
scale combination. Note that the output shown in Figure 4.20 uses Lisp syntax
(parentheses, no commas, and so on) rather than the scale/set syntax described in
previous examples. I have printed the output exactly so that program users will not
be surprised by the differences. Conversion to scale/set syntax should be obvious
upon comparing this example and, say, Figure 4.18.
05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 182

182 HIDDEN STRUCTURE

((0 1 2 3 4 5 6 7 8 9 10 11) V1200000 (0 1 2 3 4 5 6 7 8 9 10 11))

((0 1 2 3 4 5 7 8 9 10 11) V1010000 (0 1 2 3 4 5 7 8 9 10 11))
((0 2 3 4 5 7 8 9 10 11) V820000 (0 2 3 4 5 7 8 9 10 11))
((0 2 3 4 5 8 9 10 11) V711000 (0 2 3 4 5 8 9 10 11))
((0 2 3 4 5 8 9 10) V521000 (0 2 3 4 5 8 9 10))
((0 2 3 4 5 8 9) V412000 (0 2 3 4 5 8 9))
((0 2 3 4 5 9) V311100 (0 2 3 4 5 9))
((0 2 3 5 9) V121100 (0 2 3 5 9))
((0 2 3 9) V111001 (0 2 3 9))
((0 2 9) V011010 (0 2 9))
((0 9) V002000 (0 9))
((0) V000000 (0))

FIGURE 4.20 A scale analysis of Stravinsky, Three Pieces for String Quartet (1914),
3rd movement (opening), as shown in Figure 1.19.

PROGRAM DESCRIPTION
The code used to derive linear interval vectors from scales expressed as pitch
classes can be coded quite succinctly in Lisp. As shown below, the function
count-intervals uses the Lisp primitive count to produce a straightforward
listing of the intervals from 1 to 6 provided by its incrementally ascending optional
argument. Note that the argument of count-intervals is a series of intervals
between the pitch classes of the scale and not the scale itself (the function get-
intervals-from-pitches defined in Chapter 2 will output the intervals neces-
sary for count-intervals here).
(defun count-intervals (intervals &optional (size 1))
“Counts the interval types present.”
(if (> size 6)()
(cons (count size intervals)
(count-intervals intervals (+ size 1)))))
The output of a run of count-intervals on an octatonic scale in pitch classes
will provide a simple example of how the function works. As explained in previous
chapters, combining simple functions such as this can provide powerful ways to
analyze music. This is shown in the full program found on the CD-ROM accompany-
ing this book.
The following example—as well as many of those yet to come in this book—
represents a kind of Lisp pseudo-code, where the code follows all of the proper pro-
tocols but consists of yet-to-be-written functions. Hopefully, these functions have
05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 183

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 183

names that describe their actions well enough that readers who have followed the
Lisp processes thus far described can create them themselves. Using pseudo-code
such as this allows me to describe top-level functions of some sophistication with-
out having to duplicate the entire code presented in the CD-ROM that accompanies
this book.
(defun produce-scale (events)
(order-from-zero
(remove-all-duplicates
(convert-to-pitch-classes
(collect-pitches events)))))
This function produces a pitch-class scale given any set of note-events for a work
or excerpt of a work of music once all four of the functions it calls for—order-
from-zero, remove-all-duplicates, convert-to-pitch-classes, and
collect-pitches—have been successfully coded. Since readability is extremely
important in coding, programmers will often write code that duplicates existing
primitives in Common Lisp to ensure that its name reflects exactly what it does. In
the above function, remove-all-duplicates represents just such a function:
its results are the same as those of remove-duplicates, but without the latter’s
sometimes confusing optional arguments.
The software program Scale (Macintosh and PC compatible) on the CD-ROM that
accompanies this book offers readers the ability to access all 2,048 possible equal-
tempered scale sets in a variety of ways—identifying, collecting, and/or perusing
these scales by lists of pitch classes, linear interval vectors, or a combination of
the two. Also, one may search the database in various ways. For example, the pro-
gram allows searches for all scales having the initial pitch classes of 0, 1, or 2, or all
scale sets with one or more minor seconds, or scale sets that contain one minor
third, and so on. These processes hopefully provide useful advantages to both ana-
lysts and composers. The program also analyzes collected note-events in the man-
ner previously described. I have also included all of the music in the figures in this
chapter as note-events, both as examples for readers to follow and for verifying
my findings. The manual that accompanies Scale provides detailed instructions for
operating the program.
A number of functions in the Scale program, particularly organize-by-
duration and rotate-scale, use a function called mapcar. This mapping func-
tion takes two arguments. The first argument must be a function preceded by the
special indicator # and a single quotation mark. The second argument must be a
list on each element of which the first argument will operate. For example, the code
(mapcar #’1+ ‘(1 2 3 4)) will map the function 1+ over the list provided to
produce (2 3 4 5) as output. When programmers in Lisp wish to map a function
they cannot imagine ever using again, they will often use mapcar to map what is
termed a “no-name” function. No-name functions are indicated by using lambda
which allows functions to be created on the fly. No-name functions do not clutter
05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 184

184 HIDDEN STRUCTURE

programming workspace with function names that have one use only. Another way
to program the above code, then, is (mapcar #’(lambda (x)(+ x 1)) ‘(1 2
3 4)) where “x” is a local variable name.
As mentioned earlier in this chapter, scales represent special kinds of pitch-class
sets. Like major and minor scales in tonal music, post-tonal scale sets represent en-
tire passages or complete works and do not account for the appearance of every
note. Scale sets are nonetheless pitch-class sets and, as such, suffer from the limita-
tions described in Chapter 3—specifically, they lack information about register.
Register can play an important role in scale determination, just as it does in ex-
panded definition of pitch-class sets. For example, many composers use what are
called pitch fields in their compositions. Pitch fields have register-rigid pitches that
suggest scales of more than one octave to more clearly portray their sonic relation-
ships. Figure 4.21 presents just such a pitch field. It covers a two-octave span and

Flute & ∑

#œ
& nœ #œ #œ nœ #œ #œ nœ nœ # œ# œ n œ nœ #œ #œ #œ

Piano
π
& #œ nœ #œ nœ nœ #œ #œ #œ nœ #œ #œ nœ #œ
nœ nœ nœ
una corda

Fl. & n˙. n˙ n˙ Œ

p
#œ #œ #œ nœ #œ #œ #œ
& nœ nœ #œ #œ #œ #œ

Pn.

& # œ# œ n œ # œ n œ n œ #œ # œ n œ # œ n œ> n œ nœ nœ
#œ nœ #œ >n œ # œ
f π
>
fπ fπ
FIGURE 4.21 A pitch field that covers a span of two-plus octaves and contains two
exclusionary hexachord pitch-class sets.
05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 185

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 185

contains two exclusionary pitch-class sets. Because all of the notes in this brief
passage have inexact durations, a scale drawn from this music would, according to
the principles thus far discussed, present a dodecachord—a panchromatic scale
consisting of all twelve tones of the octave. Such a chromatic representation would
reveal very little about the music they represent. A more logical scale for this pas-
sage appears in Figure 4.22, with its two-octave range.
Although pitch fields do not occur in all, or perhaps even much, music, pitches
relegated to particular registers (see Chapter 3) for periods of time occur in signifi-
cant numbers of musical works. Figure 4.23 presents just such an example, where the
pitches C-sharp, G, and B appear only in the upper register and where the pitches
E-flat, F, and A appear only in the lower register. Thus, this passage would lose true
scale definition if we reduced all the pitches to pitch classes. Figure 4.24 demon-
strates the scale differences between a one-octave representation and a two-octave
representation.

Fl. & n˙. n˙

p
#œ #œ #œ#œ #œ #œ #œ nœ #œ #œ
nœ nœ#œ #œ nœ nœ nœ
&
Pn.

& #œ nœ #œ #œ n œ #œ #œ n œ n œ>#πœ
nœ nœ#œ
nœ
nœ
>π
f f
Fl. & n˙

#œ
& nœ #œ #œ #œ nœ nœ nœ

Pn.

& nœ #œ #œ #œ nœ #œ
> nœ nœ
f π
FIGURE 4.21 continued
05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 186

186 HIDDEN STRUCTURE

#w #w #w
w w w w
&w #w w #w w

FIGURE 4.22 A scale with a range of two-plus octaves for the example shown in
Figure 4.21.

&c Œ j j j œ. j
Voice œ. œ # œj œj œj œj œj . œr œj # œj œ œ œ
Hain in die– sen Pa– ra– die– sen wech–seit ab mit Blü– ten–

#w
? c ww
www œœœ # œœœ

Piano

?c j
w ˙ .. bœ w
w ˙ .. bœ w

FIGURE 4.23 An example where the notes C-sharp, G, and B appear only in the up-
per register and the notes E-flat, F, and A appear only in the lower
register (Arnold Schoenberg, Das Buch der hängenden Gärten, II).

& #w w bw w w w w w #w # w w w w w #w w
w
w bw w w

FIGURE 4.24 The scale differences between a one-octave representation and a

two-octave representation of the music in Figure 4.23.

Determining whether a passage requires a multiple-octave or a single-octave

analysis requires a simple program that counts pitches in various registers and in-
forms users when more than a certain threshold of pitches unique to a given oc-
tave has been reached, suggesting that one or more further octaves be added to
analyses to reveal the registral importance of these pitches. Given the somewhat
05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 187

COMPUTER ANALYSIS OF SCALES IN POST-TONAL MUSIC 187

unusual nature of two-octave scales, I have not provided a program here to analyze
for them, leaving it to readers to determine whether and when music would require
such extended scales.
Scales that vary in their upward and downward motion should also be consid-
ered an important contributor to some music. The melodic minor tonal scale is a
good example of such a scale, with its raised sixth and seventh degrees in the as-
cent and natural—according to the operative key signature—forms of these scale
degrees in the descent. To analyze for such scales, a computer program would re-
quire code to differentiate between pitches occurring in only rising or falling mo-
tion in music and catalog their presence in the resulting upward and downward
scales. In effect, two scales will result from such analysis.
Scales need not be limited to pitch. After all, scales result from the cumulative
membership of any given parameter in music—duration, articulation, dynamics,
and so on. For example, durational scales can reveal interesting if not significant
detail about a work or a passage. A durational scale that excludes or has relatively
infrequent occurrence of certain durations can explain significant analytical as-
pects about the music it represents. Music that creates a durational scale of  ,  ,
and  , for example, reveals a great deal about its source. Lacking dotted rhythms,
quarter notes, triplets, and so on, this durational scale indicates the music’s more
or less squareness and probable jerky quality due to shifts between durations of
quite differing values. Of course, without any reference to the quantity of each of
these durations, we are left with the possibility that the passage consists almost
entirely of eighth notes, with a half note and whole note providing rhythmic ca-
dences. Therefore, adding proportions—providing a sense of “chromaticism” to the
“diatonicism” of a durational scale—can indicate subtle nuances in it, as in

r r j j w
œ œ. œ œ. œ œ. ˙
22 / 29 / 48 / 1 / 6 / 14 / 12 / 1

which provides a great deal more information about the overall durational context
of the music. The whole note represents but 1/133 of the overall number of dura-
tions, while the eighth note holds a 48/133 advantage. Both the whole note and the
eighth note hold such marginal roles in this music that one could catalog them as
nondiatonic to the durational scale in use, even though the whole note clearly has
a substantial durational weight lasting eight times the duration of the eighth note
and most likely occurring as the ultimate note of the music. Articulations, dy-
namics, timbre, and so on can be extracted from music in much the same way as
durations. Aligning various parametrical scales could then indicate consonant or
dissonant relationships, much as the dynamic algorithmic information theory
graphs do as discussed in Chapter 2 of this book.
05_DAS_Chap4_pp145-188 1/29/09 1:35 PM Page 188

188 HIDDEN STRUCTURE

CONCLUSIONS
In the four or so centuries of its predominance in Western music theory, tonal
analysis has presented two particularly important doctrines. The first is that tonal
music relies on scales—typically major and minor scales—to provide continuity
and consistency. The second is that these scales produce hierarchy in the sense of
scale degree and musical function. If post-tonal music too can be analyzed accord-
ing to scales, then it seems logical that post-tonal music also has hierarchy based
on scale degree and function. In the next chapter I describe one of the possible
methods for discovering such hierarchy in post-tonal music based on the scales re-
vealed in the present chapter.
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 189

FIVE
Function and Structure
in Post-Tonal Music

The whole history of music, from, say, Berlioz to Boulez can be understood as
one in which nonfunctional class membership becomes increasingly important,
ending in serialism, set theory, and the compositions arising from them. Thus, in
the analysis of much contemporary music, the pitch class of a tone is more im-
portant than its function.

Leonard Meyer, The Spheres of Music (2000), 267

This chapter begins by introducing object-oriented programming in Common Lisp,

an extension of the more traditional form of Lisp programming covered thus far in
this book. This discussion is immediately followed by a proposal of the possibly
provocative notion that post-tonal music has function just as tonal music does,
and that these post-tonal functions can be analyzed and represented in logical
ways. This concept of post-tonal function emerges from the scale analysis de-
scribed in Chapter 4. Because the pitches contained in scales do not have the same
importance, or the same implied destinations, defining their various roles directly
relates to musical function. Although certainly not new, the idea of function in post-
tonal music continues to cause controversy, and even among those who agree that
function exists there is little consensus as to what form such functions take. In
turn, post-tonal functions portend structural hierarchy, a subject that itself pro-
vokes a degree of controversy. The lack of agreement on these subjects presents
complex problems for music analysts, problems that computational technology
may be well suited to engage and resolve. This chapter, then, concerns Principle 3
as described in Chapter 1.

OBJECT-ORIENTED PROGRAMMING
Thus far in this book, I have described Common Lisp as a functional language. The
word functional here relates to programming and does not carry the same meaning
as it does in music. As we have seen since Chapter 2’s initial presentation, func-
tions in programming languages operate by command-line input—typing function

189
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 190

190 HIDDEN STRUCTURE

names and their arguments into Listener windows, resulting in some sort of desired
output. For example, the code
? (first '(1 2 3))
represents a functional operation, with the function first returning the first ele-
ment of its argument. Functional programming such as this provides an excellent
environment for prototyping and testing ideas. However, most applications we
use today have a different look—they have menus, windows, buttons, scroll bars,
and so on. We can enter and manipulate data by using real-world parallels such as
mouse-controlled checkboxes, sliders, and pop-up menus, rather than typing on
keyboards. Creating mouse-clickable visual interfaces requires a different type of
programming called Object Oriented Programming (OOP). Common Lisp’s version
of OOP is called CLOS, for Common Lisp Object System. Most versions of Common
Lisp include CLOS. Using a mouse rather than a keyboard is one very simple way
to distinguish between object-oriented programming techniques and functional
programming techniques.
Building interfaces that require integrated menus, windows, and so on falls well
outside the scope of this book. However, knowing how OOP works in Common Lisp
will prove valuable for the ensuing discussions of musical function, hierarchy, and
other concepts presented later in this book that require a broader view of program-
ming than that provided in Chapter 2. OOP offers a rich vocabulary of processes
better matched to these more complex formal and structural musical ideas.
As an example, imagine that a user wishes to analyze the pitches of a group of
note-events representing beat 129 of a long work containing, say, one thousand
such note-events. Locating the note-events of beat 129 in the long list of note-
events, collecting this beat’s note-events from the longer list of note-events, extract-
ing the pitches from these collected note-events, and so on would require the use
of a series of functions each and every time this information was needed. However,
if this beat and all other beats were initially collected by the series of functions just
described and then stored in separate objects named by their beat number—such
as beat-129—locating individual beats would be simple. OOP offers perfect ways to
do this.
A study of object-oriented programming, even a brief study such as this, usually
begins with an example of defclass, such as:

(defclass beat ()
((note-events :initarg :note-events
:initform nil
:accessor note-events)
(pitches :initarg :pitches
:initform nil
:accessor pitches)
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 191

FUNCTION AND STRUCTURE IN POST-TONAL MUSIC 191

(beat-number :initarg :beat-number

:initform 1
:accessor beat-number)))
Classes—such as this one called beat—are not functions, but rather abstract dec-
larations of how this class’s instances will behave. As we shall shortly see, classes
and instances, although closely related, have different purposes. Classes describe
how their instances will act. Instances represent individual objects that follow their
class descriptions. Classes resemble recipes, while instances constitute the actual
food produced from such recipes.
The above class named beat has three types of slots (also called arguments)—
note-events, pitches, and beat-number. Each of these slots has several key-
words that follow its declaration. Keywords allow arguments to be called in any
number and any order as long as the actual argument is preceded by a keyword
prefixed with a colon. In the case of note-events, for example, the slot has an initial
argument (:initarg, itself described by the keyword note-events), an initial
form (:initform with its default value of nil), and an accessor (:accessor with its
default value of note-events), the last of which can be used just like a function to
access the data in instances of the associated class without being defined further. I
will describe these slots in more detail momentarily.
In order to place a musical work into OOP-class beats requires that each beat of
that work be placed a separate instance of the beat class. To use the class beat,
then, we have to create an instance of it containing the data for a particular beat.
To access the data in each of the slots, we call their accessors as if they were func-
tions. For example, the following succession of code represents the creation of one
instance of the just-described beat class.
? (setq beat-129
(make-instance 'beat :note-events
'((129000 50 1000 1 127))
:pitches '(60)
:beat-number 129))
#<BEAT #xF1C676>
? (note-events beat-129)
((129000 50 1000 1 127))
? (pitches beat-129)
(60)
?(beat-number beat-129)
129
Here, make-instance creates an instance of the class beat with all of its initial
slots established. Note that the slots presented are each referenced by their colon-
prefixed slot names. This type of slot calling allows for the arguments to appear in
any order; slots that are not called revert to their default values, presented in the
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 192

192 HIDDEN STRUCTURE

class definition itself. Note how the accessors look and act like functions, returning
information here from the object beat-129.
Each new instance of an object of the class beat can have different information
in its slots. The information stored in the slots of these instances can also be easily
changed. To do this, programmers typically use setf, which takes two arguments
—a location and the new data. The following code, for example, alters the informa-
tion located in the beat-number slot of beat-129:
? (setf (beat-number beat-129) 128)
128
? (beat-number beat-129)
128
The first argument to setf here defines the location as the beat-number slot of
beat-129. The second argument to setf provides its new value. This can be read
as “set the beat number of beat-129 to 128.” We now have a class of objects de-
fined, instances of these objects each containing information in appropriate slots,
and a process enabling us to alter any of this information whenever we wish. Be-
cause objects now exist as independent entities, we can find them simply by refer-
encing their names, as in
? beat-129
#<BEAT #xF5432E>
where the returned information indicates that the object belongs to the beat class,
and a reference number represents an internal symbol to distinguish objects from
one another.
OOP also has class-related methods to differentiate them from standard Common
Lisp functions. Methods are defined using defmethod. As seen below in pseudo-
code (i.e., play-events has not yet been defined), defmethod must reference the
object class to which it belongs so that when called it can expect the proper slot
accessor information.
(defmethod play ((beat-class beat))
(play-events (note-events beat-class)))
The above method definition play takes a list of arguments, each of which consists
of a class name (to the right—beat) and a nickname for that class name (to the
left—beat-class) for use within the method definition. Classes may have as many
methods as needed. The method play is then used much like a function or acces-
sor, as in
(play beat-129)
with the note-events of beat-129 played as a result.
Methods, defined separately for each class, have many advantages over func-
tions. For example, in functional programming a function has but one form, de-
scribed by its definition. The function first, for example, returns the first element
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 193

FUNCTION AND STRUCTURE IN POST-TONAL MUSIC 193

of its list argument in all circumstances. Provide this function with a non-listed
number, for example, and it will return an error. OOP methods offer programmers
the opportunity to use polymorphic processes. The word polymorphic here means
that methods with the same names may have many different definitions, each defi-
nition associated with a particular object class. Although initially confusing, espe-
cially for functional programmers, polymorphism has many advantages. For
example, two accessors both called play but bound to two different classes may
take very different kinds of information as arguments, depending on the associated
class definition. One method play could play note-events in a list as argument,
while another method play could play individual note-events not in a list as argu-
ment. To users of such a system, the accessor play serves both needs transpar-
ently, as long as it is provided instances of the proper class. If we did not have OOP
polymorphism, we would have to create several functions such as play-events,
play-event, and so on, for such computations. These extra names complicate
and clutter the programming environment.
Thus far, we have discussed objects only as representing instances of individual
classes. OOP programming includes another feature, one that most closely resem-
bles one of the subjects of this chapter: hierarchy. In OOP, classes may inherit from
other classes, and these other classes may in turn inherit from still other classes,
and so on. These variously related classes inherit the slots and values of the
classes to which they belong. As example, the following definition of the beat class
described earlier in this chapter
(defclass beat (measure)
((note-events :initarg :note-events
:initform nil
:accessor note-events)
(pitches :initarg :pitches
:initform nil
:accessor pitches)
(beat-number :initarg :beat-number
:initform nil
:accessor beat-number)))
now includes the name of another class called measure in the list following its
name. The definition of measure given below includes different slots than the class
beat, but by definition, the class beat will inherit the slots of the class measure.
(defclass measure ()
((beats :initarg :beats
:initform nil
:accessor beats)
(meter :initarg :meter
:initform 4
:accessor meter)
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 194

194 HIDDEN STRUCTURE

(number :initarg :number

:initform nil
:accessor number)))

In this case, the class measure is called a superclass, and the class beat is called
its subclass. Therefore, subclasses inherit the slots and their values from their su-
perclasses. Thus, creating instances of measure and beat produces the following
output:

? (setq measure-50 (make-instance 'measure))

#<MEASURE #xF4D02E>
? (setq beat-129
(make-instance 'beat :note-events '((129000 50 1000 1
127))
:pitches '(60)
:beat-number 129))
#<BEAT #xF4D236>
? (meter beat-129)
4
? (meter measure-50)
4
? (beat-number measure-50)
> Error: No applicable method for args

Note that the beat class, defined as a subclass of the measure object, inherits all
of the slots and their values of the measure class, but that the reverse is not true.
In other words, the hierarchy moves in one direction only.
The following code creates a phrase object that itself contains a list of beat ob-
jects for each beat in its phrase of music. The code presented below, although not
self-contained, appears in full in the file called “Root Object Code” in the folder
“Root” on the CD-ROM accompanying this book.

(defclass phrase ()
((beats :initarg :beats
:initform ()
:accessor beats)))
(defmethod initialize-instance ((object phrase) &rest initargs)
(apply #'shared-initialize object t initargs)
(setf b-number 0)
(setf (beats object) ())
(loop for beat in *beat-event-lists*
do (setq *note-events* beat)
do (setq beat-number (create-beat-number object))
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 195

FUNCTION AND STRUCTURE IN POST-TONAL MUSIC 195

do (set beat-number (make-instance 'beat))

do (setf (beats object)(push beat-number (beats
object))))
(setf (beats object)(reverse (beats object))))
(defclass beat (phrase)
((note-events :initarg :note-events
:initform ()
:accessor note-events)
(pitches :initarg :pitches
:initform ()
:accessor pitches)
(root :initarg :root
:initform ()
:accessor root)))
(defmethod initialize-instance ((object beat) &rest initargs)
(apply #'shared-initialize object t initargs)
(setf (note-events object) *note-events*)
(setf (pitches object) (get-pitches object (events object)))
(setf (root object) (get-root object (pitches object))))
(defmethod get-pitches ((object beat) note-events)
(loop for note-event in note-events
collect (second note-event)))
(defmethod create-beat-number ((object phrase))
(incf b-number)
(implode (list 'beat- b-number)))
(defmethod get-root ((object beat) ordered-pitches)
(get-the-pitch (arrange ordered-pitches)
(find-root
(mapcar #'get-intervals (arrange ordered-
pitches)))))
The class called phrase here has but one slot, a list of beat objects initially set to
the empty list (). These beat objects are created in the fifth through ninth lines of
the method initialize-instance of the beat object, within the loop call
(see Chapter 4 for a discussion of loop). The beat object has three slots, one for
the note-events of a beat, one for the extracted pitches of those beat note-events,
and one for the root pitch of the beat. The initialize-instance method for
the beat object assigns the proper values for the beat object’s slots. The code
(apply #'shared-initialize object t initargs) in the two initialize-
instance methods fills the slots of the class with its default values. The call to
mapcar in the get-root method maps the function get-intervals across every
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 196

196 HIDDEN STRUCTURE

sublist of its second argument. The function incf—the increment function—in

the create-beat-number method incrementally increases the current value of
b-number. The function incf is “destructive” in that it actually alters the state of
its b-number argument. Note that all of the methods defined here are bound to
their appropriate class.
To use this object-oriented version of the root identification program requires a
single call to make-instance, such as
? (setq phrase-object (make-instance 'phrase))
#<PHRASE #x11997DE>
? (beats phrase-object)
(BEAT-1 BEAT-2 BEAT-3 BEAT-4 BEAT-5 BEAT-6)
? (root beat-4)
65
where creating the phrase object names these objects, invests all of the slots of
the beat objects with appropriate data, and then lists the beat objects by name in
numerical beat order in its beats slot. A call to any of the three accessors of any
beat object returns the appropriate analysis of that particular beat. Looking at the
file “Root Object Code” reveals a mix of OOP and functional code similar to that in
the “Root” file. As an alternative, one could redefine these functions as methods to
produce the same results, thus keeping the entire process within OOP style. Many
programming languages (Java, for instance) pride themselves on being OOP exclu-
sive, while others (C, for instance) do not include object procedures (C++ does,
however). Common Lisp, on the other hand, allows for what many feel represents a
healthy mix of functional programming and OOP techniques.
Storing musical works in hierarchies such as the one just described can be very
useful. Interestingly, deciding whether the top level consists of phrase objects and
the bottom level consists of beat objects, or vice versa, poses a challenge for mu-
sic analysts. Clearly musical structure follows a pyramidal form, with notes at the
bottom, followed—possibly—by beats, measures, phrases, sections, movements,
and then works. However, one might not need to give beats access to information
about the works to which they belong, whereas works may very well need to have
access to the information that notes possess. Certainly, works require access to the
notes that belong to them. For composers, especially those who improvise their
music, notes may logically reside at the top level of the class structure. In Experi-
ments in Musical Intelligence (see Cope 1996 and 2005), however, object hierarchy
begins with a work superclass and ends with a beat sub-sub-subclass. This process
results from the program’s need for notes, beats, and so on, to always have access
to their location in their associated work, which is accomplished only by an ap-
proach where inheritance follows a downward flow from the work to the note. Ana-
lysts must obviously decide such issues of hierarchy in class structure prior to
analysis.
Interestingly, programming technique itself also tends to follow top-down or
bottom-up paradigms. For example, professional programmers tend to work from
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 197

FUNCTION AND STRUCTURE IN POST-TONAL MUSIC 197

the top down, creating top-level functions first and then fleshing out the details
while always maintaining an overview of the program they have in mind. Unfortu-
nately, when using top-down approaches one cannot actually test a program until
all of its parts are coded. Therefore, novice programmers, who need to feel confi-
dent that lower level functions will work before they use them, tend to program us-
ing bottom-up processes. The problem with bottom-up programming is that
although each successive function can be debugged until it actually works, one
may lose sight of the ultimate goal or, worse yet, digress into solving irrelevant
problems that have little to do with the overall program. Like analysts, program-
mers must decide which approach suits a particular project and never attempt to
work from both directions at once in a vain hope that the two will somehow magi-
cally connect in the middle.
I typically choose top-down programming as a model. Therefore, as previously
mentioned, readers should be aware that although the code on the CD-ROM accom-
panying this book works as described, the code shown in the latter part of this
book will not operate as it stands. This code represents top-level functions and ob-
ject classes that contain code not shown. Unfortunately, it may seem counterintu-
itive to some to analyze music from a top-down perspective. For example, analyzing
form without knowing the actual musical ideas present does not bode well for suc-
cess. Therefore, many analysts may prefer using a bottom-up approach, beginning
with musical functions that inform concepts of phrase and ending with sections
informing form.
My reasons for presenting OOP at this point in Hidden Structure have to do with
the nature of the material about to be discussed: musical function and hierarchy.
Deciding how to represent the analyses of function and hierarchy in post-tonal mu-
sic will greatly affect the understanding of these two important facets of music. On
a practical level, it seems highly advantageous to hide such elements as function
names, hierarchical representations, and so on in objects rather than nesting them
in lists along with note-events. Even more important, however, creating hierarchical
object classes that inherit from one another allows storage of important structural
information that allows access to every level. In fact, it requires that an analyst ana-
lyze structurally. As an example, Figure 5.1 presents one possible overview of the
structure of a work of music that simultaneously represents visual and musical par-
allels and an OOP hierarchy. Seen in this way, OOP processes not only serve as
analogies for musical structure, they actually serve as vehicles for explaining that
musical structure. As we shall soon see, this approach has many computational
advantages as well.

DEFINITIONS
Musical set theory—see Chapter 3 for more information—offers analysts opportu-
nities to analyze post-tonal music by grouping it into sets and then reducing these
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 198

198 HIDDEN STRUCTURE

FIGURE 5.1 One possible overview of the structure of a work of music that is at
one and the same time a visual/musical representation and an OOP
representation.

sets into various categories. As important as this reduction-and-categorization

process is, however, arguing that sets represent an alternative to tonal function for
post-tonal music poses many difficulties. In fact, one could rationally posit that set
theory does little beyond grouping post-tonal music into chord types similar to the
various triadic form types of tonal music (e.g., major and minor triads).
I therefore do not agree with Leonard Meyer’s implication, presented at the out-
set of this chapter, the implication that pitch-class membership can act as a substi-
tute for function (“the pitch class of a tone is more important than its function”;
Meyer 2000, 267). An equivalent view for tonal music would be that a major triad
can substitute for a tonic function, something we know to be false because many
different tonal functions consist of major triads. Interestingly, both tonal scales and
functions represent pitch-class sets, but we do not typically use pitch-class sets to
analyze tonal music, because doing so reveals little of the music other than its tert-
ian origins.
Defining the word “function” as it applies to music here seems necessary, espe-
cially before attempting to use the term in reference to post-tonal music. Webster’s
Collegiate Dictionary (1991) defines the word function as “the purpose for which
something is designed or exists.” Because I find very little if anything without pur-
pose, every note, certainly every chord and pitch-class set in music—tonal, post-
tonal, or otherwise—has purpose and function, and attempting to discover these
purposes is not only appropriate but paramount to the analysis of post-tonal music.
Therefore, for this book at least, the concept of musical function derives from
the notion of hierarchy that itself derives from scale membership, and chord roots
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 199

FUNCTION AND STRUCTURE IN POST-TONAL MUSIC 199

in cases where two or more pitches occur simultaneously. All pitches in

music, any music, have “class membership,” as Meyer terms it, which does not nec-
essarily change simply because musical styles themselves change. That said, func-
tion exists in post-tonal music. The only question that remains is how to discern
function and determine its purpose.
Musical function represents one of the most salient features of any kind or style
of music. Musical function helps define syntax—expectation, fulfillment, deception,
and so on—and semantics—tension, release, voice-leading constraints, and so forth.
Discovering function in post-tonal music, however, presents enormously challeng-
ing difficulties. Using computers may make this discovery more feasible and hope-
fully more likely, and accomplished in less time than the millennium it roughly took
to discover function in modal and tonal music.
As previously mentioned, each pitch in a scale, no matter how far removed from
tonality that scale may be, suggests—possibly even demands—a unique relation-
ship with the other members of that scale. Paul Hindemith comments that
anyone to whom a tone is more than a note on paper or a key pressed down,
anyone who has ever experienced the intervals in singing, especially with others,
as manifestations of bodily tension, of the conquest of space, and of the con-
sumption of energy, anyone who has ever tasted the delights of pure intonation
by the continual displacement of the comma in string-quartet playing, must come
to the conclusion that there can be no such thing as atonal music, in which the
existence of tone-relationships is denied. (Hindemith 1942, 155)

Hindemith’s unqualified statement here suggests that function can be found in all
music, regardless of its complexity.
Defining function in post-tonal music obviously requires more specific criteria
than has thus far been discussed. Certainly computer programs will need clear in-
structions on how to compute these functions. The following three categories rep-
resent my own views of post-tonal function:
1. Root, acoustically defined and referenced to a current scale
2. Tension, as defined by acoustical principles
3. Context, as defined by the proximity of the music under analysis
Certainly roots (category 1) of groupings play an important role in function, be-
cause these roots help define the various degrees of a scale in use. However, roots
alone cannot determine the function of groupings, particularly in the case of post-
tonal music, where a single phrase may consist of many different chord types. Ten-
sion (category 2) also contributes to function, because some groupings require
resolution to more stable groupings. However, even roots and tension together can-
not define function. As an example, Figure 5.2a presents a seven-group phrase, each
group of which has the same root and thus represents the same scale degree. How-
ever, the interval content of these groupings differs in such ways that all the groups
produce different tensions. In this example, the tensions increase steadily until the
final grouping, which returns to a tension similar to the opening grouping. Clearly,
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 200

200 HIDDEN STRUCTURE

the penultimate grouping here has a different function from the first grouping of the
passage, even though it shares the same root and thus the same scale-
degree representation. In fact, one could easily imagine the final two groupings
representing a dominant-tonic authentic cadence were it not for their identical
roots.
The same problems exist when using tensions to determine function in post-
tonal music. As an example, Figure 5.2b presents a phrase containing groupings of
similar tension, so that tensions alone seem unlikely to produce significant enough
differences between groupings to assign them anything but the same function
throughout. However, the chord roots, and thus the scale-degree representations,
differ substantially. In this example, it would make more sense to give greater
weight to roots than to tensions.
It would seem that both roots and tensions contribute in varying degrees to the
determination of function in post-tonal music but cannot reveal function by them-
selves. Context provides the necessary conclusive element in defining musical
function. Just like the relationship of chords to keys in tonal music, function in
post-tonal music requires the context of the roots and tensions of neighboring
groupings, as well as not-so-neighboring groupings that provide larger structural
contexts.
Of course, one could further include dynamics, articulations, metric placements,
and so on in determining function. Concentrating as I have in this book on pitch,
however, suggests that although not all inclusive, pitch offers a good start to identi-
fying function as a shared composite of many factors.
We shall now see how these three important fundamentals—root, tension, and
context—triangulate to define function in post-tonal music.

a.

& www www www www wwwww wwwww www

w # ww w ww

b.

& www # www

ww ww www ww w
w w w ww

FIGURE 5.2 Examples where (a) roots remain the same but tensions change and
(b) tensions remain fairly equal but roots change, demonstrating how
neither process alone can determine function in post-tonal music.
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 201

FUNCTION AND STRUCTURE IN POST-TONAL MUSIC 201

THE ACOUSTIC THEORY OF CHORD ROOTS

The following abridged description of Paul Hindemith’s acoustic root theory (Hin-
demith 1942) has been modified following both my own interpretation and informa-
tion revealed since his initial presentation of the theory (Backus 1969; Benade
1976). Much of my interpretation derives from computational, mathematical, and
empirical data (see Cope 2005).
Hindemith’s acoustic root theory originates—as do many theories of music
analysis—with the overtone series, an incomplete example of which appears in Fig-
ure 5.3. The overtone series occurs naturally in one way or another in all but pure
sinusoidal waveforms and acts as the basis for timbre and many other important
musical phenomena. The overtone series consists of a fundamental—usually the
loudest-sounding pitch of the series and the lowest pitch C in Figure 5.3—and a the-
oretically infinite series of overtones figured as incremental multiples of the funda-
mental frequency. Because humans generally hear little above 20,000 cycles (20
kHz) per second, and because the amplitude of the overtones generally diminishes
as they extend above the fundamental, the series here consists of only fifteen over-
tones above the fundamental. The overtone series is also numbered in partials, be-
ginning with the fundamental as the first partial. Thus, sixteen partials appear in
Figure 5.3. These two different numbering systems—overtones and partials—can
produce confusion. However, because using one system or the other has its advan-
tages depending on circumstance, I will reference both systems, clearly identifying
which I am using at the time.
Determining a grouping’s root begins by (1) identifying all of the separate inter-
vals present in the grouping, (2) locating each interval’s lowest occurrence in the
overtone series, (3) determining each interval’s root, and (4) finding the strongest
root among the roots present. Steps 1 and 2 seem simple enough. Steps 3 and 4, how-
ever, require some explanation. To define a single interval’s root involves locating the

w #w w w bw nw w
& w w bw w w
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
w
? w w
w

FIGURE 5.3 The overtone series from the fundamental C.

06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 202

202 HIDDEN STRUCTURE

member of that lowest-appearing interval that represents an upward octave projec-

tion of the fundamental in any overtone series (Hindemith 1942). For instance, the
perfect fifth’s root is its lower pitch, because the lowest fifth appearing in the over-
tone series has as its lower pitch an octave projection of the fundamental (over-
tone 1 or partial 2). The interval of a fourth—the inversion of the fifth—has its
upper pitch as the root, because this pitch falls two octaves above the fundamental
(overtone 3 or partial 4). Without figuring each interval separately, the roots of all
intervals smaller than an octave can be easily remembered as the lower pitches of
all odd-numbered intervals (thirds, fifths, and sevenths), and as the upper pitches
of all even-numbered intervals (second, fourths, and sixths). The tritone has an am-
biguous root because it falls midway between octave projections of the fundamen-
tal. Differences between major and minor versions of seconds, thirds, sixths, and
sevenths will not be distinguished here owing to their close approximations to one
another. Roots of compound intervals will also match the roots of their correspond-
ing intervals within the octave in this definition. For example, the root of a ninth is
its upper pitch, because it reduces to a second, and the root of a second is its up-
per pitch.
Interval derivations based on the overtone series along with their root designa-
tions appear in Figure 5.4a, from the largest to the smallest interval within the oc-
tave. Lower-occurring and lower-rooted intervals produce the strongest roots, and
upper-occurring and upper-rooted intervals produce weaker roots. From this list,
then, seconds have the least root strength because they occur highest in the series
(16/15) and have upper roots. Sevenths have slightly stronger roots because of
their lower projections, but nonetheless occur very high in the series (15/8). Sixths
have more root strength than sevenths because they occur lower in the series
(5/3), but they have less root strength than thirds because thirds occur lower in
the series (5/4 and 6/5) and have lower roots. As just mentioned, the tritone, even
though it has low placement in the series (7/5), divides the octave in half, and
therefore does not have a clear root. The perfect fifth has the strongest root of all
intervals owing to its low placement (3/2) and lower root. The perfect octave—not
shown—ultimately projects the strongest root but represents doubling, because
both members belong to the same pitch class. The interval root strengths appear
in Figure 5.4b in order of strength. The lowest strongest interval occurring more
than once in a grouping produces the grouping’s root.
Figure 5.5 provides six groupings as examples of the above-described root-
identification process. In grouping (a), the fifth D–A represents the strongest inter-
val present, with its lowest member the root of the grouping. Figure 5.5b poses
more complex problems, but C–G-sharp (A-flat) is the strongest interval, with its
upper pitch (G-sharp) as root. In Figure 5.5c, the root results from the perfect fifth
D-flat–A-flat. The perfect fifth E–B in Figure 5.5d produces the root E. Figures 5.5e
and Figure 5.5f have the same root, C, the result of the perfect fifth C–G. From these
simple examples, one can see how this root-finding process poses few problems
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 203

FUNCTION AND STRUCTURE IN POST-TONAL MUSIC 203

a)
a. b)
b.
int ratio root int ratio root
M7 (15/8) lower P5 (3/2) lower
m7 (16/9) lower P4 (4/3) upper
M6 (5/3) upper M3 (5/4) lower
m6 (8/5) upper m3 (6/5) lower
P5 (3/2) lower M6 (5/3) upper
A4 (7/5) unclear m6 (8/5) upper
P4 (4/3) upper M2 (9/8) upper
M3 (5/4) lower m2 (16/15) upper
m3 (6/5) lower m7 (16/9) lower
M2 (9/8) upper M7 (15/8) lower
m2 (16/15) upper A4 (7/5) unclear

FIGURE 5.4 (a) Interval derivations from the overtone series along with their
root designations, and (b) interval root strengths shown in order of
strength.

& www
w # # www n b b wwwww # n wwwww w
n n wwww b b wwwwww
a. b. c. d. e. f.

FIGURE 5.5 Six groupings to serve as examples of the root-identification process.

compared to tonal harmony built in thirds, where unwinding thirds in inverted

chords can require quite complicated operations.
Interestingly, one might imagine that the leftmost zero of prime-form pitch-class
sets would be the root of the set. However, although this is true for some prime-
form sets, many other sets have different roots. For example, the twelve prime-form
trichords turn out as follows, with the prime-form pitch-class set presented to the
left and the set’s pitch-class root to the right.
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 204

204 HIDDEN STRUCTURE

0,1,2 2
0,1,3 0
0,1,4 0
0,1,5 5
0,1,6 1
0,2,4 0
0,2,5 5
0,2,6 2
0,2,7 0
0,3,6 0
0,3,7 0
0,4,8 0

With more complex chords (pentachords, and so on), the roots vary as well. For ex-
ample, the first four pentachords in pitch-class-set order have 0, 5, 0, and 5 as roots,
and the first four hexachords in pitch-class-set order have 5, 0, 5, and 5 as roots.
A simple example of pseudo-code appears below, where an object-oriented
method assigns a root to instances of an object class called set.

(defmethod initialize-instance ((object set) &rest initargs)

(declare (ignore initargs))
(if (> (tension-range object)(root-range object))
(put-functions-into object tensions)
(put-functions-into object roots-and-scales)))

Here, the conditional if computes which of the ranges—tension or root—will de-

termine the function. Although the functions tension-range, root-range, and
put-functions-into do not actually exist in the files accompanying this book on
CD-ROM, readers who have followed the discussion of Lisp functional and OOP
processes should, with the assistance of one or more of the Lisp books listed in the
bibliography (Graham 1995; Steele 1990; Touretzky 1990; Wilensky 1986), be able
themselves to produce a working model based on this pseudo-code.
Whereas roots in tonal music tend to be the sole determining factor in deciding
function, roots in post-tonal music can have much less influence on function. As
has just been shown, roots of groupings can be strong, as determined by perfect
fifths and fourths, or relatively weak, as determined by sixths and thirds. As we will
soon see, a strongly defined root tends to support its determination of function,
while a weak root does just the opposite.
Although all pitches contribute to a scale in use, chord roots especially tend to
define scale membership. In order to clearly place roots into scales, we need a no-
tation that is both simple and easy to understand. Therefore, the notation I will use
in this book consists of movable-do pitch-class numbers. Members of scales in this
post-tonal approach appear simply as pitch classes in the scale in which they belong.
Hence, a 7 indicates a perfect fifth above 0. All numbers not present in the scale
have a small suffix “ø” to indicate their out-of-scale nature. Therefore, the [0 1 2 3 4
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 205

FUNCTION AND STRUCTURE IN POST-TONAL MUSIC 205

5 6 7 8 9 t e] set indicates a full chromatic scale in t-normal form. The major and
octatonic scales are listed below as further examples of this notation:
Major Scale
1ø 3ø 6ø 8ø tø
0 2 4 5 7 9 e 0
Octatonic Major Scale
1ø 4ø 7ø tø
0 2 3 5 6 8 9 e 0
Obviously, any scale based in the twelve-tone temperament can be represented in
this way.
This notation for representing scale membership of pitch class roots, although it
has several drawbacks, has many virtues. The principal drawback involves the
movable-do use of 0–e pitch classes. In other words, 0 no longer represents C but
rather the first pitch of the analyzed scale. Of course, looking at this representation
as a t-normal set resolves this problem. The main virtue of this representation is
that it requires no memorization of new symbols aside from the ø, nor does it in
any way contradict or cause confusion with the Roman numerals of tonal functional
analysis.
As alluded to earlier, the notation just discussed indicates only a numbering no-
tation for scales and roots within those scales. This notation does not necessarily
indicate actual musical function. The pitch class 0 might represent a function such
as tonic (home base, etc.) in tonal music but may not indicate the same for post-
tonal music. As we will see, this concept of scale degrees and roots projecting pos-
sibly different functions is extremely important.

MUSICAL TENSION
Figure 5.6 presents a list of intervals from a perfect unison to a perfect twenty-ninth
and my own views of their tensions. The SPEAC Analysis software—discussed in
more detail later in this chapter—that accompanies this book on CD-ROM allows
users to adjust these levels according to their own personal tastes. Although the
list presented here involves some simple mathematics—especially when consider-
ing octave projections of intervals within the octave—I have not produced it using
any overall mathematical formula. I prefer to gauge tension based on refinement of
rough estimates derived from formulae (see Cope 2000).
Computations of groupings require the addition of the tensions of all of the inter-
vals present. For example, to find the tension of a major triad in closed position
would require the addition of the tensions of a major third, a perfect fifth, and a mi-
nor third, the three intervals present. Doublings, dynamics, and so on, all viable
factors in the effective evaluation of grouping tension, have not been included here,
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 206

206 HIDDEN STRUCTURE

unison 0.0 minor sixteenth .925

minor second 1.0 major sixteenth .725
major second .8 minor seventeenth .175
minor third .225 major seventeenth .125
major third .2 perfect eighteenth .475
perfect fourth .55 augmented eighteenth .575
augmented fourth .65 perfect nineteenth .025
perfect fifth .1 minor twentieth .2
minor sixth .275 major twentieth .175
major sixth .25 minor twenty-first .625
minor seventh .7 major twenty-first .825
major seventh .9 perfect twenty-second 0.0
perfect octave 0.0 minor twenty-third .91
minor ninth .95 major twenty-third .71
major ninth .74 minor twenty-fourth .135
minor tenth .175 major twenty-fourth .11
major tenth .15 perfect twenty-fifth .46
perfect eleventh .5 augmented twenty-fifth .56
augmented eleventh .6 perfect twenty-sixth .01
perfect twelfth .05 minor twenty-seventh .185
minor thirteenth .225 major twenty-seventh .16
major thirteenth .2 minor twenty-eighth .61
minor fourteenth .65 major twenty-eighth .81
major fourteenth .85 perfect twenty-ninth 0.0
perfect fifteenth 0.0

FIGURE 5.6 A list of intervals weighted according to their tension levels.

because these computations are so particular to individual contexts that these cal-
culations would require several hundred pages to describe.
Figure 5.7 provides simple examples of this process. The major and minor triads
have three positions, each with increasing tension. The tensions described in the
figure caption reasonably fit standard expectations, with the second-inversion triad
having the most tension and the first-inversion triad having less tension than the
second-inversion but more tension than the root-position triad. Figure 5.8 shows four
other familiar triads for tension analysis. These triads show ever-increasing levels of
tension, though these results may not exactly agree with some readers’ interpretations.
From this point on in this book I will often use the terms stable and unstable
interchangeably with tension levels. Stable infers less tension and less need to re-
solve that tension. Unstable infers more tension, suggesting the necessity of resolu-
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 207

FUNCTION AND STRUCTURE IN POST-TONAL MUSIC 207

ww ww ww b ww
& www w w b www w w

FIGURE 5.7 Figuring interval weights from the bass note by using the chart in
figure 5.6 produces 0.3 (M3 at 0.2 + P5 at 0.1), .5 (m3 at 0.225 + m6 at
0.275), and 0.8 (P4 at 0.55 + M6 at 0.25), respectively. The minor triad
and its inversions produce 0.325 (m3 at 0.225 + P5 at 0.1), .45 (M3 at
0.2 + M6 at 0.25), and 0.825 (P4 at 0.55 + m6 at 0.275).

& # www b b www n n b wwww b b wwww

FIGURE 5.8 The augmented triad at 0.475 (M3 at 0.2 + m6 at 0.275), the diminished
triad at 0.775 (m3 at 0.225 + A4 at 0.55), the dominant seventh chord
at 1.0 (M3 at 0.2 + P5 at 0.1 + m7 at 0.7), and the diminished seventh
chord at 1.125 (m3 at 0.225 + A4 at 0.65 + M6 at 0.25).

tion of that tension. Although these terms are vague, they serve a useful purpose
when one compares two groupings with different tensions. As we shall see shortly,
tension is relative: a chord that may seem unstable in the abstract may in fact be
either stable or unstable, depending on its context.
Figure 5.9 shows four less common chord types, with their respective tensions
presented in the figure caption. The low tension score of the third chord in this
series—sometimes referred to as an added-sixth chord—may indicate why it often
occurs in final cadences of popular music, its tension remaining relatively low even
with four different chord members present. Interestingly, the strongest interval in
this chord—a perfect fifth—suggests C as its root instead of A, the root when it is
calculated as a submediant chord in standard tonal harmonic analysis.
Figure 5.10 shows four more chords based on projections of intervals other than
thirds. These four chords show how instability can increase in more complex
music. The results here—with the calculations presented in the figure caption—
indicate the third chord as the most unstable and the first as the most stable.
Again, however, I remind readers that later in this chapter we will see how stability
and instability are relative, with chords like the first chord in Figure 5.10 becoming
unstable in otherwise consonant circumstances and chords like the third chord be-
coming more stable when surrounded by chords containing several more minor
seconds.
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 208

208 HIDDEN STRUCTURE

& wwww b b wwww n wwww n wwww

w

FIGURE 5.9 Four chords producing tensions of 1.2 (M3 at 0.2 + P5 at 0.1 + M7 at
0.9), 1.25 (m3 at 0.225 + P5 at 0.1 + m7 at 0.7), 0.55 (M3 at 0.2 + P5 at
0.1 + M6 at 0.25), and 2.0 (M3 at 0.2 + P5 at 0.1 + M7 at 0.9 + M2 at 0.8).

& www b bn www

b b www # # n www nw

FIGURE 5.10 Chord tensions adding to 1.0 (M2 at 0.8 + M3 at 0.2), 1.225 (m2 at 1.0
+ m3 at 0.225), 1.8 (m2 at 1.0 + M2 at 0.8), and 1.475 (P4 at 0.55 + m7
at 0.7 + m3 at 0.225).

Each of the examples shown thus far appears outside of any such context. I ar-
gue that context very much influences how we perceive stability. Metric placement
and duration represent two of these contextual factors (I will discuss others in the
next section). To calculate the metric placement of musical rhythm I employ the
simple mathematical formula p = (b × 0.1)/m, where p equals placement value, b
refers to beat number, and m represents metric beat. This latter variable accounts
for changes in tension from, for example, beat 1 to beat 2 in 4/4 meter, where beat 2
has more tension than beat 1, and so on. Concomitantly, this linear formula also
gives beat 3 in 4/4 meter more tension than beat 2, thus contradicting the more
commonly considered opposing point of view. Interestingly, my own conception of
beat tensions more closely matches the formula’s output than tradition suggests.
Although this tension formula may seem surprisingly simple, I have experimented
with a variety of different tension-calculating processes and always return to the
formula presented here. However, no simple mathematics can compute the com-
plex metric relationships of beats in all meters. Therefore, I use lookup tables for
this purpose.
Figure 5.11 presents a simple lookup table of metric tensions for principal beats
in twelve different meters. I have omitted offbeats in order to present a more read-
able example, and because offbeats can be computed using the same formula as
presented above. This lookup table results from a strict application of the previ-
ously described formula. I then adjust these calculations to suit my personal con-
cepts of stability and instability of beats and sub-beats in standard meters. These
adjustments rarely exceed 0.05 in either direction and amount to fine-tuning rather
than radical change.
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 209

FUNCTION AND STRUCTURE IN POST-TONAL MUSIC 209

1 2 3 4 5 6 7 8 8 10 11 12
12 .008 .03 .06 .083
9 .011 .044 .078 x x x
8 .0125 .038 .063 .088 x x x x
7(3+2+2) .014 .06 .09 x x x x x
7(2+3+2) .014 .057 .09 x x x x x
7(2+2+3) .014 .042 .07 x x x x x
6 .02 .67 x x x x x x
5(3+2) .02 .08 x x x x x x x
5(3+2) .02 .06 x x x x x x x
4 .025 .05 .025 .075
3 .033 .067 .1
2 .025 .05

FIGURE 5.11 A simple lookup table of metric tensions for principal beats in twelve
different meters.

Notes of extended duration provide agogic weight. For example, a half note be-
ginning on beat 2 of a 4/4 measure extends over the higher-tension third beat of the
measure. Although this extension should not include the complete metric weight of
this third beat, the half note should receive some kind of added durational tension.
In my book Computer Models of Musical Creativity (2005) I describe some of the
complicated processes required for calculating duration as stability and instability:
Figuring durational weightings is complicated due to my belief that consonant
groupings are made more consonant by extending duration, while dissonant
groupings are made more dissonant. To account for this, I use a combination of
percentages derived from adding a grouping’s duration multiplied by .1 to the
same grouping’s tension multiplied by .1. Thus, a quarter note major triad, for
example, receives an additional durational weight of .025 (1/4 × .1) added to .03
(.3 × .1) or .055, while an eighth note dominant-seventh chord receives a dura-
tional weight of .0125 (1/8 × .1) added to .1 (1.0 × .1) or .1125. (Cope 2005,
232–233)

and I will continue using this formula here.

In that book I also compute approach tension figured between the roots of two
successive groupings. Because the interval tensions given previously can serve as a
guidepost for horizontal root motions, I use them multiplied by 0.1 in calculating
approach tensions (my latest variation in a long succession of variations of this
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 210

210 HIDDEN STRUCTURE

particular factor). Thus, a grouping approached by the interval of a fifth from a pre-
ceding grouping receives an additional 0.01 weight, while this same grouping ap-
proached by a minor second from a preceding grouping receives an additional 0.09
weight. As an example, the tension of a strong dominant-tonic motion in a tonal
cadence is little affected by the approach motion, while a weaker root motion of
a third—as in a tonic-to-mediant progression—has more tension and achieves a
greater additional weight as a result.
Figure 5.12a provides an example of tension computation based on combinations
of interval content, metric placement, duration, and motion of approach to the root
in tonal music. Some of the tensions shown in this example from Bach might ap-
pear counterintuitive. For example, the lowest-tension chord—the second chord of
the passage—represents a subdominant rather than a tonic function. However, in
terms of computation, based on the four previously mentioned factors, these ten-
sions make sense. The first, fourth, and last chords, for example, though exactly
the same in pitch, register, voicing, and duration, do not have exactly the same ten-
sions because the chords occur on different metric beats and thus play different
roles in the music. As well, the most dissonant—and hence unstable—chord of the
phrase has the highest tension (the second-to-last chord).
Figure 5.12b presents the same type of analysis as Figure 5.12a but for a post-
tonal passage by Schoenberg. This music obviously has a greater range of tensions
than does the Bach of Figure 5.12a. The Schoenberg also appears to have a roughly
sinusoidal shape, peaking first at 9.47 and then again at 16.34. The tensions here
are compounded by my figuring all pitches present rather than ignoring the decay-
ing pitches of previously struck chords. I have also freely grouped the second half
of m. 4 to avoid using each new entrance as a new grouping. Obviously, when
groupings such as these have high tension levels, tensions based on metric place-
ment, duration, and approach to the root play significantly smaller roles than in
tonal music.
Thus, four factors contribute to the tension of a given grouping: interval content,
metric placement, duration, and root-to-root approach interval. Of these contribu-
tors, only one involves context between groups—root-to-root approach interval. I
mention this here because context will play an increasingly important role in inter-
preting tension and thus stability and instability when determining function as we
proceed. This tension-calculating process, dependent as it is on intervals, meter,
and duration, does not fully account for all of the factors involved with tension.

CONTEXT AND SPEAC

No two musical pitches or groups of pitches in a musical composition are equiva-
lent, even if they appear to be so. This statement represents a simple truth, for no
two objects can occupy the same time and space, and thus they vary from one an-
other in at least these two aspects. Musical context varies pitches or groups of
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 211

FUNCTION AND STRUCTURE IN POST-TONAL MUSIC 211

U
a.
## œœ œœ œœ
& # c œœ œœ œœ œœ œœ
U
Piano
œœ œ œ œœ œœ œ œ œ œœ
? # ## c œ œ œ
interval: 1.25 .525 1.6 1.25 1.325 1.49 2.375 1.25
p = (b * .1)/m: .075 .025 .05 .025 .075 .025 .05 .025
approach: .0 .055 .055 .15 .01 .01 .08 .01
duration: .23 .16 .27 .24 .25 .26 .35 .23
1.56 .77 1.94 1.7 1.75 1.79 2.74 1.52

b.
# œœ ww ˙˙ # ˙˙ ww ˙ œ ˙
œ ˙˙ # œ Œ # œœ ˙˙
& c Ó Œ œ w ˙ ˙ w #˙
Ó
π
Piano
Ø ‰ j #˙.
?c ∑ Ó ˙˙ Ó Œ n ˙ . n œ # œ wœ ˙˙ .
&
n nn ˙˙˙ ˙ n n ˙˙ . n œ w ˙.
interval: 1.54 8.4 1.54 8.4 14.64 6.79 6.37 1.54
p = (b * .1)/m: .075 .025 .025 .05 .025 .05 .075 .025
approach: .0 .09 .09 .09 .09 .0 .0 .09
duration: .26 .95 .27 .95 1.58 .78 .75 .25
1.88 9.47 1.93 8.55 16.34 7.62 7.2 1.91

FIGURE 5.12 (a) J. S. Bach, Chorale no. 42, demonstrating the same chord in
differing contexts; (b) Schoenberg, Sechs kleine Klavierstücke op. 19,
no. 6, opening measures, demonstrating the same types of contextual
differences.

pitches yet further. The notion of apparent equalities being unequal owing to dis-
crepancies of time and space is an extremely important real as well as conceptual
one.
In order to fully understand how the same music may have different functions,
we need to look beyond roots and tensions of groupings toward a kind of musical
semantics. This musical semantics can inform analysts about music’s expectation
and fulfillment or deception, predictability and unpredictability. In short, the funda-
mental motion of music—all music—results from its context. Furthermore, analyz-
ing musical context reveals important relationships about music regardless of the
style or type of music involved.
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 212

212 HIDDEN STRUCTURE

Language can provide a good analogy. In language, for example, we assume that
context reveals a great deal about the meanings of words and phrases. The follow-
ing sentence presents an instance of how words that are spelled and pronounced
the same can have quite different functions and meanings, even when placed in
close proximity to one another:
I light the light with a light touch.

The word light appears three times in this sentence, with each appearance having a
different meaning and making a different syntactic contribution. Only the context
distinguishes each word’s true function and meaning. We adjust to such subtle
shifts in meaning by using contextual references, giving language another powerful
way in which to express ideas.
“Verbing” represents another example of this kind of language context. Verbing
involves treating a noun or other non-verb as a verb. In context, this typically
makes immediate sense to anyone familiar with the language being spoken. Exam-
ples include “I was carded at the door” and “I will text him your answer.” In both
cases here, the word verbed—interestingly, itself an example of verbing—is a noun
relying on syntax and context for understanding. Languages develop in this way, by
providing a rich source of new words coupled with a true sense of context.
The same is true of music. In tonal music, leading tones represent one way in
which context differentiates between apparently identical functional motions in
music. The leading-tone pitch in C major, B, for example, strongly leans toward the
tonic pitch when appearing alone or when found in dominant, dominant seventh,
and leading-tone harmonies. However, when this pitch appears as the fifth of the
mediant triad, it does not particularly lean toward the tonic pitch. In fact, the
leading-tone pitch in the mediant chord often moves more naturally elsewhere—to
the submediant pitch, for example. Thus, the same leading-tone pitch can be ana-
lyzed differently depending on its context.
Schoenberg took into account this notion of context overriding predisposed
function when he stated that “every tone that is added to a beginning tone makes
the meaning of that tone doubtful” :

If, for instance, G follows after C, the ear may not be sure whether this expresses
C major or G major, or even F major or E minor; and the addition of other tones
may or may not clarify this problem. In this manner there is produced a state of
unrest, of imbalance which grows throughout most of the piece, and is enforced
further by similar functions of the rhythm. The method by which balance is re-
stored seems to me the real idea of the composition. (Schoenberg 1983, 123; ital-
ics in original)

As we have seen, the music in Figure 5.12 provides two examples of contextual
differentiation in music. In Figure 5.12a, the chords that sound on the upbeat to full
measure 1, the third beat of full measure 1, and the third beat of full measure 2 con-
sist of exactly the same pitches. However, the upbeat chord to full measure 1 intro-
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 213

FUNCTION AND STRUCTURE IN POST-TONAL MUSIC 213

duces the melody and key of the chorale. The third beat of full measure 1 substanti-
ates that key as well as acting as the harmonic grounding for the leap that appears
subsequently in the melody. The third-beat chord of full measure 2, on the other
hand, verifies the key and acts as a cadential tonic function. These three chords,
then, although they contain precisely the same pitches, play very different roles
and, one could posit, have very different functions. I would argue that the same
thing occurs in the post-tonal passage in Figure 5.12b, from op. 19, no. 6, by
Schoenberg, where the three-pitch chord in the right hand repeats twice, as does a
three-pitch chord in the left hand, though what the different functions are is not as
clear in Figure 5.12b as it is in Figure 5.12a.
One might argue that traditional analyses indicates different functions for identi-
cal chords as well, particularly when such chords appear in different keys. For ex-
ample, a C–E–G tonic chord in C major in one location in a work does not have the
same function as a C–E–G subdominant chord in G major elsewhere in the same
work. One could further argue that pivot chords in modulation provide even more
complicated double meanings. However, these tonal examples clearly represent
special cases. To offer the kind of differentiations referred to in the previous para-
graphs requires that an analysis process delineate differences between C–E–G
chords in exactly the same register within the same key. Therefore, the previously
discussed processes might actually designate a C–E–G chord in C major in one loca-
tion in a work and a C–E–G in G major elsewhere in the same work as equivalent in
function. In each of these last two cases, the context determines the analysis, not
the opposite.
For many years I have usefully employed a contextual analysis process called
SPEAC (pronounced “speak”). SPEAC stands for S (statement) for stable, P (prepara-
tion) for weakly unstable, E (extension) for fairly stable, A (antecedent ) for very un-
stable, and C (consequent) for strongly stable. SPEAC identifiers thus follow the
order of stability of A–P–E–S–C, with the most unstable identifier on the left and the
most stable identifier on the right. Basing analysis on a combination of root, musi-
cal tension, and context most clearly parallels my musical sensibilities and repre-
sents the core of the analysis component of my algorithmic composing programs
(see Cope 1991 and 1996). SPEAC antecedents have more tension than statements,
preparations, or extensions. Likewise, statements have more tension than either
preparations or extensions. Following the work of Heinrich Schenker (1979), SPEAC
provides useful insights into musical function, especially in the analysis of post-tonal
music.
SPEAC analysis first requires segmenting music into groupings defined either by
some arbitrary length or by each note onset and then calculating vertical tension
separately for each grouping following the processes described earlier in this chap-
ter. To determine a grouping’s tension involves slicing the grouping vertically, ac-
cording the slice with the least tension the overall tension of the grouping, because
the program assumes that the higher-tension slices contain nonharmonic tones
that resolve. In Figure 5.12a, for example, beat 1 of full measure 1 divides into two
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 214

214 HIDDEN STRUCTURE

vertical slices (onbeat and offbeat); the onbeat slice is used for SPEAC analysis be-
cause it contains the least tension.
To assign SPEAC identifiers generally requires the setting of the highest tensions
of a series of groupings to antecedents (A) and the lowest tensions to consequents
(C), with statements (S) closest to consequents (C), preparations (P) closest to an-
tecedents (A), and extensions (E) falling equidistant between Cs and As. Other fac-
tors must also play a role. For example, the first-heard chord or grouping will most
likely receive a P or an S, depending on its tension level. These assignments result
from a lack of context within which to judge the music.
As a result of these processes, no two groupings labeled A in SPEAC have pre-
cisely the same amount of “A-ness.” Therefore, not only can SPEAC differentiate be-
tween two identical groupings, but it can also further differentiate between two
groupings, both of which have the same SPEAC analysis. Suffixed subscripts allow
SPEAC to represent such differentiations. Thus, a C1 and a C2 both represent conse-
quences, but they have different degrees of consequence. The hierarchy of these
subscripts flows from high value to low value in inverse proportion to the size of
the number. Therefore, a C1 has more C-ness than a C2. For example, in the passage
P1 S1 A2 C2 A1 C1 the final C1 has deeper consequence than the earlier occurring C2.
Figure 5.13 presents the same examples as does Figure 5.12. In Figure 5.13, how-
ever, the numerical tensions have been translated into SPEAC functional descrip-
tors. In Figure 5.13a we see the duplicate chords that sound on the upbeat to full
measure 1, the third beat of full measure 1, and the third beat of full measure 2 with
different representations—P1, C2, and C1. Bach’s phrase, when viewed in terms of
its most unstable and most stable elements provides, an S1, A1, C1—or statement,
anticipation, consequent—logic. Interestingly, the Schoenberg passage in Figure 5.13b
has precisely the same functional progression.
SPEAC assignments are not definitive. Analysts have the right (even the obliga-
tion) to assign SPEAC symbols as appropriate to their personal interpretations. As
an example, consider the extension (E2) that I have assigned to beat 1 of m. 3 in Fig-
ure 5.13a. Following the anticipation of the final chord of the previous measure and
the fact that this extension is a variant of the chord that appears on beat 1 of m. 2,
it might more likely represent an S or C. Situations such as this require that the bal-
ance between root and tension, even between the various contributors to tension
itself, contribute their weightings in different proportions to one another, as I men-
tioned earlier in this chapter. This strategy allows one to interpret numerical scores
more liberally, a fact that may give some readers pause. However, more than likely,
such personal interpretations should be viewed as refinements to the approach to
function described here, rather than aberrations of it. Because computers do not as
of yet have personal views of anything, programming these interpretations is possi-
ble only by using artificial intelligence techniques, many of which I discuss in Chap-
ter 7 (association networks, fuzzy logic, etc.).
Obviously, many other factors play roles in the tensions we hear in music. Chro-
maticism, timbre, dynamics, register, and so on can affect and even override the
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 215

FUNCTION AND STRUCTURE IN POST-TONAL MUSIC 215

U
##
a.
œœ œœ œœ
& # c œœ œœ œœ œœ œœ
U
Piano
œœ œ œ œœ œœ œ œ œ œœ
? # ## c œ œ œ
P2 S1 A2 C2 A2 E2 A1 C1

# œœ ww ˙˙ # ˙˙ ww ˙ œ ˙
œ ˙˙ # œ Œ # œœ ˙˙
b.

&c Ó Œ œ w ˙ ˙ w #˙
Ó
π
Piano
Ø ‰ j
#˙.
?c ∑ Ó ˙˙ Ó Œ n˙. nœ #œ œw
˙˙ .
&
n nn ˙˙˙ ˙ n n ˙˙ . n œ w ˙.
P2 E2 S1 P2 A1 E2 E2 C1

FIGURE 5.13 SPEAC analysis of the music in Figure 5.12a and b.

tension that SPEAC generates through its various pitch analyses. However, such
factors are often mitigated by the small degree of effect that they impose on ten-
sion. Therefore, the approach described above provides enough substantive infor-
mation to be useful in defining function.

FUNCTION
Allen Forte frequently writes about function in tonal music as well as post-tonal mu-
sic, taking note of how analysts have difficulty applying systematic approaches:
When the trained analyst examines a musical score, he associates certain signs
with others to form units and makes a series of basic decisions about the tempo-
ral spans of such units and their internal structuring. In a metaphorical sense he
places a template over the score to frame patterns and to show how they are in-
terwoven to form local contexts. Although the decisions the analyst makes may
rest upon years of practical experience, they are often unsystematic and are sub-
ject to many influences that are not easily identified. (Forte 1993, 56)
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 216

216 HIDDEN STRUCTURE

In contrast to what Forte describes here, the approach described below follows a
more systematic application of the basic principles discussed thus far in this chap-
ter. The functional notation I use, resulting as it does from a triangulation of root,
tension, and context, appears as a combination of scale degree (defined by root),
tension, and context:
0S1
Here 0 represents the scale degree of the root of the grouping being analyzed and
the notation S1 indicates a first-level Statement, the combination of root, tension,
and context. Any root notation can have any SPEAC tension and context identifica-
tion. For those familiar with more static notation, such as I, IV, V, and so on in tonal
music, the dynamic representation of post-tonal function can be initially confusing.
The root-SPEAC identifications pose interesting resonances and contradictions, be-
cause 0—what one might presume as a kind of tonic—can appear as 0S, 0P, 0E, 0A,
or 0C, giving it a full range of stable and unstable characterizations.
If one further identifies each member of the movable-do 0–e notation as belong-
ing to either T, D, or S (tonic, dominant, or subdominant) functions, as envisioned
by Hugo Riemann (Harrison 1994; see Chapter 1 of this book for more information),
the problems of comparing scale-degree-SPEAC function with tonal functions are
yet further exacerbated. The following list provides possible T, D, and S equivalents:
0=T
1=D
2=D
3=T
4=T
5=S
6=S
7=D
8=S
9=S
t=D
e=D
In SPEAC terms, it would seem natural that T(onic) would resemble S and C, that
D(ominant) would resemble A and possibly P, and that S(ubdominant) would re-
semble P and possibly E. Thus, potential contradictions increase significantly. Such
contradictions, however, can provide fascinating analytical information. A work
ending on, say, a 0A1 would suggest that either one or more aspects of the analysis
system I propose here are incorrect, or that the music under analysis terminates
with something significantly not representing resolution or stability. Thus, even
though I do not typically evoke the TDS approach in tandem with roots and SPEAC, I
find more than occasional interest in following their connections (and lack thereof ).
In summation, function in post-tonal music as I define it involves the analysis of
scale, roots, tension, and context to derive a simple but inclusive representation
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 217

FUNCTION AND STRUCTURE IN POST-TONAL MUSIC 217

such as 0C1. When SPEAC and root-scale assignments agree, the results support,
but do not necessarily confirm, a strong functional definition. When SPEAC and
root-scale assignments disagree, one of two possibilities exists: the music’s func-
tion is confused, or the interpretation is wrong or misguided. In cases where the
latter interpretation is suspected, I find that the earlier reference to personal inter-
pretation can often resolve the problem. Of course, such solutions can prompt
the question of whether one is simply massaging the solution to fit the problem. In
general, however, tweaking tensions or SPEAC interpretation based on aurally per-
ceived contradictions resembles, for example, how one determines modulation
versus secondary regions in tonal music.
Fred Lerdahl describes a similar—but nonetheless fundamentally different—
approach to post-tonal functional analysis (Lerdahl 2001; see particularly 347 ff.).
Lerdahl’s theory involves five classifications: RS (referential sonority—a kind of
tonic replacement), Dep (departure—forward-branching), Ret (return—backward-
branching), N (neighboring—subordinate within a strong progression), and P
(passing—subordinate particularly to Dep or Ret). Lerdahl analyzes (in 2001, 353–
368) Schoenberg’s Three Piano Pieces, op. 11, no. 1 according to these principles
and includes a hierarchical prolongational analysis as well. Lerdahl points out that

without an ordered space and with the concomitant perceptual dependency on

salience for organizing prolongational structure, there is a relatively weak corre-
lation in most atonal music between elaboration and tension. Hence, it is prefer-
able to refer not to atonal prolongational tension and relaxation but just to atonal
departure and return. (Lerdahl 2001, 348)

Thus, Lerdahl distances his analytical approach to post-tonal functional music

analysis from tonal functional music analysis. RSDepRetNP, then, like SPEAC, relies
on contextual rather than static definitions. Nonetheless, music thus analyzed gen-
erates rich hierarchical functions that gather strength as they ascend toward the
higher and more abstract structural levels.

FORM AND STRUCTURE

Musical form and structure represent very different aspects of music, though invok-
ing these names often produces similar interpretations. For this book, musical form
refers to aspects of the musical surface that state, contrast, and develop themes
and motives. For example, terms such as A and B—representing contrasting areas
of material—often catalog the formal aspects of a musical work. Musical structure,
on the other hand, refers to the process of hierarchical differentiation of more sig-
nificant elements from less significant elements. (I should note here that the word
structure in this book’s title refers to the generic notion of structure as order, not
the notion of structure as hierarchy. Unfortunately, Hidden Order is the title of a
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 218

218 HIDDEN STRUCTURE

well-known book by John Holland, one that I reference several times here, and thus
was not available.)
Computationally discovering formal thematic areas, as well as their similarities
and differences, requires pattern matching to indicate when themes return, vary, or
contrast:
One of the purposes of analyzing musical structure and form is to discover the
patterns that are explicit or implicit in musical works. Pattern in music is gener-
ally described in ordinary language, supplemented by technical musical terms:
for example, “the movement is in sonata form,“ “the opening section is in the key
of C major[,] it is followed by a section in the dominant, then a return to the orig-
inal key[,]” “the chord is a G seventh[,]” “the slow movement is written in 3/4
time.” No complete formalism has existed for describing a musical pattern pre-
cisely, its exact nature can be communicated only by writing out the music in
extenso—the actual notes in musical notation. Although some abbreviation is
achieved by such notation as figured bass, all established notations set forth the
notes essentially in the order of their temporal occurrence. (Simon and Sumner
1993)

The use of the word pattern in this passage reveals its importance when one is deal-
ing with both form and structure in music. As we shall see, my analysis programs
use pattern recognition to discover both form and structure in post-tonal music.
As opposed to signatures, which require the matching of patterns in at least
two separate works for the identification of style (Cope 1991, 1996, 2000, and 2001),
some patterns have local importance and relate to harmonic, thematic, and rhyth-
mic elements of a theme or other distinctive musical idea. Discovering such pat-
terns can be quite important for revealing musical form. I have named these
patterns unifications (see Cope 2000, 171–74, for more information). Unifications are
patterns found most frequently in a single section of a work that unify the section
and distinguish it from other sections. Unifications maintain unity in music at the
foreground, or formal, level.
I use three basic techniques to discover unifications in music. First, and most ob-
viously, patterns are collected only from the music of a phrase or section of music.
Second, my programs match patterns using intervals, not pitches. Lastly, my pro-
grams look for patterns that conform to certain prescribed limits within individual
phrases rather than between phrases. In effect, only patterns integral to the basic
continuity of the music appear.
My analysis programs also ascertain the number, location, and variance types of
unifications. For example, knowing that a certain pattern is repeated during a pas-
sage is of little use if one does not also know whether this pattern repeats a few or
many times, repeats only in certain circumstances, repeats only after predictable
delays, repeats only with certain variations such as sequence, and so on. Unlike
other types of patterns, unifications may be quite numerous and, rather than
spreading their use over entire compositions, appear in close quarters, even in
combination and contiguously.
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 219

FUNCTION AND STRUCTURE IN POST-TONAL MUSIC 219

In order to discover unifications, music must obviously be separated into

phrases. The processes used to computationally accomplish this are complex and
often redundant—to ensure that supposed phrases are phrases in fact. Rather than
repeat the descriptions of the processes I use for detecting phrases in music here,
particularly because these processes require significant space to describe, I refer
readers to Cope 1996, 2000, and 2001, in particular. Once phrases are detected and
separated out from their original work, unifications can be revealed computation-
ally in several ways. The method described below provides the best way that I
have found to accomplish this analysis.
Figure 5.14 presents an example of a unification (Figure 5.14a) as derived from
Schoenberg’s Three Piano Pieces, op. 11, no. 1 (Figure 5.14b). As seen here, the

a.

? 3 ‰ œ #œ œ #œ œ
4 #œ. ‰ Œ

b.
3 j
&4 Œ œ #œ nœ. nœ nœ n˙ b˙ nœ Œ
n œœ b œ n n œœ
Œ n˙ Œ
Piano
p
? 43 n˙ n˙ #œ
∑ Œ
b˙ Œ b˙ Ó
# n œœ

5
j
& œ ‰ ‰ œ. œ. ‰ Œ Œ œ œ Œ Œ
œ œ bœ nœ. ‰ œ œ b œ .œ nœ
J ‰ Œ rit.

? nœ #œ nœ ‰ Œ ‰
œ #œ œ #œ œ Œ bœ œ œ # œœ # œ
œ #œ œ
œ. ‰ #œ œ. ‰ Œ ‰ . œ

FIGURE 5.14 An example of a unification as derived from Schoenberg, Three Piano

Pieces op. 11, no. 1: (a) the unification, and (b) the first 8 measures
of the work from which the unification was drawn. Note the other
unifications present here as well.
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 220

220 HIDDEN STRUCTURE

chosen unification appears three times in various guises—the same pitches, but
different metric placement. Pattern matchers have little problem with discovering
unifications such as this. In fact, pattern matchers can typically discover unifica-
tions without much difficulty once phrases are delineated clearly. Thus, unifica-
tions reveal material and this material’s variations for different sections in a work,
and they help reveal that work’s form.
In order to match patterns that resemble but do not duplicate one another, I use
controllers, variables set to numbers representing thresholds defined by their
names. For example, an *intervals-off* controller contains the number of al-
lowable incorrect intervals before the pattern matcher considers two patterns un-
equal. Figure 5.15 presents an example of how this controller works with two
patterns presented, a target pattern (a) and a potential matching pattern (b). If the
*intervals-off* controller is set to 2, the two patterns will not match. However,
if the *intervals-off* controller is set to 3, the two patterns will match. Many
controllers can exist simultaneously in such situations, controllers such as
*interval-amount-off*, a controller that defines how much an interval can dif-
fer in order not to cause even one incorrect interval to produce a failed match.
A good music pattern matcher typically contains as many as a dozen or more
such controllers in order to allow, say, tonal variations to pass easily, rejecting vari-
ations that do not conform to typical tonal alterations. Given enough time and ex-
perimentation, most works, even very complex contrapuntal works, will reveal
their formal designations in this way. In particularly problematic circumstances, I
use several small programs (outlined in Cope 2000, chapter 5) in concert with
those described here.
Obviously, discovering form represents a greater problem with some works than
I have described here. The processes for discovering such forms, however, remain
the same. The difficulties must be overcome by rephrasing the work under analy-
sis, varying controller levels, or some combination of these two processes.
One of the ways to help discover structure in music, as opposed to form, in-
volves meta-patterns, patterns whose members occur noncontiguously (see Cope

a. b.
3 3
bœ œ #œ œ. bœ nœ œ bœ œ bœ œ. œ nœ
&c œ œ œ œœ

FIGURE 5.15 Two patterns, a target pattern and a potential matching pattern. If
the *intervals-off* controller were set to 2, the two patterns
would not match. However, if the *intervals-off* controller were
set to 3, the two patterns would match.
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 221

FUNCTION AND STRUCTURE IN POST-TONAL MUSIC 221

1996). Meta-patterns are discovered by hierarchical pattern-matching processes

that utilize unique pattern matchers. This hierarchical pattern matcher reduces the
actual number of matchings necessary by first evaluating foreground material in
terms of importance and using only those pitches and intervals that surpass cer-
tain thresholds.
The discussion of pattern matching in Chapter 2 of this book provides general in-
sights into the usefulness of pattern matching for music analysis. However, for
more intricate pattern-matching processes—such as those required of formal analy-
sis and the discovery of meta-patterns—using object-oriented programming and
controllers greatly refines the process. In the first section of this chapter, we
learned how to store data in objects, name them according to how best to locate
them, and, whenever needed, access them by referencing their names rather than
storing them in long lists and then searching for them. This process works as well
with locating a particular stored pattern as it does for groupings. By locating cer-
tain patterns and then storing them in well-named objects, they become available
in exactly the same way.
The pattern-matching process that produces meta-patterns resembles what Ex-
periments in Musical Intelligence uses to find patterns with interpolated pitches.
Figure 3.7 of Experiments in Musical Intelligence (Cope 1996, 89) shows how Chopin
buries repetitions of his themes in extremely ornamented variations. In such cases,
two highly varied patterns can match owing to a high setting of the controller des-
ignated to allow interpolated notes. Although effective for matching patterns in
music by composers who, like Chopin, use such ornamental techniques for varia-
tion, a high setting of the interpolation variable can also allow all manner of un-
wanted patterns—noise—to enter into the analysis of music of composers who do
not use such techniques. Hierarchical analysis, on the other hand, does not depend
on the approach composers use to vary their music, but rather pattern matches
music and derives its inherent meta-pattern.
Unlike formal analysis, structural analysis does not rely on material repetition,
variation, and/or contrast. Structural analysis attempts to reveal the architecture of
the music, the harmonic and melodic superstructure. The hierarchical processes I
use resemble those used in Schenkerian or layer analysis (Schenker 1979). Typi-
cally, however, Schenkerian analysis follows middleground patterns of descending
scale degrees. My programs make no such assumptions about the music they ana-
lyze, but simply assign various tensions to pitch groupings and then apply filter
thresholds. Such filter thresholds typically produce results that, on the one hand,
have more pitches than those usually provided by Schenker-type middlegrounds,
but on the other hand, fewer notes than the foreground or surface of the pattern
being analyzed.
Figure 5.16 provides an example of this process. Figure 5.16a shows a simple
phrase of music composed by Mozart (the theme of the first movement of the Pi-
ano Sonata in C Major, K. 545). Figure 5.16c presents the meta-pattern revealed by a
hierarchical pattern matcher. This meta-pattern surfaces from the matching
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 222

222 HIDDEN STRUCTURE

process by virtue of its agogic tensions, placement in the metrical structure, and
relevance to the key or scale in use (not unlike the processes described earlier in
this chapter for defining function). The various patterns in the music in Figure
5.16a have lines connecting them to their correlates in newly composed music (Fig-
ure 5.16b). The meta-pattern in Figure 5.16c then demonstrates how, through the
composition process, the fragments reorganize while concomitantly retaining much
of their original coherence. The meta-pattern serves as a guide for a simple com-
posing program. Although only half the pitches fall in the same place in their re-
spective measures in Figure 5.16b, the variation of the original succeeds because
the meta-pattern occurs so prominently in both phrases. This obviously simplistic
example nonetheless demonstrates the basic principles of deriving and success-
fully modeling meta-patterns.
Software filters can help to reveal structure in music by discarding pitches dur-
ing analysis that for one reason or another do not survive certain tests—tension,
duration, and so on. These tests in many ways resemble structural analysis in tonal
music, where certain pitches do not survive middleground analysis (see Cope
1991). In structural analyses of tonal music, these tests include scale degree impor-
tance, implied or explicit function, register, phrase placement, and so on. The no-
tion of filtering out less important pitches in order to highlight more important
ones is the basic principle of structural analysis. Such filtering of post-tonal music
can. I believe, produce equally rewarding structural maps. For example, filtering
could easily include tests of varying degrees of severity such that a wide range of
Schenker-like foregrounds, middlegrounds, and backgrounds could emerge.

a.

&˙ œ ˙ œ œ œ œ
œ œ. œ œ ˙ ˙

b.

&˙ œ œ œ ˙ œ œ
œ. œ œ œ ˙ ˙

c.

&w w w w

FIGURE 5.16 A phrase of music by Mozart (a); a machine-composed replication

(b); and the meta-pattern that binds them together (c).
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 223

FUNCTION AND STRUCTURE IN POST-TONAL MUSIC 223

Applying five simple but useful filtering elements—duration, dynamic, register,

metric placement, and scale membership—can provide useful and possibly mean-
ingful gradient maps of post-tonal music, as presented in Figure 5.17. Figure 5.17a
represents unanalyzed music and Figure 5.17d a background as realized by the
filters imposed. The intervening maps show increasing note removal produced by
filters imposing ever-harsher restrictions on the music of the previous maps. The
structural mapping software available on the CD-ROM accompanying this book pro-
duced this map and requires only the use of the scale software from Chapter 3 in
order to analyze music in the manner described here. The function call and note-
event output that produced the musical notation in this figure appear at the bot-
tom of the example.
SPEAC analysis at the structural level informs analysts that repeated material
need not necessarily serve the same function as the material it repeats. For exam-
ple, material initiating a work will most likely provide a statement (S). However, this
same material recognizably returning at a later point in this same music may be an
extension (E), preparation (P), anticipation (A), or even consequence (C), depend-
ing on its role in the structure. A SPEAC analysis at a structural level represents
music that contains many smaller SPEAC functions.
Averaging lower SPEAC functions to derive an overall tension and then using this
more inclusive formal SPEAC identity in competition with the required syntax of its
own level produces the most effective analysis. Thus, a structural level (S) may ac-
tually represent music that at a lower level has a predominantly (A) identity. Al-
though this may seem contradictory, various hierarchical levels in music can serve
very different purposes, just as in language (see Cope 2000).
Many other approaches to form and structure have been attempted, often with-
out the aid of computers. For example, Joseph Straus comments that Schoenberg’s
op. 33a reveals a form (and structure of a type) defined by the transpositions of its
primary pitch-class set:

The large-scale progression of the piece, then, is A0–A2–A7–A0. Obviously Schoen-

berg has in mind some kind of analogy to the tonal motion I–II–V–I. But there is
more than an analogy at work here. Look again at the first three notes of the
series—they form set class 3-9 (0,2,7). The set of areas A0–A2–A7 thus reflects the
initial set of pitch classes. The large-scale progression of the piece composes out
its initial melodic idea. (2005, 259)

Straus’s approach very much resembles a meta-pattern as described in this chapter.

Although my own discussions of form and structural analysis here barely
scratch the surface of what can and will be accomplished with the aid of comput-
ers, the processes I have described represent strategies that have proven quite
useful in my computer-composing programs such as Experiments in Musical In-
telligence (see Cope 1996 and 2001, in particular) and Alice (see Cope 2000 and
accompanying software). Therefore, although I am not claiming any degree of
thoroughness for these techniques, I do believe that pattern matching in the ways
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 224

224 HIDDEN STRUCTURE

a. bœ.
œ Œ J
& c Œ # # œœ œJ ‰ Œ ∑
Piano

? c Ó. # ## œœœœ Œ œ b Rœ ≈ ‰ Œ œ
œ
b. bœ.
œ Œ J
& Œ # # œœ œJ ‰ Œ ∑

? Ó. # ## œœœœ Œ
œ
Ó
œ
œ
c. bœ.
& ∑ ∑ Œ J

? Ó. # # œœœ Œ
œ
Ó ∑

d. bœ.
& ∑ ∑ Œ J

? Ó.
# œœ
∑ ∑
? (structural-levels 4 schoenberg)
(((1000 61 500 1 49) (1000 63 500 1 49) (2000 71 500 1 49)
(2000 77 500 1 49) (3000 51 1000 1 49) (3000 56 1000 1 49)
(3000 61 1000 1 49) (3000 62 1000 1 49) (5000 59 1000 1 49)
(6000 58 250 1 49) (8000 48 1000 1 49) (8000 59 1000 1 49)
(9000 87 750 1 49))
((1000 61 500 1 49) (1000 63 500 1 49) (2000 71 500 1 49)
(2000 77 500 1 49) (3000 51 1000 1 49) (3000 56 1000 1 49)
(3000 61 1000 1 49) (3000 62 1000 1 49) (5000 59 1000 1 49)
(8000 48 1000 1 49) (8000 59 1000 1 49) (9000 87 750 1 49))
((3000 56 1000 1 49) (3000 61 1000 1 49) (3000 62 1000 1 49)
(5000 59 1000 1 49) (8000 59 1000 1 49) (9000 87 750 1 49))
((3000 61 1000 1 49) (3000 62 1000 1 49) (9000 87 750 1 49)))

FIGURE 5.17 Various levels (gradient map) of music revealed by applying filters.
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 225

FUNCTION AND STRUCTURE IN POST-TONAL MUSIC 225

described here will prove valuable as prototypes for more complex and encom-
passing programs to come.

MUSICAL EXAMPLES
The structural graph (a result of using the Structure Map program described in
more detail in the next section of this chapter) partially shown in Figure 5.18 of the
opening of the third movement of Stravinsky’s Three Pieces for String Quartet
(1914) presents a dense and complicated set of connections that produces equally
complicated connections between one node (pitch) and all other nodes (pitches).
The information of the initial lowest pitch of this piece is displayed in a separate
window, the result of clicking on the relevant node. The scroll bar to the right in
this rectangle suggests that there is a great deal more information about the rela-
tionships between this node and all other nodes in this passage. Even with the
small amount of information shown, this window within the graph presents nine-
teen separate tensions for closely associated pitches, with event-2 having the most
tension (0.6) in relation to the initial low pitch. Looking back at the music of this
passage, shown in Figure 1.19 we see that this low D-natural has the most tension
relative to the following D-flat.
Interestingly, the opening passage of the third movement of Stravinsky’s Three
Pieces for String Quartet offers as much tension with its often chromatic voice-
leading as that created by vertical groupings. This voice-leading tension creates
a very difficult problem for harmonic analysis. However, applying the [0,2,3,4,5,7,8,9,t,e]
C-based scale first described in Chapter 4 of this book produces
1ø 6ø
0 2 3 4 5 7 8 9 t e
which itself presents a very interesting root analysis, as shown in Figure 5.19. Note
here that all of the roots occur within the principal scale, that pitch class 1 occurs
often in the bass voice of the violoncello part, and that the single pitch class 6 oc-
curs in m. 4 of the viola part.
A simple SPEAC analysis of this same opening passage from the third movement
of Stravinsky’s Three Pieces for String Quartet, shown in Figure 5.20, contains no
statements or extensions, with a predominance of anticipations and consequences
and with an occasional preparation, especially near the beginning of the passage.
The tensions provided above each column confirm the analysis with tensions rang-
ing from 1.4 to 3.3.

PROGRAM DESCRIPTION
The Root program on the CD-ROM that accompanies this book returns the root of
any grouping of pitches provided as an argument. This simple five-function program
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 226

226 HIDDEN STRUCTURE

FIGURE 5.18 The opening few beats of a graphical analysis of Stravinsky, Three
Pieces for String Quartet (1914), 3rd movement.
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 227

FUNCTION AND STRUCTURE IN POST-TONAL MUSIC 227

À & 45 46 œ b œ 45
V1
b˙ œ œ œ œ bœ ˙ œ œ ˙ œ
π
V2 & 45 46 45
b˙ œ œ œ œ bœ ˙ œ œ bœ œ ˙ œ
π
B 45 œ nœ bœ œ 46 œ 45
Va
b˙ œ b˙ nœ œ bœ ˙ nœ
π
Vc
à ? 45 π˙ bœ œ œ bœ œ ˙ œ 46 b œ œ œ ˙ œ 45
3 2 3 0 2 3 0 3 2 3 0 0 3

À & 45
4
≤ ≤ ≤ ≤
C œ œ œ œ
3
ϲ w
œ bœ œ œ œ- ˙. ˙ F p
3
≤ ≤ ≤ ≤ ≤
& 45 C œ œ œ œ œ w
œ bœ œ œ œ- ˙. ˙ F p
3
≤ ≤ ≤ ≤ œ≤ w
B 45 C œ œ œ œ
bœ bœ œ n œ b œ- b˙. ˙ p
F
≤ ≤ ≤ ≤ b œ≤
3
? 45 œ C #œ œ œ œ #w
à œ bœ œ œ- ˙. ˙
F p
2 3 2 3 0 0 2 2 2 4 T 2

FIGURE 5.19 A root analysis of Stravinsky, Three Pieces for String Quartet (1914),
3rd movement.

first reduces pitches to intervals and then calculates their individual stengths based
on a series of strengths found in the variable called *interval-strengths*.
The Root program can be used with any software that requires roots for function
determination.
The Structure Map program returns selected note-events that the program has
determined represent the music analyzed on ever-higher structural levels. Users
can determine for themselves how many levels they wish to view. Structure Map re-
quires a separate program from Chapter 4 that returns the scale in use to help the
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 228

228 HIDDEN STRUCTURE

FIGURE 5.20 SPEAC analysis (by chord change) of Stravinsky, Three Pieces for
String Quartet (1914), 3rd movement.

structural functions determine the various levels. Structure Graph creates struc-
tural graphs of segments of music, each node of which, when clicked, produces a
window containing important information about the note which it represents. As an
example, Figure 5.21 shows a structural graph of a section from the author’s
Triplum for flute and piano. All the pitches in this passage appear in this graph,
with the circled notes representing pitches that the program has found structurally
important. Although standard functional code produced the structural information
here, the graphic output results are almost entirely from OOP.
The three pitches circled in the graph in Figure 5.21 result from precise cumula-
tive tensions that, again, appears when one clicks upon the circled pitch. These cir-
cled pitches do not represent a tonal center, but rather the strength of the pitch
based on the various parameters in the internal tension less the multiple of the
constant of the average of the relational tensions, as discussed earlier in this chap-
ter. Combined with the other parameters influencing the tension of pitches, these
high-tension pitches arguably represent important structural pivots in the post-
tonal music being analyzed. Computing the tensions of all of the pitches in even a
simple one-line phrase poses such extraordinary obstacles that I would not attempt
it—nor, perhaps, even have conceived of the process—without access to computer
software such as Structure Map.

CONCLUSIONS
The functional and structural programs presented in this chapter, along with the ba-
sics of object-oriented programming, provide a frame for the chapter that follows—
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 229

FUNCTION AND STRUCTURE IN POST-TONAL MUSIC 229

FIGURE 5.21 A structural graph of a section from the author’s Triplum for flute
and piano.

a chapter concerned with, first, building models of short segments of music as con-
tinuations of existing works, and, second, creating complete musical works in the
style of music under analysis. Although the software described in the current chap-
ter does not directly contribute to either of the modeling processes described in
Chapter 6, this software serves an important role in comparing the simulations of
Chapter 6 to the originals and thus helping to verify the success or failure of the
modeling processes.
06_DAS_Chap5_pp189-230 1/29/09 1:35 PM Page 230
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 231

SI X
Generative Models
of Music

A computer is an ideal instrument by means of which analytical ideas can be

tested, since the investigator starts with certain hypotheses from which he
formulates operating principles; he supplies this information to the computer;
the computer then generates music based upon these principles; and the investi-
gator then analyzes the results to further his investigation. This, of course, is
essentially nothing more than a standard example of experimental scientific
method, but the unusual thing is that computers provide a practical experimen-
tal technique for carrying out such research in the musical field.

Lejaren Hiller and Leonard Isaacson, Experimental Music (1959), 166

Music theories and even resultant analyses can appear logical and effective, but
unless they can produce output similar to the music under analysis, they lack a cer-
tain validity. In this chapter, I will describe several methods for using probabilities—
implicit and explicit—to analyze music and then, using that analysis, create com-
plete compositions roughly equivalent—at least stylistically—to the analyzed music.
I will also describe a program that can compose missing portions of the input mu-
sic exactly. This process of regenerating music from its analysis—similar to the
concepts of algorithmic information theory presented in Chapter 2 of this book—
may indicate, at least to a degree, more than just the success or failure of a theory;
it also may indicate many of the subconscious techniques that composers actually
use in creating their music. Most of this chapter therefore deals with Principle 4 as
discussed in Chapter 1 of this book.

MODELING
Throughout the common-practice period in Europe, young composers often learned
their craft by modeling recognized works. Established masters typically assigned
their apprentices the creation of new music following the classics of the past. Even
when composers did not explicitly study their craft, their personal styles often re-
vealed the music they used for models. For example, it would be hard to imagine a
J. S. Bach without an Antonio Vivaldi, a Wolfgang Mozart without a Joseph Haydn, a

231
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 232

232 HIDDEN STRUCTURE

Ludwig van Beethoven without a Wolfgang Mozart, and so on. It seems no less logi-
cal to use these same modeling techniques as an approach to music analysis.
Modeling offers analysts the ability to view music during its composition and
thus the opportunity to test a work in progress as well as when complete. Steven
Wolfram (2002) and many others have postulated that such modeling, rather than
reverse engineering, represents the better way to accurately understand how
things work. Wolfram uses dynamic two-dimensional cellular automata to produce
visual models to prove that extremely complex output can result from extremely
simple rules (see Cope 2005, 68, figure 3.6 for examples). Wolfram comments that
much of the emphasis over the past century or so has been on breaking systems
down to find their underlying parts, then trying to analyze these parts in as
much detail as possible . . . but just how these components act together to pro-
duce even some of the most obvious features of the overall behavior we see has
in the past remained an almost complete mystery. (Wolfram 2002, 3)

and
it has seemed that whenever one tries to get to another level of accuracy, one en-
counters more complex phenomena. And at least with traditional scientific intu-
ition, this fact suggests that models of progressively greater complexity will be
needed . . . could it even be that underneath all the complex phenomena we see
in physics there lies some simple program which, if run for long enough, would
reproduce our universe in every detail? (Wolfram 2002, 465)

Like a salmon swimming upstream against a current too strong for it, Wolfram sug-
gests, we are currently waging a losing battle in attempting to understand the
universe by taking it apart and analyzing its individual components. Science, ac-
cording to Wolfram, should rather create computer models of the universe, choose
the one that achieves the closest resemblance to reality, and study the principles of
our modeling processes. I believe the same to be true of understanding music.
John Holland comments that “although model building is not usually considered
critical in the construction of scientific theory, I would claim that it is” (Holland
1998, 4). Although Holland refers here primarily to the sciences, I believe that his
words have as much use for music analysts. He adds that
in the beginning were rule-bound sacrifices to the gods—we modeled the world
in terms of personalities and ways of propitiating those personalities. Later, we
discovered mechanisms (gates, pumps, and wheels) and ways of using them to
control parts of the world, and we began to model the world with mechanisms
instead of personalities. Eventually, we arrived at complex computer-controlled
devices and models, and scientific models that employ abstract mechanisms.
(Holland 1998, 10)

The Melody Predictor software found on the CD-ROM accompanying Cope 2000
represents a musical modeling program that extends simple tonal melodies. Melody
Predictor has only three basic rules in its algorithm:
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 233

GENERATIVE MODELS OF MUSIC 233

1. If the preceding three pitches represent a scale, continue the scale in the same
direction.
2. If the preceding pitch represents a leap from its preceding pitch, either continue
leaping in the same direction following triadic formation, or fall back a scale step
in the opposite direction based on pitch motion previous to the leap.
3. If the preceding three pitches represent a pattern, continue the pattern including
repeated pitches.
Melody Predictor uses the scale of a piece under analysis and does not predict
chromatic pitches.
Figures 6.1a and 6.1c show two melodies given to the predictor, the initial
themes of Mozart’s Piano Sonatas K. 279 and K. 533. Figure 6.1b and d shows the re-
sults of the program’s attempt to predict the pitches Mozart chose. Of its guesses,
64 and 71 percent correctly match. These percentages are quite high when com-
pared to the percentages—between 0 and 12 percent—achieved by the program
when attempting to match randomly created tonal melodies.

a.
œ œ œ œ
‰ œ œ œ œ œ œ œ
œ œ œ œ œ œ œ œ œ œ œ
&c œ œ œ

b.
œ œ œ œ œ œ œ œ œ
&c œ ‰ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ

c.

& c Ó. œ œ œ œ œ œ œ œ œ œ
œ
˙ œ œ œ œ œ œ œ œ w

d.

& c Ó. œ œ œ œ œ œ œ œ œ œ œ
˙
œ œ œ œ œ œ œ œ w

FIGURE 6.1 (a) Mozart, Piano Sonata K. 279, 1st movement, mm. 1–3; (b) results of
the Melodic Predictor; (c) Mozart, Piano Sonata K. 533, 1st movement,
mm. 1–4; (d) results of the Melodic Predictor.
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 234

234 HIDDEN STRUCTURE

Although approaching accuracy in the case of the Mozart examples, especially

given the simplicity of the program, this accuracy does not prove either that
Mozart was a predictable composer or that by extending the program it could com-
pete with Mozart’s melodic composing skills. The Melody Predictor’s results do
demonstrate, however, that Mozart used rather straightforward rules to compose
his melodies and that his pitch choices were often based, at least in part, on the
rules listed above. Although we might rather imagine that each pitch of Mozart’s
music was an inspired original, the fact remains that a great deal of his music, like
that of most of his contemporaries in the classical period, follows algorithms, and
that these algorithms produce relatively predictable results. The extraordinary na-
ture of Mozart’s musical genius, of course, arises from those occasions when he
avoids the routine choices.
Many analysts believe that composers compose and analysts analyze. However,
one of the most effective ways for analysts to understand the music they study is
to use their theories to compose music and compare this music to the originals.
When this comparison fails, theories lose much if not all of their validity. Over the
past twenty-five years or so, I have attempted to verify my theories of music analy-
sis by creating programs that produce demonstrable results. Experiments in Musi-
cal Intelligence represents such a program. The Extend program described later in
this chapter represents another example of this kind of verification process.

RECOMBINANCY
The thousand or so compositions produced by Experiments in Musical Intelligence
can be reverse engineered. These works have a kind of compositional provenance.
In fact, Virtual Music (Cope 2001) provides a full description of the creation of a
machine-composed sonata movement in the style of Mozart. The processes de-
scribed in that book follow a principle known as recombinancy, a principle I will
briefly describe here. I will also point readers to several more complete descrip-
tions presented elsewhere (see Cope 1990, 1991, 1996, and 2005). Interestingly, the
compositional processes used by Experiments in Musical Intelligence do not resem-
ble any of the analytical processes presented thus far in this book. Although com-
position and subsequent analyses need not match one another precisely, analytical
processes that have no relation to their associated compositional processes seem
more contradictory than complementary.
When I began Experiments in Musical Intelligence in 1981, having an intermediary
—myself—form abstract sets of rules for composition seemed artificial and unneces-
sarily premeditative. Also, having to code new rules for each new style or work
proved daunting. I therefore set out early on to create new output derived from
music stored in a database. The concept—of first deriving rules from a database of
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 235

GENERATIVE MODELS OF MUSIC 235

music and then using those rules to compose new music—seemed logical then and
still does. As with Wolfram’s approach to discovering the rules of the universe and
then modeling those rules to prove their robustness, my programs have created
musical compositions that hopefully stand as testimony to the validity of the
processes used to create them.
I argue in my book Experiments in Musical Intelligence (1996) that recombinancy
appears everywhere as a natural evolutionary and creative process. The recombi-
nation of atoms, for instance, produces new molecules. Complex chemicals derive
from the recombination of more rudimentary particles. Humans evolve through ge-
netic recombination and depend on recombination for communication, because
language itself results from the recombining of letters, words, and sentences. Music
is no different: most of the great works of Western art music result from recombina-
tions of the twelve different pitches of the equal-tempered scale, as well as resul-
tant groupings of those pitches. The secret of creativity lies not in the invention of
new letters or words for language, or of pitches or groups of pitches in music, but
in the subtlety and elegance of their recombination. Musical recombinancy can
thus be defined as a method for producing new music based on the recombination
of elements of extant music.
Simply breaking musical works into smaller parts and randomly combining them
into a new order, however, will almost certainly produce nonsense. Effective recom-
bination requires extensive musical analysis and very careful recombination to be
effective. Therefore, my first attempts at recombinant analysis and composition be-
gan by using music with which I had familiarity—Bach chorales (Riemenschneider
1941). My initial programs grouped this music into beats and saved these beats as
OOP objects. I had this program also store the name of the pitch to which each
voice moved in the subsequent beat (the destination pitch). I further had the pro-
gram collect these beats into groupings called lexicons, delineated by shared
pitches and registers of their entering voices, as in C1–G1–C2–C3, with the arabic
numerals referring to the octaves in which the pitches appear. To compose, then,
the program simply chooses the first beat of any chorale in its database, examines
this beat’s destination pitches, and then selects one of the stored beats with those
same first pitches, assuming enough chorales have been stored to make more than
one choice possible. New choices then create the potential for different offbeat mo-
tions and different destination chords while maintaining the integrity of Bach’s orig-
inal voice-leading rules.
A simple pseudo-code example of the above processes might look like this:

(defun compose-recombinantly (many-works-in-note-events)

(let* ((grouped-music (group-music many-works-in-note-
events))
(music-objects (objectify-the-music grouped-music))
(lexicons (put-groupings-into-object-lexicons music-objects))
(composition (compose lexicons)))
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 236

236 HIDDEN STRUCTURE

where let* acts as a kind of binding function wherein each of its listed arguments
consist of a variable name (e.g., grouped-music) followed by functions and
their arguments, defined elsewhere. The asterisk following the let here allows
for previously bound variables to be used within the same call to let as in
grouped-music used in the binding of music-objects. In short, compose-
recombinantly groups the music, puts these groups into objects, places these
objects into appropriate lexicons, and then uses these lexicon-stored grouped
objects for recombinant composition.
Figure 6.2 demonstrates how this recombining process can lead to the comple-
tion of a new Bach-like chorale phrase. Each of the new grouping choices here is
absolutely correct in terms of chord-to-chord voice-leading, but without the pro-
gram having to initiate any user-supplied rules. The music, a product of very sim-
ple syntactic networking, inherits the voice-leading rules of the works upon which
it bases its replications. In effect, the program extracts its data’s algorithms and
then uses these algorithms to create new instances of stylistically faithful music.
Although I do not think that Bach composed his chorales by recombinantly
assembling them from parts of previously composed chorales, clearly his voice-
leading rules led him to revisit certain voicings time and time again. One can certainly
imagine him remembering previous solutions and either reiterating those solutions
or deliberately attempting to avoid them to create more original-sounding music.
Whatever his exact process, however, Bach’s voice-leading recombinancy at least
partially agrees with the one Experiments in Musical Intelligence uses for composing.
Most music does not consistently move as Bach chorales do, in four-voice tex-
tures with similar character and few rests. Experiments in Musical Intelligence,
therefore, analyzes character in the original music before breaking this music into
beat-size groupings. This character analysis ensures that changes of musical char-
acter take place at reasonable locations during recombinancy. The program then
stores representations—typically rhythmic and textural values—of this musical
character in each grouping object so that continuity can be maintained in output.
Having information about an originating work’s beat-to-beat character, for example,
allows the program to produce works that have unifying elements. One can imagine
that if the program were not able to access this kind of information, many beats of
a resulting work would have different types of internal patterns, suddenly shifting
from one musical context to another (see Figure 6.3). This lack of unity would most
likely be uncharacteristic of either the style or the intent of the original music. Its
stored character attribute allows the program to produce works having more logic
and continuity than would otherwise be the case.
Even with the textural continuity just described, music composed in this way of-
ten wanders, with very unbalanced and uncharacteristic phrase lengths. No real
musical logic exists beyond chord-to-chord syntax. Further, phrases are simply
strung together randomly, without any large-scale form or structure, usually in a
single key, and without any sense of the types of repetition and development neces-
sary for intelligent music composition.
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 237

GENERATIVE MODELS OF MUSIC 237

102 137 322 159 128 183

1/1 1/3–4 9/4 10/2 11/1–3 11/3

# œœ
& c œœ œœ œ œ œ œœ œœ œœ œ œœ
J J
œ œ.
Piano
œ œœ œ œ œ œ
? # c œœ œœ
œ œ œ œ œ
œ œ

FIGURE 6.2 Recombination process leading to the completion of a new Bach-like

chorale phrase.

œ œ œ œ œ
a. b. c.
˙˙
&c œ œ œ œ œ ˙ ˙˙
˙
œ œ ˙˙
˙
œœ œœ œœ œœ œœ
Piano
œ œ œ œ œ œ
?c œ œ œ œ œ œ œ œ œ œ œ
œ

FIGURE 6.3 An example of differing types of internal patterns causing problems in

recombination: (a) and (b) originals, (c) a recombination without re-
gard to texture sensitivity.

In order to avoid such wandering, Experiments in Musical Intelligence also

makes formal and structural aspects of the music inherent in its database. This in-
volves separate analyses so that the program can store formal and structural infor-
mation with each grouping along with destination pitches and character. For
example, Experiments in Musical Intelligence analyzes the original distance to ca-
dence, the position of the grouping in relation to meter, and other formally and
structurally sensitive features. Each new substitution must now meet criteria in re-
lation to destination pitches of each voice, cadence, and other phrase-dependent
factors.
Having programs also compare the locations of cadences one to another helps
to produce effective phrase and section endings over the course of an entire com-
position. Using a single work in the database as a model helps to produce more
stylistically faithful recombinant multi-phrase music creations. Of course, retaining
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 238

238 HIDDEN STRUCTURE

too much information with each grouping causes the program to simply reiterate
one of the works in its database. Therefore, whenever possible, the program
chooses new possibilities rather than old ones. Providing very large databases also
helps guarantee that more than one possible correct grouping exists at each recom-
binant point of choice.
The process of discovering thematic areas, their similarities and differences, and
ensuring that such relationships transfer in principle to output music requires
several sub-programs. In summary, however, formal analysis relies on pattern
matching models in a database to indicate when themes return, vary, or are con-
trasted with other themes (see the discussion of unifications and meta-patterns in
Chapter 5). Once located, these themes are labeled and then classified in as many
ways as possible—scales in use (see Chapter 4), length, tessitura, and so on. This
composing program then attempts to emulate this formal analysis in new music
(for more detail on these processes see Cope 1996, 2000, 2001, and 2005).
Many examples of Experiments in Musical Intelligence’s output appears in my other
books (e.g., Cope 2001), on the Internet, and in publication (e.g., spectrumpress.com).
Figure 6..4 presents the beginning of the Rondo Capriccio for cello and orchestra ar-
guably in the style of Mozart (Cope 2006). This work resulted from a database of
various divertimentos, serenades, violin concertos, and so on, rather than works
by Mozart for cello and orchestra, because Mozart did not compose any works for
cello and orchestra. This example demonstrates the elements discussed above:
consistency of texture (character), formal thematic statement, structural integrity,
and so on. Though not an example of post-tonal recombinancy, Figure 6.4 nonethe-
less presents a good model of strict recombination as described here and else-
where (Cope 1991 and 1996). Like the Experiments in Musical Intelligence Mozart
sonata described in Virtual Music (Cope 2000), this work too could be analyzed for
its constituent parts and each part traced to a particular passage in a particular
work composed by Mozart. Of course, Mozart’s works can be reverse engineered in
much the same way as I describe in several books (see in particular Cope 2005,
149).
Reverse engineering the music of Experiments in Musical Intelligence can pro-
duce quite valuable results. At the least, such analysis would prove or disprove
the numerous criticisms that this music either sounds too little or too much like
the originals on which it is based. Discovering the differences between computer-
generated music and human-composed music might help us to reevaluate our un-
derstanding and appreciation of the latter. To date, very little serious research
exists on the relative merits of computer-composed replications. It would, for ex-
ample, seem natural for anthologies of music designed for students of music theory
and musicology to encounter some of this music in order to clearly demonstrate
their knowledge of both style and theory of the music upon which these computer-
created works are modeled.
Such reverse engineering is not simple, however. One must be aware of the com-
plete possibilities for recombination and equally aware of where the particular recom-
binant bits may originate. Of course, computers can aid analysts in reconstructing
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 239

GENERATIVE MODELS OF MUSIC 239

Rondo Capriccio
For Cello and Orchestra

Flute
À& q = 96
c ∑ ∑ ∑
Emmy Mozart
∑

Oboe & c ∑ ∑ ∑ ∑

#
Clarinet 1, 2 & # c ∑ ∑ ∑ ∑
in B b

Bassoon 1, 2
Ã? c ∑ ∑ ∑ ∑

# c ∑ ∑ ∑ ∑
Horns 1, 2 &

À? c ∑
tutti
‰ œ. œ. œ. ‰
Solo Cello
œ. œ. œ. ‰ œ. œ. œ. ‰ œ. œ. œ. ‰ œ. œ. œ. ‰ œ. œ. œ.
F
c Ó Œ j j œ œ
Violin 1 & œ. œ- œ œ œ œ ˙ œ. œ . . œ. œ.
œ -
f
& c ∑ ‰ ‰ ‰ ‰ ‰ ‰
œ. œ. œ. œ. œ. œ.
Violin 2
œ. œ. œ. œ. œ. œ. œ. œ. œ. œ. œ. œ.
F
B c ∑ ‰ œ œ œ ‰ œ. œ. œ. ‰ ‰ ‰ ‰
Viola
. . . œ. œ. œ. œ. œ. œ. œ. œ. œ. œ. œ. œ.
F
Cello/Bass
Ã? c ∑ ‰ œ. œ. œ. ‰
F œ. œ. œ. ‰ œ. œ. œ. ‰ œ. œ. œ. ‰ œ. œ. œ. ‰ œ. œ. œ.
FIGURE 6.4 An example of Experiments in Musical Intelligence output: the begin-
ning of the Rondo Capriccio for cello and orchestra arguably in the
style of Mozart.
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 240

240 HIDDEN STRUCTURE

Fl.
À&
5
∑ ∑ ∑ ∑

Ob. & ∑ ∑ ∑ ∑

#
Cl. 1, 2 & # ∑ ∑ ∑ ∑
in B b

Bn. 1, 2
Ã? ∑ ∑ ∑ ∑

# ∑ ∑ ∑ ∑
Hn. 1, 2 &

À? ‰ œ. œ. œ. ‰ œ. œ. œ. ‰ œ. œ. œ. ‰ œ. œ. œ. ‰ ‰ ‰ ‰
S. Vc.
œ. œ. œ. œ. œ. œ. œ. œ. œ. œ. œ. œ.

œ. j œ j œ œœœœœ œ œ œ. œ œ. œ
& #˙ œ œ œœ œ œ. œ. œ œ
Vn. 1
œ

& ‰ ‰ ‰ ‰ ‰ œœ œ ‰ ‰ ‰
. . . œ. œ. œ. œ. œ. œ. œ œ œ
Vn. 2
œ. œ. œ. œ. œ. œ. œ. œ. œ. œ. œ. œ. . . .

B ‰ œ œ œ ‰ œ œ œ ‰ œ œ œ ‰ œ œ œ ‰ œ. œ. œ. ‰ œ. œ. œ. ‰ œ œ œ ‰
. . . œ. œ. œ.
Va.
. . . . . . . . . . . .

Vc./Cb.
Ã? ‰ œ. œ. œ. ‰ œ. œ. œ. ‰ œ. œ. œ. ‰ œ. œ. œ. ‰ ‰ ‰ ‰
œ. œ. œ. œ. œ. œ. œ. œ. œ. œ. œ. œ.

FIGURE 6.4 continued

07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 241

GENERATIVE MODELS OF MUSIC 241

Fl.
À&
9
∑ ∑ ∑ ∑

Ob. & ∑ ∑ ∑ ∑

#
Cl. 1, 2 & # ∑ ∑ ∑ ∑
in B b
1.
? ˙ ˙ w ˙ ˙
Bn. 1, 2
à ∑
P
# ∑ ∑ ∑ ∑
Hn. 1, 2 &

œ- . œ- œ œ œ œ ˙ œ. œ. œ. œ.
À? ‰
Solo
œ J œ. œ
J
S. Vc.
œ. œ. œ. Œ
f
& ‰ ‰ ‰ ‰ ‰ ‰
˙ œœœ œœœ œ œœœ œœœ
Vn. 1
˙ œœœ œ œ
F
& ‰ ‰ ‰ ‰ ‰ ‰ ‰ ‰
œœœ œœœ
Vn. 2
œ. œ. œ. œ. œ. œ. œœœ œœœ œœœ œœœ
F
B ‰
Va.
œ. œ. œ. ‰ œ. œ. œ. ‰ œ œ œ ‰ œ œ œ ‰ œ œ œ ‰ œ œ œ ‰ œ œ œ ‰ œ œ œ
F
Vc./Cb.
Ã? ‰
œ. œ. œ. ‰ œ. œ. œ. ‰ œ œ œ ‰ œ œ œ ‰ œ œ œ ‰ œ œ œ ‰ œ œ œ ‰ œ œ œ
F
FIGURE 6.4 continued
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 242

242 HIDDEN STRUCTURE

Fl.
À&
13
∑ ∑ ∑ ∑

Ob. & ∑ ∑ ∑ ∑

#
Cl. 1, 2 & # ∑ ∑ ∑ ∑
in B b

w w ˙ ˙ ˙ ˙
?
Bn. 1, 2
Ã
# ∑ ∑ ∑ ∑
Hn. 1, 2 &

œ. œ œ œ œ œ œ œ œ œ œ. œ œ. œ œ
À? #˙ œ œ œ œ œ œ. œ .œ œ
S. Vc. J J

‰ ‰ ‰ ‰ ‰ œœ œ ‰ œ ‰
Vn. 1 & œœœ œœœ œœ œ œœœ œ œ œœ œ‰ œœ œ

& ‰ ‰ ‰ ‰ ‰œ ‰ ‰ ‰
œ œ œœœ œœ œ œœ œ
Vn. 2
œœœ œœœ œœ œ œœœ

B ‰ œ œ œ ‰ œ œ œ ‰ œ œ œ ‰ œ œ œ ‰ œ œ œ ‰œœœ ‰ œœ œ‰
Va.
œœ œ

Vc./Cb.
Ã? ‰œ œ œ‰œ œ œ ‰œ œ œ ‰œ œ œ ‰
œœ œ
‰œœœ ‰ ‰
œœ œ œœ œ

FIGURE 6.4 continued

07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 243

GENERATIVE MODELS OF MUSIC 243

Fl.
À&
17
∑ ‰œœœœ
œ # œ œ œ œ œ œ œ #œ œ œ
œ Œ Ó
F
Ob. & ∑ ‰ œ œ œ œœ œœ œ œ œ œ œœ œœ œ Œ Ó
F
#
& # ∑ ‰ œ œ Œ Ó
œœ œœ œœ œœ œ # œœ œœ œœ œœ œœ œœ œœ œ #œœ œœ œœ
Cl. 1, 2
in B b

˙
F
˙ w w œ
Bn. 1, 2
Ã? w w œ Œ Ó
F
# ∑ w w œ Œ Ó
Hn. 1, 2 & w w œ
F
À? ˙ ˙
∑ ∑
tutti
œ œ œ œ
S. Vc.
œ œ œ œ
P
œ œœœ œœ œœ œ œ œ œ œ œ œ œ
div. unis.

& ‰ ‰ œ Œ Ó ∑ œ œœœ œœ œœ
œœœ œœœ
Vn. 1

F
œœœœœœœœ
Vn. 2 & ‰ ‰ Œ Ó ∑ œ œœœœœœœ
œœœ œœœ œ F
B ‰ œ Œ Ó œ œœœœœœœ œ œ œ œ œ œ œ œ
Va.
œ œ œ‰œ œ œ ∑
F
Vc./Cb.
Ã? ‰
œ œ œ‰œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ
œ œ œ œ
F
FIGURE 6.4 continued
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 244

244 HIDDEN STRUCTURE

sources in this way, as evidenced by my work with my computer program called Sor-
cerer (see Cope 2005).
One of the problems encountered in creating stylistically correct new music
based on existing pieces of music involves their allusions to other music. Such allu-
sions can cause problems for recombinancy by breaking these allusions into unrec-
ognizable smaller groupings. On page 175 of my book Computer Models of Musical
Creativity (Cope 2005) I discuss the opening measures of the third movement of
Beethoven’s Piano Sonata in F Minor, op. 2, no. 1, and account for every note as an
allusion to Bach, Mozart, or to Beethoven himself (figure 5.39 in that book). Each of
the fragments accounted for in earlier pieces is documented by composer, work,
movement, and measure number. Of course, such allusions represent only one part
of analysis:

Every work of music, unless it has been composed entirely by a formalism (and
possibly even then), contains within it many pointers to the musical culture that
helped to create it. These pointers, whether they be quotations, paraphrases,
likenesses, frameworks, or commonalities, help us to relate to that work, even if
we are hearing it for the first time. These pointers also point to other musical
styles and works that themselves have pointers, providing us with a rich history
of the cultural evolution of the work being heard. (Cope 2005, 175)

Considering allusions as compositional technique at one extreme (the music of

Ives and certainly the music of Experiments in Musical Intelligence, for example)
and pure formalism (isorhythmic motets and fugues, for example) at the other ex-
treme places most music in a more or less centrist position—combinations of allu-
sions and rules.
Discovering the sources of music, except in the most obvious cases, can prove
extremely difficult for analysts. Composers using formalisms can inadvertently pro-
duce allusions. Likewise, composers using allusions from very esoteric sources that
follow the rules of composition currently in play can make the discovery of these
allusions nearly impossible. The resultant mix of allusions and rule-created music
can produce incredible difficulties for those of us intent on using the proper tech-
niques for modeling the music we wish to better understand.
However, analyzing music for its quotations, paraphrases, and allusions to other
music would seem just as important to understanding music as attempting to de-
fine context-independent reductions by using, for example, roman numerals, pitch-
class sets, and letter representations of formal sections. Computer programs with
large databases of compositions have the potential to identify such quotes, para-
phrases, and allusions, for which we would otherwise have to rely entirely on ana-
lysts’ listening experiences. As mentioned earlier in this book, in Computer Models
of Music Creativity (Cope 2005) I describe the computer program called Sorcerer,
which returns a subset of possible references from a large database (e.g., see par-
ticularly Cope 2005, chap. 5).
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 245

GENERATIVE MODELS OF MUSIC 245

However, Sorcerer represents just a beginning to this kind of study. Programs fol-
lowing a similar design but with significantly larger databases should help deter-
mine potential sources for hermeneutic, semiotic, referential, and other types of
analysis that provide both semantic and syntactic information about the fundamen-
tal nature of musical compositions.
Figure 6.5 provides an example of allusion, part of a computer composition in
Beethoven’s style (a) followed by one of the originals upon which it was based (b).
The entire second movement of a recombinant Beethoven symphony appears as an
appendix in Cope 2005 and will give readers further opportunity to analyze the
structure, form, SPEAC, and other subjects discussed in this chapter in more detail.
Obviously, music does not exist in a vacuum. Discovering a German augmented
sixth chord in Bruckner does not carry the same weight as discovering a similar
German augmented sixth chord in Bach. Context provides an extraordinary frame
for even the simplest analytical discovery. Even the music we choose to analyze
has survived at least in part because of its context—idiomatic performability,
uniqueness in the repertoire, meaning of its title, life of the composer, and so on all
play important roles. How we analyze this music should be determined at least in
part by that context. For example, the German augmented sixth chord in Bach men-
tioned above will probably be analyzed as a series of nonharmonic tones, rather
than an individual chord, while Bruckner’s version will stand on its own as a Ger-
man augmented sixth chord. Because analysts cannot invent individual systems for
each work they analyze, they often depend on analytical processes that are fairly
devoid of these important individual contexts. Theorists such as Robert Gjerdingen
(1988) have attempted with some success to approach music analysis more from a
contextual point of view. However, far too often—and particularly with post-tonal
music—contextual relationships appear more often in program notes than in musi-
cal analyses.
Interestingly, it may be the context of the musical analysis—apart from the dis-
covery of musical allusions—that most effectively defies the computer’s ability to
assist human analysts. Although one could imagine a computer connected to the
Internet gathering mountains of potentially contextual information about a com-
poser or work, deciphering that information and siphoning out the data of actual
relevance to the music under analysis still requires human analysts. At the same
time, understanding musical provenance—meaning here the ties of a work to past
music—would seem as important to its understanding as discovering its functions
or other features. Computers make such studies much more feasible than depend-
ing on human memories of past music (see Cope 2005).
Judging new compositions that approximate but do not exactly match that of
the original music require some kind of criteria other than source identification.
Logic alone cannot evaluate the effectiveness of such musical solutions. The use of
the term musical here, a term with which most readers of this book are familiar,
is difficult to define. Webster’s Collegiate Dictionary (1991) defines musical as
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 246

246 HIDDEN STRUCTURE

David Cope - Experiments in Musical Intelligence

a. Allegretto
# . . . . .
œ. œ. œ. œ. œ œ œ œ œ. œ. œ.
& 43 œ . œ œ œ œ œ. Œ œ œ œ œ. œ. œ œ.
Œ œ œ
œ. . .
.
3
œ. œ. œ. œ. œœ. œœ. # œœ. œ. œ. œ. œ.
? # 43 Œ œ œ œ œ Œ Œ œ Œ Œ œ. œ œ.
.

. . . . .
8
# œ. œ. œ. œ. œ œ œ œ œ. œ. œ.
& œ œ œ. œ. œ œ œ. Œ œ . œ œ œ œ œ. Œ œ œ œ œ. œ. œ Œ
œ. œ œ
œ. . .
.
3
œ. œ. œ. œ. œœ. œœ. # œœ. œ. œ. œ. œ.
? # œ. œ .œ Œ œ œ œ œ Œ Œ œŒ Œ œ. œ œ.
. œ. œ. .

16
# œœœœœ
& œ œ œ œ œ œ œ˙ . œ œ œ œ˙ œ œ œ œœ n œœ œœ œœ œœ œœ œ ˙˙ œœ œ œ œ
˙. œ

? # œ. œ. œ œ œœœœœœ
œ. œ œ œ ˙ ˙ œ ˙ œ
œ œ
œ œ

23
# œ œ #œ œ œ
& œ˙ œ œ œ œœ n œœ œœ # œœ œ œœ Œ œ œ ˙˙
œ œ #˙ œ ˙ œ œ #œ œ

?# ˙ œ œ œœ œ œ œ œ œ #œ
œ œ œ #œ œ œ œ œœ ˙ œ ˙ # œ œ œœ œœ # œœ
œ

30
#
& ˙ œ #˙ œ ˙ œ œ #œ œ ˙ #˙ œ ˙ œ

? # œ #œ œ œ œ œ œ œ
œ œ # œ œ ˙ œœ ˙ # œ œœ œœ œœ # œœ œ # œ œ œ œ œ ˙ œœ ˙ # œ # œœ

FIGURE 6.5 An example of allusion in (a) an Experiments in Musical Intelligence

replication, and (b) a Beethoven bagatelle Op. 119, no. 1 (1820) upon
which it is partially based.
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 247

GENERATIVE MODELS OF MUSIC 247

b.
b3
&b 4
œ . œ œ. œ. œ. œ Œ œ . œ œ. œ. œ
. #œ Œ œ œ
œœ. œœ. œœ. œœ. œ. # œ.
Piano
œ œ
? b b 34 ∑ #œ Œ Œ œ œ œ Œ Œ

6
b #œ œ œ œ œ œ œ œ. œ
&b œ œ œ œ œ œ œ œ œ œ œ #œ Œ

? bb œ œ œ œ œ #œ œ Œ
œ #œ œ œ

FIGURE 6.5 continued

“melodious” and the word melodious as “musical,” offering little help. Attempting to
avoid a polemic, I offer a more quantifiable definition of the term musical:
The word musical means that, within the context of a particular piece of music,
logical, intuitive, and physical interpretations agree. Being logical infers the fol-
lowing of explicit rules. Being intuitive infers the following of implicit rules. Being
physical infers the following of natural physical laws (referring here to human
performability). A musical passage is therefore one in which the user of the term
finds all of the above criteria acceptable and in coincidence.

Not wanting to belabor this definition, all outputs of the program described here
that precisely coincide with the original human work are musical, and that those do
not match the original work should be taken as musical only as they individually
fulfill the above criteria.
Interestingly, one of the tests that computer music analysis makes possible is
what I call the jigsaw test. Like a jigsaw puzzle, a single piece of music is carved into
thousands of multiple-note varying-size note-event groupings and shuffled so that
they do not occur in any particular order. Ontimes of the note-events are reset so
that they relate to one another in their respective groupings but no longer give any
hint as to their original temporal location. Analysts then must put the pieces back
into proper order again, using whatever rules they wish to employ. This Humpty
Dumpty approach poses very difficult problems for even the simplest of tonal
works, no less for complex post-tonal music. Familiarity with the jigsawed work
can certainly give analysts an advantage. However, for most, the test poses severe
problems—even for this single work process—as the example of a Bach chorale in
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 248

248 HIDDEN STRUCTURE

À & bb 41 1

œ
2

œ
3

œ
4
œ
5 6
œ
7
œ
8
œ
9 10
œ
11
œ
Soprano œ #œ

b
Alto & b 41 œ œ œ œ #œ #œ œ œ œ œ
œ
b œ œ œ œ
Tenor V b 41 n œ œ œ œ œ œ œ œ

? b b 41 œ œ #œ œ
Bass
à œ
œ œ œ œ œ œ

À & bb œ 12 13 14

œ
15
œ
16
œ
17
œ
18
œ
19
œ
20 21
œ
22

œ
S œ œ

b
A &b œ œ œ œ œ œ #œ œ œ
œ œ œ

b œ œ œ œ œ œ œ œ
T V b œ œ œ œ

œ œ œ œ œ
B
à ? bb œ œ
œ
œ œ œ

À & bb 23 24
œ
25
œ
26 27 28
œ
29
œ
30

œ œ œ
31 32 33
œ
S
œ œ œ #œ

b
A &b œ œ œ œ œ œ œ œ œ #œ
œ œ
b œ œ œ œ œ œ œœ
T V b œ œ œ #œ #œ œ

œ œ
B
à ? bb œ œ œ œ œ œ œ œ œ œ œ

FIGURE 6.6 A Bach chorale as example of a jigsaw puzzle.

07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 249

GENERATIVE MODELS OF MUSIC 249

Figure 6.6 proves. The solution for this puzzle is discovering the source chorale
number in Bach’s 371 chorales as published in Riemenschneider (1941). Solving
this puzzle will require all four principles described in Chapter 1 of this book and
will further demonstrate the complexities involved in recombinancy and the need
for using computers in analysis.

PROBABILITIES
As discussed in Chapter 1, probabilities played an important role in early attempts
at computer music composition and analysis (Hiller and Isaacson 1959). The Monte
Carlo method, stochastics, Markov chains, and so on that these individuals used in
their early experiments all fall under the general rubric of probability theory, where
a limited range of choices regarding the occurrence of certain events governs com-
positional and analytical processes. Probabilities can reveal both the predictable
and the unpredictable and can even provide access to those special moments that
sometimes indicate more about a musical composition than all of the more generic
materials that surround them. Statistically small-probability passages often provide
the key to a work’s musical character. Using Markov chains (discussed later in this
chapter in more detail) and similar processes first to analyze music and then to
compose using the information thus gathered can provide invaluable information
about the workings of music.
Probabilities contribute to the understanding of almost everything. The world is
gray, not black and white, and where we find grayness, probabilities offer one of
the best potentials for our comprehension. Probability theory helps to predict out-
comes so that decisions made from among many choices will be informed, not ran-
dom (Freund 1973).
Understanding probabilities requires the understanding of possibilities. For ex-
ample, the probabilities of tossing a coin and producing heads after having already
tossed a coin and producing heads fifty times in a row remains 50/50, or 0.5. Every
toss of a coin—all other factors being equal—has a 0.5 probability of producing
heads and a 0.5 probability of producing tails. Coins know nothing about previous
tosses. Although most likely obvious, knowing the true outcome possibilities helps
us to understand probability.
Many such possibilities, though, do not prove as simple, even if they initially ap-
pear simple. For instance, ask someone to select a single pitch from an imaginary
piano keyboard and keep the choice secret. Now, try to figure out the probability of
choosing which pitch they selected. The answer—the simple part of this question—
is obviously 1 in 88, or 0.01136. However, discovering the probability of having two
people independently choose the same pitch in their heads does not prove so sim-
ple. The correct response to this more difficult question depends on the number of
composite possibilities involved. In this situation, the composite number rises sig-
nificantly as one would expect, from 88 to 882, or 7744. However, the number of
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 250

250 HIDDEN STRUCTURE

possible correct matches between the two participants increases as well. Each key
on the imaginary keyboard can be correctly chosen by both participants once, pro-
ducing a total of 88 possible correct matches. Thus, the probabilities of both indi-
viduals choosing the same key is 88 in 7744, surprisingly the precise same
probability as one person choosing the same single pitch—0.01136.
Understanding probabilities of the musikalisches Würfelspiele discussed in Chap-
ter 1 also requires carefully figuring of possibilities. These combinatorial musical
dice games created by composers such as Kirnberger, C. P. E. Bach, Haydn, and
Mozart involve the tossing of dice to make selections from matrices containing
numbers keyed to measures of music. These measures of music are then joined to-
gether to create musical phrases and short musical compositions (see Figure 1.10).
Computing the probabilities based on tosses of dice would seem as simple as com-
puting the probabilities of coin tosses and the mental selection of pitches on a key-
board. However, dice tosses hold many surprises. As the examples in Figure 1.10
point out, the total number of different outcomes from tossing a pair of dice is
eleven (rolling the number 1 is not possible). Furthermore, the possibilities of toss-
ing additions of two dice totaling 2 through 12 are not equal. There is, for example,
only one way for the combination of two dice to add up to either the number 2 (1
and 1) or the number 12 (6 and 6). However, there are several ways for two dice to
add up to the number 6 (i.e., 3 and 3, 2 and 4, 1 and 5). In fact, the possible ways to
toss numbers decreases toward the extremes (2 and 12) in ways that make the dice
throws in musikalisches Würfelspiele probabilistically quite uneven. Thus, under-
standing the possibilities involved in games such as this will lead to better compu-
tations of probabilities. One would imagine the composers of these dice games
knew this and created their music in ways that matched the increased likelihood of
certain choices with their most interesting musical measures.
Studies of probability also include concepts of probability functions. Probability
functions take variables as input and provide the probability of the variable’s hav-
ing a certain value as output. The set of probabilities that a function assigns to all
of its variables is known as a distribution. Variables can be independent, as with the
coin example given previously, or dependent, as with the example given previously
of two individuals making keyboard pitch choices and with Musikalisches Würfel-
spiele. Dependent variables can also be conditional—that is, the probability of one
variable may differ given the state of the other variable.
Probabilities can also change over time. In other words, new information may al-
ter probabilities. The Bayes rule follows the relatively simple notion that adding
new information to existing information impacts probabilities and their distribu-
tions. For example, consider a situation in which a bass drum strikes on the down-
beat of the first measure of a work. The probabilities of this bass drum strike
occurring on the downbeat of m. 2 of this same work are 50/50 or 50 percent—the
bass drum strike can either occur or not occur with equal certainty. However, if the
bass drum strike does occur on the downbeat of m. 2 of this same work, the proba-
bility for it occurring on the downbeat of m. 3 changes to 66.7 percent. A third oc-
currence increases the odds to .75 for a fourth occurrence, and so on.
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 251

GENERATIVE MODELS OF MUSIC 251

The Bayes rule appears in Figure 6.7 in mathematical form. This mathematical
form roughly translates into English as “The probability of A given B equals the
probability of B given A times the probability of A, divided by the probability of B.”
The variable A represents the current state of data, while the variable B refers to
new information that will most likely cause a change in the probabilities of A. There
are many such rules that can affect probability—more than we can investigate here
(for more information see particularly Temperley 2007, 7–19; Ghahramani 2005).
The Markov chains (discussed later in this chapter) that underlie much of this
chapter’s approach to musical analysis rely on such probabilities.
Initially, probabilities might seem more important to composers than to music
analysts. After all, many composers (most notably Xenakis, who used stochastic
probabilities to compose much of his music) have explicitly used probabilistics in
their compositional processes. However, understanding the probabilities of the
choices that composers make in their compositions, whether they do so explicitly
or not, represents a significant component of music analysis (Temperley 2007).
Most composers make the same or similar choices more than once, with their
choices dependent on other choices that then change over time. Knowing about
such probabilities enables music analysts to develop methods for understanding
and representing such choice making.
Using probabilities in computer programs requires the presence of as much rele-
vant data as possible. Data-driven computer programs, the ones that I prefer, pro-
vide perfect environments for probability analyses of music. Interestingly, the
manner in which data is stored in databases greatly influences both the effective-
ness of computation and, even more important, the types of applications that pro-
grammers might want to develop. Methods of data representation can enhance or
limit the imaginations of those who use them. For example, large amounts of data
stored according to pitch numbers rather than ontime numbers will most likely
support the counting of pitches in a work while concomitantly discouraging pitch
pattern recognition. Unfortunately, such storage processes would also make perfor-
mance, formal, and structural analyses less feasible, as well as posing problems for
analysts intent on studying temporal analysis strategies. Therefore, data represen-
tation and storage should be as flexible as possible to allow as many approaches to
analysis as conceivable.

P (B\A) P (A)
P (A\B) =
P (B)

FIGURE 6.7 The Bayes rule.

07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 252

252 HIDDEN STRUCTURE

RULES AND MARKOV CHAINS

As mentioned previously, generating new music from an analytic system and then
comparing the results with the actual music used for generation represents one of
the most straightforward forms of validation of the effectiveness of that analytical
system. Completing a known work separated from its ending and then comparing a
new ending with the real ending provides another example of such validation. Inter-
estingly, it is much more difficult to add music to an incomplete version of an ex-
tant score than to compose stylistically credible music from scratch. For example,
interrupted voice-leading requires a newly composed extension to initially meet
goals that the rules may not adequately prepare the program to meet. Indeed, if an
interrupted interval does not already exist in the database as a completed voice-
leading (i.e., it represents its first appearance), then a work cannot be completed at
all. Even if this interval exists once in the database, there may be no way the pro-
gram can possibly complete the piece in the same or a similar manner to that used
by the work’s originator.
I have chosen music from Béla Bartók’s Mikrokosmos to computationally analyze
for its rules and produce viable new endings. All four of the examples used for
analysis here appear in the third volume of the Mikrokosmos: 80, 81, 91, and 92.
These works share a number of common features that make them attractive to use
for research as described here. First, each work resembles a two-part invention,
rarely exceeding two voices in texture. This textural consistency provides excellent
clarity for discovering rules. Second, each of the four pieces exhibits extensive
chromaticism appearing at roughly similar rates. Third, the two-part counterpoint
in these selections exhibits the kind of tension and release typical of many other
post-tonal works.
Deriving rules from music is a critically important process for computer analysis
programs (Maxwell 1992; Schwanauer 1993; Widmer 1992). Markov chains can
greatly help this analysis by classifying rules by their probability of occurrence.
Markov chains are probabilistic processes often expressed in terms of orders (Ames
1989). A zero-order Markov chain, for example, makes random decisions, with no
applicable rules. A first-order Markov chain, however, bases new decisions on imme-
diately preceding choices. A fifth-order Markov chain bases its decisions on the pre-
vious five choices. Figure 6.8 shows a matrix (called a state transition matrix) for a
first-order Markov chain involving all twelve pitches of the chromatic scale. The
numbers in the matrix here represent probabilities that add to 1.0—or 100 percent—
in each horizontal line. To interpret this matrix, a program would find a current
pitch in the left-hand column, and then read the various probabilities of potential
following pitches along the associated horizontal row. The probability of the pitch
A being followed by another pitch A in the state transition matrix given here, for in-
stance, is 10 percent (0.1); of the pitch A being followed by the pitch B-flat, 20 per-
cent (0.2); and so on.
As a more musical example of this process, imagine that a melody-composing
program chooses new pitches by virtue of their immediately preceding pitches
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 253

GENERATIVE MODELS OF MUSIC 253

FIGURE 6.8 A state-transition matrix for a first-order Markov chain involving all
twelve notes of the chromatic scale.

(first-order Markov chain). If only a major or minor second in any direction is al-
lowed between consecutive pitches, such a first-order Markov chain would produce
simple meandering, stepwise melodies. A more elegant program might have new
choices depend on two preceding pitches. For example, if the previous two pitch
choices represented a leap upward, then new choices might be limited to pitches
moving a second in contrary motion. Other rules would then need to exist for other
pitch combinations. This second-order Markov chain would produce considerably
more interesting music than a first-order Markov chain.
Second-order Markov chain state transition matrixes appear much like expanded
first-order state transition matrixes. Figure 6.9 shows a portion of the upper left-
hand corner of a twelve-pitch matrix such as that shown in Figure 6.8. Here, each
row states the probability of pitches following two previously established pitches.
Note that the probabilities in each horizontal rank must still total 1.0. Second-order
Markov chains have far more constraints—conditional probabilities—than do first-
order Markov chains, as do third-order Markov chains over second-order Markov
chains, and so on.
Common Lisp provides useful ways of organizing data to represent state transi-
tion matrices. Instead of describing explicit probabilities as in state transition
matrixes, pitches appear in lists such that duplications increase the probabilities
proportionately. For example, the following simple list
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 254

254 HIDDEN STRUCTURE

FIGURE 6.9 The upper left-hand corner of a second-order Markov chain state tran-
sition matrix.

(60 (62 64 64 64 65 71 71 71 71 71))

indicates that the MIDI pitch 62 (D) follows the MIDI pitch 60 (C) with a 0.1 proba-
bility (1 of 10 possible following pitches), the MIDI pitch 64 follows the MIDI pitch
60 with a 0.3 probability, with the probability of MIDI pitch 65 at 0.1, and that of
MIDI pitch 71 at 0.5. Duplicated pitches indicate the higher probability of their be-
ing chosen in a resulting composing process. Only existing probabilities appear in
this matrix, with all other pitch possibilities understood to be 0. Although this
notation may not be as readable as a more traditionally notated state transition
matrix, it works especially well for computational purposes. The Markov program
accompanying this book on CD-ROM collects the first-order probabilities of any
succession of pitches in music in a database and places the resulting probabilities
in lists representing a simplified state transition matrix. The resulting analysis is
then used by this Markov program to produce new music in a style roughly compa-
rable to that of the original upon which it is based.
Complete rules in the Markov program look like the following, stored as a Lisp-
style list form state transition matrix here given the symbol name *stm*.
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 255

GENERATIVE MODELS OF MUSIC 255

? *stm*
((60 (61))
(61 (65 64))
(64 (63 63))
(63 (61 65))
(65 (68 68))
(68 (71 69))
(69 (67 71))
((71 (72 72))
((72 (69))
(67 (64)))
This form of rule storage represents just one of the many possible—arrays, tables,
objects, and so on, being obvious alternatives—but follows the Lisp list vocabulary
especially well. The asterisks surrounding the “stm” symbol here indicate that it
represents a global variable. Global variables can be accessed anytime by users
and code function, and their values typically have relevance to more than one
process. For example, the global variable *tempo* would likely need to be avail-
able to many different functions. Global variables differ from local variables defined
in function argument lists or in let statements. Such local variables can only be
accessed from within their parenthetical function definitions, and calling them out-
side of these environments will cause errors.
A large enough amount of data must be used for a state transition matrix; other-
wise, the program output can easily lapse into repetitive back-and-forth motions or
other types of dead ends most likely atypical of the analyzed music. Figures 6.10
and 6.11 present both successful and unsuccessful examples of Markov output rep-
resenting extensions of the melody from Bartók Mikrokosmos nos. 71 and 77. In Fig-
ure 6.10, the Markov program required just five runs to duplicate the actual pitches
that Bartók composed.
The first argument to the function compose-new-music-based-on-markovian-
probabilities, the top-level function of the Markov program on the CD-ROM ac-
companying this book, includes all but the final five pitches that the program then
returns as its best attempt to duplicate the original music. The data used as the
third argument to the function compose-new-music-based-on-markovian-
probabilities requires that the final pitch entry occur at least one other time in
the list, so that it is not followed by a nil pitch. Such nils do not cause errors, but
create further nil pitches, which then continue indefinitely because nil pitches do
not otherwise exist in the state transition matrix. The code below presents the calls
to compose-new-music-based-on-markovian-probabilities that produced
the music in Figure 6.10.
(defVar bartok-2
'(67 65 64 65 67 65 67 64 65 67 67 65 64 67 65 64 62 64 65)
"mikrokosmos 71.")
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 256

256 HIDDEN STRUCTURE

a.
3 3
& C Ó ˙ 2 œ œ œ œ œ œ C œ Œ œ œ œ Œ ˙ 2 œ œ œ œ œ œ ˙. Œ œ œ C œ œ œ œ œ

b.
3 3
& C Ó ˙ 2 œ œ œ œ œ œ C œ Œ œ œ œ Œ ˙ 2 œ œ œ œ œ œ ˙. Œ œ œ C œ œ œ œ œ

c.
3 3
& C Ó ˙ 2 œ œ œ œ œ œ C œ Œ œ œ œ Œ ˙ 2 œ œ œ œ œ œ ˙. Œ œ œ C œ œ œ œ œ

d.
3 3
& C Ó ˙ 2 œ œ œ œ œ œ C œ Œ œ œ œ Œ ˙ 2 œ œ œ œ œ œ ˙. Œ œ œ C œ œ œ œ œ

e.
3 3
& C Ó ˙ 2 œ œ œ œ œ œ C œ Œ œ œ œ Œ ˙ 2 œ œ œ œ œ œ ˙. Œ œ œ C œ œ œ œ œ

f.
3 3
& C Ó ˙ 2 œ œ œ œ œ œ C œ Œ œ œ œ Œ ˙ 2 œ œ œ œ œ œ ˙. Œ œ œ C œ œ œ œ œ

FIGURE 6.10 (a) Original, Bartók, Mikrokosmos, no. 71, transposed to begin on G;
(b–f) five computer extensions beginning five notes from the end of
the passage.
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 257

GENERATIVE MODELS OF MUSIC 257

a.
2
& 4 œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œj ‰

b.
2 j
&4 œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ ‰

c.
2 j
&4 œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ ‰

d.
2
& 4 œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œj ‰

FIGURE 6.11 (a) Original, Bartók, Mikrokosmos, no. 77, transposed to begin on D;
(b–d) three computer extensions beginning five notes from the end
of the passage.

(compose-new-music-based-on-markovian-probabilities
65
5
(67 65 64 65 67 65 67 64 65 67 67 65 64 67 65 64 62 64
65))
(67 65 64 65 64)
(compose-new-music-based-on-markovian-probabilities
65
5 (67 65 64 65 67 65 67 64 65 67 67 65 64 67 65 64 62 64
65))
(67 65 67 67 65)
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 258

258 HIDDEN STRUCTURE

(compose-new-music-based-on-markovian-probabilities
65
5
(67 65 64 65 67 65 67 64 65 67 67 65 64 67 65 64 62 64
65))
(64 65 67 64 67)
(compose-new-music-based-on-markovian-probabilities
65
5
(67 65
64 65 67 65 67 64 65 67 67 65 64 67 65 64 62 64 65))
(64 65 64 65 64)
(compose-new-music-based-on-markovian-probabilities 65 5
(67 65 64 65 67 65 67 64 65 67 67 65 64 67 65 64 62 64
65))
(64 62 64 65 64)
The fifth run of compose-new-music-based-on-markovian-probabilities
here produces the actual ending Bartók composed. Even the previous four unsuc-
cessful attempts reveal logical orders of pitches that make musical sense, lacking
only cadence and other patterns that higher-order Markov chains would catch.
Viewing the state transition matrix for this example proves that the resulting cor-
rect solution exists in the data itself.
((67 (65 65 67 64 65 65))
(65 (64 64 67 67 67 64))
(64 (65 62 67 65 65))
(62 (64)))
Figure 6.11 presents an unsuccessful example of first-order Markovian output. No
matter how many runs the program makes, it never achieves the results that
Bartók created. The reason, of course, lies in the state transition matrix, where the
final pitch 62 does not follow pitch 67 in the original and thus will never appear in
that order in the output.
((62 (65 65 65 65 65))
(65 (69 62 62 69 62 62 67 69))
(69 (67 67 71))
(71 (69))
(67 (65 65 65)))
Like all but the final example presented in Figure 6.10, these three unsuccessful out-
puts in Figure 6.11 retain much of the logic of Bartók’s original music, even though
they do not match Bartók’s ultimate solution.
The Markov program on the CD-ROM accompanying this book cannot produce
pitch orders or pitches themselves that have not already occurred in the data. This
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 259

GENERATIVE MODELS OF MUSIC 259

represents a significant problem, because composers commonly transpose pitches,

motives, harmonies, and so on, creating new pitch orders and new pitches. For
Markov to accomplish this requires the use of intervals instead of pitches in its
rules. Using intervals enables the program to follow the intervallic template of the
analyzed music, and, using that template, output music that more likely follows
what a composer would create. Because the Markov program deals with numbers,
translating pitches to intervals at the outset and then converting the intervals back
to pitches at the output level will convert the program properly. The problem with
using intervals, of course, is that these intervals will not necessarily conform to a
scale in use. This shortcoming can cause numerous out-of-scale pitches to occur,
pitches that might never appear in the original music. Including analyzed scales re-
quires adapting code from Chapter 4 of this book, and thus more discussion of the
Markov program than is deserved here. Therefore, rather than convert Markov to
an interval base at this point, I will now describe a more elaborate program that
attempts to extend Bartók’s two-part contrapuntal music using a similar Markovian
model, but one using intervals instead of pitches.
The Extend program presented on the CD-ROM accompanying this book gathers
interval rules from its database music and then uses these rules to compose. Rules,
particularly voice-leading rules as in the Extend program, may be considered sepa-
rately from the pitch classes that define them and thus reveal other important as-
pects of the music being analyzed. Examining how the rules produce output
identical to that of a human composer can lead analysts to important discoveries
as to what constitutes proper and improper voice motions. As with the Markov pro-
gram, even studying the rules of unsuccessful output can reveal significant insights
into compositional process and thus produce a better understanding of a work un-
der analysis.
As an example, Figure 6.12 presents a simple one-against-one two-voice counter-
point in which each voice moves stepwise. The pitches, the consonance or disso-
nance they produce when sounded together, and other aspects all contribute to the
complete phrase in which they occur (not shown here). Each of these attributes
also contributes to the functions implied here, with phrases beginning on tonic,
ending on tonic, and having penultimate dominant functions that help provide final
cadences. Clearly, the voice-leading of this simple counterpoint supports this sim-
ple analysis. The voice-leading contributes to a sense of functional continuity with-
out which a phrase’s implied harmonic direction would suffer.
Figure 6.13, in contrast to Figure 6.12, presents an example with various two-
pitch groupings that lack clear traditional functions. Interestingly, this example con-
sists of the same voice motions (without regard for vertical simultaneities) in
different orders than the music of Figure 6.12. Therefore, one might argue that not
all of the functions in Figure 6.12 have disappeared from the music in Figure 6.13.
Figure 6.14b and 6.14c present two computer extensions to the original Bartók
first phrase presented in Figure 6.14a (mm. 5–7). Figure 6.14b uses a first-order
Markov process, while Figure 6.14c uses a second-order Markov process. Figure
6.14b is substantially different from the original, with awkward leaps, unusual
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 260

260 HIDDEN STRUCTURE

&c ˙
˙ ˙ ˙ ˙ ˙
Piano

?c ˙ ˙ ˙ ˙ ˙ ˙

FIGURE 6.12 A simple two-voice counterpoint in which each voice moves stepwise
in various directions.

&c ˙ ˙ #˙ ˙ b˙ ˙
Piano
˙
? c ˙ b˙ b˙ b˙ ˙

FIGURE 6.13 A post-tonal example, where the various two-note groupings do not
have clear functionalism.

consecutive vertical intervals, and almost continually parallel motion. Figure 6.14c,
however, matches the original exactly. Thus, second-order Markov extensions prove
more robust than first-order, particularly when trying to match the original rather
than varying it (as is the case with most style composing programs).
Evaluating different extensions such as that in Figure 6.14b also poses many
problems—for example, determining whether completing one of, say, Bartók’s
works by anyone other than Bartók can be as good or better than Bartók’s own
completion. Anyone aware of my other books on algorithmic music (see bibliogra-
phy) will probably predict my response to these problems: certainly such exten-
sions can be as good as or better than the original, as well as worse than the
original. Making such choices, however, requires that listeners shed their notions
that the only music worth the definition must originate with human composers.
The extensions presented in this chapter do not include rhythm, meter, dynam-
ics, articulation, phrasing, and so on, all critical ingredients in the full musical expe-
rience. In order to make these extensions fit reasonably with their appropriate
works, I simply use the rhythm of the extended passage, tying pitches that repeat.
Hence, the Extend program available on the CD-ROM accompanying this book
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 261

GENERATIVE MODELS OF MUSIC 261

Hommage à R. Sch.

a. Andantino, piacevoie q = 72
b œ. œ bœ. œ
& b b 42 œ . n œ œ . œ ˙ œ. ‰ Œ
J œ. nœ œ. œ

b 2
&b b 4
nœ œ nœ œ #œ bœ bœ œ nœ œ nœ œ #œ bœ bœ œ nœ œ nœ œ

b œ. œ bœ. œ
6
j #œ bœ bœ œ nœ œ nœ œ #œ œ bœ œ
&b b ˙ œ. ‰ Œ

b
& b b#œ bœ bœ
œ nœ œ nœ œ #œ bœ œ œ œ. œ bœ. œ œ. nœ #œ. œ ˙

12
b nœ œ #œ bœ bœ œ nœ œ nœ œ œ. œ œ.
& b b nœ nœ ˙ nœ
F dim.
b
&b b j ‰ Œ œ nœ. nœ œ. œ nœ. œ œ. œ nœ. œ b˙
œ. nœ.

œ. nœ œ #œ œ nœ œ bœ ˙
17
b
&b b ˙ ˙ ˙
p
b
&b b
œ. œ nœ nœ b˙ œ nœ œ nœ bœ œ ˙ ˙

FIGURE 6.14 Two possible nonidentical computer extensions (b and c) to the orig-
inal by Bartók (Mikrokosmos, no. 80) presented in (a).
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 262

262 HIDDEN STRUCTURE

b.
b ‰. œ œ. nœ œ
V-1 &b b c Œ R . œ ˙ Ó

b
V-2 & b b c Œ. j j ‰ Ó.
œ œ œ œ bœ œ

c.
b 2 ‰. œ œ. œ œ.
V-1 &b b 4 Œ R œ ˙

b
V-2 & b b 42 Œ . œ
j
#œ œ bœ œ j Œ.
nœ

FIGURE 6.14 continued

represents only a beginning to an approach to modeling. Like many of the pro-

grams presented in this book, the Extend program provides a useful beginning for
musical investigation that will ultimately help theory become reality in order to
verify hypothesis.
Other programs can complete extant works as does Extend. For example, in
Cope 2004 and 2005 I describe a relatively simple computer program called Gradus
that analyzes musical output for its rules and then creates new musical data follow-
ing those rules (Principle 4). Gradus differs from the Extend program just described
in that Gradus actually learns by decreasing the backtracking necessary when
reaching dead ends until such backtracking is no longer necessary. Originally de-
signed to produce two-voice one-against-one Fux-like vocal counterpoint, the same
counterpoint often encountered in beginning theory courses at many American
universities, the program can—by using different databases—compose music fol-
lowing any rules (including, say, twentieth-century counterpoint by Bartók). Unfor-
tunately, although flexible in terms of the two-voice music in its database, the
program only collects allowable vertical intervals and horizontal voice motions.
Therefore, although Gradus has many interesting features—learning, style acquisi-
tion, and so on—it has little flexibility in regard to completely new situations,
where neither allowable vertical intervals nor allowable horizontal voice motions
play roles. For example, it would not make much sense to use Gradus to analyze se-
rial music for, in the program’s current state, none of the more salient features of
serial music would translate to Gradus’s output. Although such a program could be
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 263

GENERATIVE MODELS OF MUSIC 263

created and, indeed, could be extremely valuable, it should also be a separate pro-
gram and not an expansion of Gradus.
Alice (the ALgorithmically Integrated Composing Environment) is a Macintosh in-
teractive computer composing program (see Cope 2000) designed to provide an in-
troduction to basic interactive music composition in the style of the composer using
the program. Alice differs from Experiments in Musical Intelligence but resembles
both Extend and Gradus in its use of inherited voice-leading rules rather than re-
combinancy. Alice presents a variety of analytical methods to aid composers in set-
ting variables to produce music more to their liking. Figure 6.15 shows three
different forms of analysis provided by Alice. In Figure 6.15a, a single window indi-
cates percentages of pitch classes in a pie chart. A column chart provides identical
information, but with a computational range of mode possibilities. These charts are
then followed by a linear diagram of when pitch classes first enter, a texture analy-
sis, a pitch/duration intersecting analysis scatter plot chart, and a channel-timbre-
use column chart. Figure 6.15b presents a rules window with rules consisting of the
current and the destination pitch classes, interval-class voice motion, and the chan-
nel in which motion takes place (discussed in detail shortly; also see Cope 2000).
Figure 6.15c presents a structural analysis defined by straight lines and a formal
analysis defined by curved lines.
Alice creates its rules by first grouping the music appropriately, converting the
pitches of these groupings into pitch-class sets, and then deriving voice-leading
rules from the results of these processes. A rule in Alice takes the form
(((3 3) 2 1) ((5 1) -2 2) ((0 0) 2 3) ((8 5) -1 4))
with each rule sublist (e.g., ((3 3) 2 1)) consisting of a head sublist followed by two
numbers. The numbers in this head sublist represent the origin and destination
pitch classes. The number immediately following the head sublist indicates the di-
rection and amount of scale-degree motion between the two pitch classes in the
rule sublist, with negative numbers indicating downward direction. The last num-
ber in the rule sublist represents the channel to which the rule part applies (each
voice in Alice must be assigned to a different channel). Therefore, the rule sublist
((3 3) 2 1) indicates the motion of an upward major second between pitch class 3 in
the chord of origin and pitch class 3 in the chord of destination in channel 1. Al-
though origin and destination pitches appear as pitch classes, motion is measured
in scale degrees according to the current scale. Hence, the 2 in rule part ((3 3) 2 1)
indicates a major second upward only given that the scale here is measured in half
steps. Using a scale defined in whole steps, this 2 would equate to a major third.
Figure 6.16 shows the rule (((3 3) 2 1) ((5 1) -2 2) ((0 0) 2 3) ((8 5) -1 4)) as it would
appear in music notation with scale degrees measured in half steps.
Voice motion in rule notation must be flexible enough to represent more than
scale degrees. This notation must also be able to represent chromaticism when us-
ing scales other than the chromatic scale. In Alice, each half step above a scale
degree that is not itself a scale degree receives a superscript caret, one per half
step. Hence, the half step raising the second degree of the pentatonic scale of
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 264

264 HIDDEN STRUCTURE

a)
a.

FIGURE 6.15 Three different forms of analysis provided by Alice.

07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 265

GENERATIVE MODELS OF MUSIC 265

b.

c)

c.

FIGURE 6.15 continued

07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 266

266 HIDDEN STRUCTURE

& b b www wwww

w
[0,3,5,8] [0,1,3,5]

FIGURE 6.16 The rule (((3 3) 2 1) ((5 1) -2 2) ((0 0) 2 3) ((8 5) -1 4)) in music notation.

C–D–F–G–A would be 2^ for D-sharp, and the double raising of this second degree
would be notated as a 2^^ for E-natural. All pitches chromatic to a given scale are
figured as raised scale degrees in Alice. Hence, no lowered form of notation need
exist. This chromatic notation allows for every possible motion between members
and nonmembers of any scale. Such inclusivity greatly enhances Alice’s ability to
adapt to various styles of music.
To create new groupings using these rules, an appropriately analyzed and named
set is matched to an OOP lexicon of such sets in a database. An alternate but iden-
tically named set is then substituted for the current set. This process of substitut-
ing similarly named sets can produce new orderings of the original set based on
matching two sets with new voice motions, a new destination set, or some combi-
nation of these two possibilities to create new music. As an example, the initial
pitches of a first set [0,2,4,7] are rearranged to match the newly found pitch-class
set’s [4,7,2,0] pitch-class order. This ensures that the new rule motion will be ap-
plied properly. The new rule then replaces the original rule.
The two distinct motions shown in Figures 6.17a–6.17c will make this process
clearer. Figure 6.17a shows the original motion of two groupings in the database.
Figure 6.17b shows the rule formed by this motion with the applicable scale in this
case measured in half steps. Note that the order of the rule follows the sets from
lowest to highest channel. Figure 6.17c shows a second motion whose first set re-
duces to the same set as in Figure 6.17a. This equivalent set, however, moves to a
different destination set with this movement shown in the rule of Figure 6.17d. Alice
then substitutes the first set of Figure 6.17a for the first set of Figure 6.17c and uses
the voice-leading of Figure 6.17c to create a different—but correct—motion to the
second set of Figure 6.17c. In order to link the elements of the first sets together
logically, the program exchanges them so that their equivalent elements connect
(i.e., so that the first set connects appropriately with the applicable rule). Hence,
the first set of Figure 6.17c must be made to have the set-member order of the first
set in Figure 6.17a. Once this is accomplished, however, the actual motion from the
first set of Figure 6.17a using the rule of Figure 6.17c–d follows channels rather than
set numbers. The results of this process appear in Figure 6.17e. The rule derived
from Figure 6.17e, shown in Figure 6.17f, demonstrates how the elements of 6.17d
have been rearranged with different channel settings.
I have recounted Alice’s rule-based analysis and composing processes here
because they relate directly to modeling and because, by creating extensions to
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 267

GENERATIVE MODELS OF MUSIC 267

w
7
&w
3 5
ww
0 -1 0
a.

? w2 -1
# ww2
w
4 2 7
[0,2,4,7] [0,2,5,7]
b. (((4 7) 2 1)((2 2) -1 2)((0 0) -1 3)((7 5) -3 4))

4 3 3
& # ww n ww
2 1 7
c.

? ww
7 2
w1
w
0 -2 0
[0,2,4,7] [0,1,3,7]
d. (((0 0) -2 1)((7 1) 2 2)((2 7) 1 3)((4 3) 3 4))

& ww
7 2
w1
bw
0 -2 0
e.

? w2 1
b w7
w w
4 3 3
[0,2,4,7] [0,1,3,7]
f. (((4 3) 3 1)((2 7) 1 2)((0 0) -2 3)((7 1) 2 4))

FIGURE 6.17 Extending voice-leading rules.

07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 268

268 HIDDEN STRUCTURE

existing music, Alice exemplifies the composing approach described previously in

this chapter. I invite those readers who find Alice’s processes difficult to follow or
who remain curious about the program’s potential for acting as an analytical tool to
carefully review the chapters in Cope 2000 that relate directly to Alice’s inferential
rules-based analytical and composing tools.
Not all music analysts are as sanguine about modeling as a process for better un-
derstanding music. For example, in a reply to Stephen Smoliar’s “Modelling Musical
Perception: A Critical View” (Smoliar 1999), Peter Desain and Henkjan Honing state:
Interestingly, even an algorithm that always produces a “correct” output is not
good enough: it does not validate the algorithm as a model of the cognitive
process itself. If we want to make statements about the architecture of human
cognition, we have to relate the architecture of the program to that of the human
subject. This is still one of major challenges of the computational modeling of
music cognition. (Desain and Honing 1999, 114)

Obviously, I too believe that more than one way to create a particular composition
exists, having proven here and elsewhere that both recombinancy and rules can
produce precise replications of existing music. Believing in this multiplicity of ap-
proaches, however, does not negate the use of modeling as an analytical process
for better understanding music. Indeed, I believe that composers compose using a
hybrid of many approaches, rather than any single approach, and that identifying
only one contributor could significantly diminish musical analysis.
Leonard Meyer comments on the possible differences between computer genera-
tions and human genius:
Why, out of all the possible alternatives that the composer might have devised
or considered for use in some work, was this particular one chosen rather than
some other? To make the point forcefully and with only slight exaggeration, the
difference between a crackpot and a genius is not primarily a matter of fecund in-
vention; both readily devise novel, imaginative, even eccentric, possibilities.
Rather the difference lies in the ability of the genius to choose with perspicacity.
(Meyer 1989, 135)

The fact that given a set of rules captured from a database of previous composi-
tions, a computer program can, among several outputs, generate the exact set of
pitches that a human composer has created does not necessarily mean these out-
puts represent a credible or all-inclusive analysis of a work. However, such replica-
tive extensions cannot be ignored either. Clearly, the rules discovered, and the
choices between various correct possibilities made, bear close resemblance if not a
strong overlap with those rules and choices that composers actually use and make.
However, even if, as those who criticize the modeling process state, such exten-
sions do not replicate those the composer actually used, they may still shed impor-
tant light on what transpires in the music—a not inconsequential truth.
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 269

GENERATIVE MODELS OF MUSIC 269

MUSICAL EXAMPLES
I occasionally assign advanced music students the completion of works of music
just as Extend does. Figure 6.18b–d presents several such student-created exten-
sions based on the Bartók work shown in Figure 6.18a. The students in this case
ranged from college juniors and seniors in upper-division music theory courses to
first-year graduate composition students. The results here do not follow the rules
as closely as those produced by the Extend program. None of the student exten-
sions remotely resemble the actual ending of the work. At the same time, most of
these student extensions are interesting, certainly creative, and musical, though
using such complimentary terms begs many questions.
Completing the opening section of Stravinsky’s Three Pieces for String Quartet
(1914) using even a program such as Extend that is limited to two rather than four
voices is fairly easy given that the program not have to deal with rhythm, meter, dy-
namics, and so on, as does Alice’s analyzing and composing processes. Stravinsky’s
style—often consisting of different-length cycles of chord repetitions—makes almost
all output similar or the same. The final two measures of the movement, while re-
sembling the opening material and mathematically generated from it, cannot, unfor-
tunately, be predicted in the same way as continuing the first few measures can.
Although the music later in the movement derives from rhythmic variations of the
opening music, new sections contain related but impossible-to-predict material.

PROGRAM DESCRIPTION
The top level of the Extend program—the function complete-the-composition—
requires three arguments: a list of symbol names representing the note-events for
musical works, the Markov order number desired, and the last pitch of channel 1 of
the music to be extended. A scale-type optional argument that defaults to the
chromatic scale follows the required arguments. The function complete-the-
composition accomplishes two things: it runs get-rules, which returns a full
set of soon-to-be-described rules, and it runs the completion function, which re-
turns one of the many possible extensions of the incomplete music indicated by
the channel-1-last-note argument to complete-the-composition. This
function appears below in pseudo-code.

(defun complete-the-composition
(event-list-names markov-order channel-1-last-note
&optional (scale-type 'chromatic))
(get-rules event-list-names markov-order scale-type)
(completion channel-1-last-note markov-order) *new-work*))
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 270

270 HIDDEN STRUCTURE

a.
œ œ #œ œ œ œœ œ œ #œ œ nœ bœ œ
&c œ œ nœ œ œ ˙ œ œ‰
P J

j j
&c Œ œœ nœ œ œ œœ œœ œœ nœ œ #œ œ œ #œ. œ #œ #œ ‰
p œ

8 œœ nœ œ œ œœ œœ œ
œœ nœ œ #œ œ œ #œ. œ œ nœ ‰
&ΠJ J
p
& œ œ #œ œ ‰
P œ œœ œ œ #œ œ nœ bœ œ œ œœœ œ ˙ œ.

15

&Œ œ œ #œ #œ œ œ œ œ #œ œ #œ œ œœ bœ
p #œ œ #œ œ œ œœ
più p P
#œ #œ œ œ œ
& œ œ #œ #œ œ Œ ? #œ ‰ œ #œ œ
nœ œ œ #œ œ œ œ J
p più p P
22

& nœ œ œ bœ œ œ œ nœ œ œ œ bœ œ œ œ bœ œ œ œ ˙ œ œ #œ œ
p più p
? b œ œ # œ œ n œ b œ œ ‰ # œJ Œ œ nœ œ b˙ œ bœ œ
J J bœ œ œ œœ
p più p
29

& œ œœ œ œ #œ œ œ nœ œ 34 œ œ œ œ 42 œ œ œ œ œ
œ ˙ ˙
#œ œ œ œ
π P
œœ
?œœœ œ 34 œ œ b œ œ b œ 42 œ œ # œ œ œ ˙œ œ ˙˙
π P
FIGURE 6.18 Several student-created extensions (b–d) based on Bartók (a), Mikro-
kosmos, no. 81.
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 271

GENERATIVE MODELS OF MUSIC 271

b.
3 2
&4 œ œ œ œ œ 4 œ œ œ bœ œ #˙ ˙
P
? 43 œ œ b œ œ œ 2 œ œ # œ œ œ bœ œ n ˙˙
4
P
c.

& 43 œ 24
œ œ œ œ œ œ œ œ œ ˙ ˙
P
? 43 œ œ b œ
œ œ 2 œ œ œ œ bœ œ b œœ œœ n ˙˙
4
P
d.

& 43 œ œ œ œ œ 42 œ œ bœ œ œ œ œ ˙
P
? 3 œ œ bœ œ œ œ œ œ nœ œ bœ ˙
4 42 œ
P
FIGURE 6.18 continued

Rules in the Extend program take the form

((((19 16) -3)) (((16 12) -4)) (((12 16) 4)) . . .
with each rule consisting of a list of the intervals between two consecutive vertical
pitches, followed by the interval motion of the top voice. In detail, this means that
the rule
(((19 16) -3))
signifies that the first sublisted set of two vertical intervals (assuming first-order
Markov) of 19 and 16 (given a chromatic scale in use and indicating a perfect
twelfth and a major tenth), and the top voice moves -3 (a minor third downward).
Although this rule notation may seem initially awkward, in Lisp terms it character-
izes all Extend requires in order to run the compositional modeling process.
Figure 6.19 presents this rule in musical notation. In this case, the bottom voice re-
mains stationary to account for the fact that the motion in the voice in channel 1
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 272

272 HIDDEN STRUCTURE

&c ˙ ˙
Piano

?c w

FIGURE 6.19 The rule (((19 16) -3)) in music notation.

creates the new interval shown. Including the motion of the lower voice would pro-
duce unnecessary redundancy.
The extension program that uses these rules, run by the Extend compose func-
tion, applies the rule with the channel-1-last-note argument to complete-
the-composition described previously to create the next vertical interval.
Depending on how much data is provided to the Extend program, many rule choices
will exist for composition, as shown below when using the data in the bartok-79, bar-
tok-80, bartok-81, bartok-91, and bartok-92 databases provided with the program:
'((((24 24) 1)) (((24 24) 5)) (((24 24) 1)) (((24 24) -7))
(((24 24) 1))
(((24 24) 5)) (((24 24) 1)) (((24 24) -1)) (((24 24) -5))
(((24 24) -1)) (((24 24) 5)) (((24 24) -1)) (((24 24) -1))
(((24 24) -1)) (((24 24) 2)) (((24 24) 1)) (((24 24) 1))
(((24 24) -5)) (((24 24) 6)) (((24 24) -4)) (((24 24) -3))
(((24 24) 2)) (((24 24) 2)) (((24 24) 1)) (((24 24) 1))
(((24 24) -5))
(((24 24) 2)) (((24 24) 1)) (((24 24) -1)) (((24 24) 2))
(((24 24) 1)) (((24 24) -3)) (((24 24) 1)) (((24 24) 3))
(((24 24) -3))
(((24 24) -1)) (((24 24) -3)) (((24 24) 5)) (((24 24) -1))
(((24 24) -1)) (((24 24) -2)) (((24 24) 1)) (((24 24) -1))
(((24 24) 1)) (((24 24) -1)) (((24 24) 1)) (((24 24) -1))
(((24 24) 4))
(((24 24) -3)) (((24 24) -1)) (((24 24) 5))
(((24 24) -5)) (((24 24) -1)))
Note the many duplicate rules that appear in this list. However, the program uti-
lizes these duplicates to bias the program’s output toward those rules that occur
more often—the basis for the probabilities inherent in Markov state transition ma-
trixes discussed earlier in this chapter.
Rules in Extend are stored in the *rules* global variable and can be examined
by simply typing *rules* into the Listener window, followed by pressing the re-
turn key. Tracing functions when running the program can also be useful. To trace a
07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 273

GENERATIVE MODELS OF MUSIC 273

function, simply use its name as an argument to trace, which allows for any num-
ber of such arguments. A trace typically looks like this

? (trace get-the-multiple-rules)
? (get-the-multiple-rules
'(bartok-79 bartok-80 bartok-81 bartok-91 bartok-
92)
'chromatic)
Calling (get-the-multiple-rules
(bartok-79 bartok-80 bartok-81 bartok-91 bartok-
92)
chromatic)
get-the-multiple-rules returned t

with the function and its arguments provided first and then whatever the function
returns given last. Tracing lower-level functions while running higher-level func-
tions provides the most effective use of trace. Typically, functions that recurse ap-
pear only once, to avoid filling listener windows with vast amounts of information
and slowing programs significantly. Common Lisp provides ways to trace functions
(see Steele 1990) with many or even all of their recursive calls, should this be nec-
essary. Both forward and backward tracing provide useful techniques for debug-
ging code.
Creating programs such as Gradus, Alice, and Extend also requires several ap-
proaches to functional computing that have not yet been discussed here. Although
this may appear as a slight digression, I present them now so that readers can keep
up with the more complicated kinds of code that such programs require. Since
Chapter 2 we have used the Common Lisp function cons for recursion. However,
recursive functions may use any similarly acting function to collect processed data
from lists of arbitrary length. For example, the mathematical function + can be
used in the following way:

(defun add-them-all (numbers-in-a-list)

(if (null numbers-in-a-list) 0
(+ (first numbers-in-a-list)
(add-them-all (rest numbers-in-a-list)))))
? (add-them-all '(1 2 3 4 5 6 7))

Note that instead of using nil as a final argument for if in line 2 here, this function
supplies the number 0, a final addition to the accumulating number.
Interestingly, there are other ways of computing the addition of all of the num-
bers in a list. For example, the code

? (apply #'+ '(1 2 3 4 5 6 7))

07_DAS_Chap6_pp231-274 1/29/09 1:36 PM Page 274

274 HIDDEN STRUCTURE

produces the same result of 28, as did the function add-them-all. Note that
in this case, the function + appears as an argument to another function—apply—
requiring the special prefix of #' to declare it so.
The function apply also has many other uses. For example, apply in the follow-
ing example has a special effect:
? (apply #'append '((1)(2)(3)(4)()()()))
resulting in
(1 2 3 4)
thus reducing one level of sublists from a list of sublists including, in this case, the
removal of a series of empty lists as well. So popular is this particular technique
that it is often called the apply-append trick. However, one should take care not
to use this trick too often, for it typically signals the correction of a mistake. After
all, programmers should produce output in the right form in the first place, rather
than correcting it after the fact. Common Lisp contains many interesting features,
and interested readers should again review the sources listed in the bibliography
to learn more (see especially Graham 1995; Steele 1990; Touretzky 1990; Wilensky
1986).

CONCLUSIONS
This chapter has explored the potentials of Principle 4—all patterns, scales, and
functions in music (see Chapter 5) are best understood by modeling their
processes—as discussed in detail in Chapter 1 of this book. With Chapters 2 and 3
concentrating on Principle 1 (all music consists of patterns), Chapter 4 dedicated
to Principle 2 (all pitch patterns can be reduced to scales), and Chapter 5 con-
cerned with Principle 3 (all elements of scales have different functions), all four
of the principles of this book have been explored in some detail. Chapter 7, then,
covers many of the future possibilities of computer music analysis and how they
too might relate to these four principles. Unlike previous chapters, Chapter 7,
though organized around several central themes, covers diverse ground—areas of
mathematics and artificial intelligence, for example. I believe that each individual
topic, however, will provide fertile subject matter for important future research in
computer music analysis.
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 275

SE VE N
A Look to the Future

The prevailing image of the present intellectual epoch, perhaps soon to be sup-
planted, is that of the machine, interpreted in its widest sense to subsume for-
malizations of any kind. The work of Frege, Russell, Gödel, Hilbert, Carnap, and
so many others laid the foundations of the current civitas mentis machinosae,
leading naturally through Church’s Thesis and Turing. . . . To explicate something
is, ultimately, to formalize it, that is, to make it into a machine at whose
metaphorically whirring and clicking parts we are happy to stare, and be enlight-
ened. As a child of my epoch, this is my belief.

John Rahn, “On Some Computational Models of Music Theory” (1989), 663

In the preceding chapters, I have concentrated on one or two ideas shared between
music and computer science, artificial intelligence, or mathematics. This chapter,
by contrast, covers a wide range of diverse topics grouped according to general
themes or categories. Many readers may thus find this chapter more postulatory
than conclusory. However, this book attempts more than the persuasion of a partic-
ular view of music analysis; rather, it argues for a coalescence of music analysis
with computers in ways that will hopefully unleash the full potentials of both. I of-
fer no apologies for the smattering of approaches to analysis presented in this
chapter. In some ways, this smattering may represent one of the more important
presentations in this book, providing a diversity of ideas, one or more of which
might encourage readers in pursuing their own particular version of the revolution-
ary new field of computer music analysis.

PRINCIPLES
I begin this chapter with what might seem to some a digression; however, my point
will become clear as the chapter proceeds. In 1654 the noted mathematician Blaise
Pascal discovered or formulated—depending on your point of view—the now-
famous Pascal’s triangle. Although earlier manuscripts from Asia present various
versions of this mathematical construct, shown in Figure 7.1, Pascal was the first to
truly begin to reveal its secrets. This simply stacked set of numbers offsets each

275
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 276

276 HIDDEN STRUCTURE

1 1

1 2 1

1 3 3 1

1 4 6 4 1

1 5 10 10 5 1

1 6 15 20 15 6 1

1 7 21 35 35 21 7 1

1 8 28 56 70 56 28 8 1

1 9 36 84 126 126 84 36 9 1

FIGURE 7.1 The top ten levels of Pascal’s triangle.

successive line such that the addition of the two numbers to the upper left and up-
per right of any number add to produce the centered-between-them number below.
One must imagine the numbers to the left or right of the outer 1s as virtual 0s.
Thus, the second row (two 1s) results from the addition of 0 and 1 and 1 and 0 from
the first row. The third row begins similarly, but the middle number is 2, the sum of
the upper-left and upper-right 1s from row 2. Following each successive row will
make this process evident. The simplicity of the resulting infinitely expanding trian-
gle, however, is deceiving, for this assemblage of numbers hides an enormous con-
centration of significant mathematical information.
The following collection of observations represents but a small fraction of the
various sequences and calculations the triangle contains:
1. The sum of each row results in increasing powers of 2 (i.e., 1, 2, 4, 8, 16, 32, and
so on).
2. The 45° diagonals represent various number systems. For example, the first diag-
onal represents units (1, 1 . . .); the second diagonal, the natural numbers (1, 2,
3, 4 . . .); the third diagonal, the triangular numbers (1, 3, 6, 10 . . .); the fourth di-
agonal, the tetrahedral numbers (1, 4, 10, 20 . . .); and so on.
3. All row numbers—row numbers begin at 0—whose contents are divisible by that
row number are successive prime numbers.
4. The count of odd numbers in any row always equates to a power of 2.
5. The numbers in the shallow diagonals (from 22.5° upper right to lower left)
sum to produce the Fibonacci sequence (1, 1, 2, 3, 5, 8, 13 . . .), discussed in
Chapter 4.
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 277

A LOOK TO THE FUTURE 277

6. The powers of 11 beginning with 0 produce a compacted Pascal’s triangle (e.g.,

110 = 1, 111 = 11, 112 = 121, 113 = 1331, 114 = 14641, and so on).
7. Compressing Pascal’s triangle using modulo 2 (remainders after successive divi-
sions of 2, leading to binary 0s and 1s) reveals the famous Sierpinski gasket, a
fractal-like diagram made up of various-sized triangles, as shown in Figure 7.2,
with the 0s (Figure 7.2a) and without the 0s (Figure 7.2b), the latter presented to
make the graph clearer.
Therefore, while initial reactions to Pascal’s triangle may suggest that it offers very
little, when carefully considered, the triangle actually yields an immense amount of
valuable information. I believe that many of the musical ideas and processes de-
scribed in this chapter will prove similarly useful.

MATHEMATICS
Data in the form of numbers, no matter how those numbers are organized or what
they might represent, offer potential for analysis using any mathematical process.
Certainly, when such numbers represent music, they invite more traditional musi-
cal analyses. However, like Pascal’s triangle, this data may also reveal valuable in-
formation when analyzed in ways that apparently have nothing whatsoever to do
with music. I will here present several examples of such techniques that, although
still experimental, have provided tangible results in my research.
Mathematical formulae represent a powerful way of manipulating numerical data
to achieve results virtually impossible using any other method. For example, c =
␲r2 helps us calculate the circumference of a circle by knowing the constant ␲ or
3.14 . . . and the current radius of that circle. The formula e = mc2 relates the square
of the speed of light (c) to mass (m) and energy (e) and represents one of the more
important mathematical abstractions of the twentieth century. Most formulae such
as these derive from observations of data that then lead mathematicians to formu-
late symbolic representations and reductions of these observations. Music can of-
ten be described using similar formulae. For example, the circle of fifths can be
represented mathematically by the formula ƒx = mod(12) (x + 7), an example of Prin-
ciple 1, and all of the equal-tempered scales within a given octave can be derived
from binary representations from 1 to 211 (see Chapter 4 and Principle 2). Com-
posers (e.g., see Morris 1987; Xenakis 1971) have generated music from mathemati-
cal sources that should in turn encourage analysts to analyze the music of these
composers using these same processes. Analysts (e.g., Rahn 1980; Benson 2006)
have also proposed mathematical procedures for analyzing post-tonal music.
Guerino Mazzola (2002), in his book The Topos of Music, attempts to describe music
purely in terms of mathematical, physical, and psychological terms.
When Milton Babbitt (1961), Allen Forte (1973), and others began using mathe-
matical set theory (see Chapter 3) to analyze post-tonal music (Principle 1), many
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 278

278 HIDDEN STRUCTURE

a.
a)
1
11
101
1111
10001
110011
1010101
11111111
100000001
1100000011
10100000101
111100001111
1000100010001
11001100110011
101010101010101
1111111111111111
10000000000000001
110000000000000011
1010000000000000101
11110000000000001111
100010000000000010001
1100110000000000110011
10101010000000001010101
111111110000000011111111
1000000010000000100000001
11000000110000001100000011
101000001010000010100000101
1111000011110000111100001111
10001000100010001000100010001
110011001100110011001100110011
1010101010101010101010101010101
11111111111111111111111111111111
100000000000000000000000000000001
1100000000000000000000000000000011
10100000000000000000000000000000101
111100000000000000000000000000001111
1000100000000000000000000000000010001
11001100000000000000000000000000110011
101010100000000000000000000000001010101
1111111100000000000000000000000011111111
10000000100000000000000000000000100000001
110000001100000000000000000000001100000011
1010000010100000000000000000000010100000101
11110000111100000000000000000000111100001111
100010001000100000000000000000001000100010001
1100110011001100000000000000000011001100110011
10101010101010100000000000000000101010101010101
111111111111111100000000000000001111111111111111
1000000000000000100000000000000010000000000000001
11000000000000001100000000000000110000000000000011
101000000000000010100000000000001010000000000000101
1111000000000000111100000000000011110000000000001111
10001000000000001000100000000000100010000000000010001
110011000000000011001100000000001100110000000000110011
1010101000000000101010100000000010101010000000001010101
11111111000000001111111100000000111111110000000011111111
100000001000000010000000100000001000000010000000100000001
1100000011000000110000001100000011000000110000001100000011
10100000101000001010000010100000101000001010000010100000101
111100001111000011110000111100001111000011110000111100001111

FIGURE 7.2 Pascal’s triangle shown stacked to the left and modulo 2 with the ze-
ros identifying the Sierpinski gasket (a), and zeros removed (b) to
make the graphic more readable.
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 279

A LOOK TO THE FUTURE 279

b
b.)
1
11
1 1
1111
1 1
11 11
1 1 1 1
11111111
1 1
11 11
1 1 1 1
1111 1111
1 1 1 1
11 11 11 11
1 1 1 1 1 1 1 1
1111111111111111
1 1
11 11
1 1 1 1
1111 1111
1 1 1 1
11 11 11 11
1 1 1 1 1 1 1 1
11111111 11111111
1 1 1 1
11 11 11 11
1 1 1 1 1 1 1 1
1111 1111 1111 1111
1 1 1 1 1 1 1 1
11 11 11 11 11 11 11 11
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
11111111111111111111111111111111
1 1
11 11
1 1 1 1
1111 1111
1 1 1 1
11 11 11 11
1 1 1 1 1 1 1 1
11111111 11111111
1 1 1 1
11 11 11 11
1 1 1 1 1 1 1 1
1111 1111 1111 1111
1 1 1 1 1 1 1 1
11 11 11 11 11 11 11 11
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1111111111111111 1111111111111111
1 1 1 1
11 11 11 11
1 1 1 1 1 1 1 1
1111 1111 1111 1111
1 1 1 1 1 1 1 1
11 11 11 11 11 11 11 11
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
11111111 11111111 11111111 11111111
1 1 1 1 1 1 1 1
11 11 11 11 11 11 11 11
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1111 1111 1111 1111 1111 1111 1111 1111

FIGURE 7.2 continued

08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 280

280 HIDDEN STRUCTURE

analysts found their work controversial or even inappropriate. Today, however, al-
most every college music theory program in the United States uses set theory in
the study of contemporary music. Most of today’s analysts find set theory indis-
pensable for deciphering the post-tonal music they study. Interestingly, set theory
represents only one of the numerous subfields of discrete mathematics, many of
which could be put to similar use in music. The following partial list of discrete
mathematical categories seem particularly apt for analyzing music:

Cryptography—the mathematical study of information transmission (Principle 1)

Complexity theory—the study of input size as it relates to computer time and
memory requirements (Principle 1)
Combinatorics—the study of the possible orderings of finite collections of
numbers (Principles 1 and 2)
Game theory—a formal modeling approach to maximize results in game play-
ing (Principle 4)
Graph theory—the study of mathematical structures used to model relations
between points in graphs (Principles 3 and 4)
Probability theory—the study of probability (Principle 4)
Logic—the study of formal systems relating to inference (Principle 4)
Number theory—the study of numbers in general and integers in particular
(Principles 1–4)

As mentioned earlier in this book, musical analysis itself represents a kind of

cryptography. Complexity theory has already been discussed to a degree in this
book (see Chapters 2 and 6). I have found that combinatorics (notably in relation
to my work with recombinant music under the rubric of Experiments in Musical In-
telligence) and graph theory particularly relevant to music. Combinatorics calcu-
late the number of ways of logically combining sets of numbers or objects. At the
smallest level, all composers use combinatorics by combining pitches in various
ways; it is the fundamental principle behind all musical composition. However,
the methods composers use to combine motives, phrases, sections, and so on also
contribute to composition and, as such, deserve study in much the same way that
musical sets do.
Game theory and music have a closer relationship than one might imagine. For
example, highly formalistic music provides interesting correlations with board
games, offering, as they both do, obedience to rules as well as a certain freedom of
choice. Graph theory calculates the ways in which objects of any type relate or
connect to one another. Such connections could certainly include the time-linear
connections between adjacent musical ideas, the nonlinear connections when ideas
are separated by significant time spans (the analysis of musical form), and so on.
Although I have not gone very deeply into these last-mentioned subjects here, the
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 281

A LOOK TO THE FUTURE 281

various sub-disciplines of discrete mathematics offer many uses for the analysis of
music in any style, with computers making these potentials more feasible.
Probability theory (see Hiller and Isaacson 1959) has already been discussed to
some degree in this book in Chapter 6. Logic supports set theory and many other
discrete mathematical processes by proving logical propositions given certain
facts. Number theory represents a broad range of discrete properties of numbers
and number patterns and thus possesses significant potential for music, as re-
vealed later in this section.
Thus, music expressed as numbers represents an extraordinary opportunity for
music analysts. Even with a small amount of data, one can produce almost endless
discoveries. Many of these may seem more recreational than analytical, but they
nonetheless can produce intriguing results. For example, the square root of the
number 12,321 is 111; the square root of 1,234,321 is 1111; and so on. The center
number of the palindrome represents the number of 1s in the square root. Another
simple puzzle involves multiplying any number by 9 and then adding the single dig-
its of the resulting number together until a single digit is reached. This process al-
ways produces 9, as in 9 × 127 = 1143 = 9, and so on.
Magic squares also belong in the arena of number theory. Magic squares consist
of matrices in which each square contains a unique member of an incremental se-
quence beginning with 1 and ending with the number representing the number of
squares in the matrix. Once completed correctly, each horizontal and vertical rank
or column of a magic square adds to the same number. Figure 7.3 presents a simple
five-by-five magic square in which all horizontal rows and vertical columns sum to
65 (even the diagonals in this particular magic square sum to 65). There are several
ways to create magic squares. For example, beginning by placing the number 1 any-
where in the otherwise empty matrix and placing incrementally advancing numbers
upward diagonally produces the desired result. When an occupied square is
reached, choosing the square directly below the current square for the next num-
ber allows the numbers to continue. When moving off the matrix, treating the ma-
trix as a torus by inserting the number in the square at the side opposite where it
would normally go extends the process. The magic square in Figure 7.3 results from
Chess-knight moves.
One way to use magic squares musically involves utilitzing nonsymmetrical
geometries while maintaining the principles so important to the magic square’s def-
inition. For example, in a two-by-six matrix such as

e 9 7 t 6 8
0 2 4 1 5 3

both rows add to 6 horizontally (the result of adding the digits of 51 and 15 together)
and all six columns add to 11 vertically. Of course, the second horizontal row repre-
sents a mirror inversion of the top horizontal row, and such mirror inversions always
produce equivalent additions such as this. Although not a magic square per se, this
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 282

282 HIDDEN STRUCTURE

combination of numbers nonetheless fulfills many of the requirements of equivalent

additions while having a semblance of musical meaning.
Anton Webern’s Concerto for Nine Instruments (op. 24; see Figure 3.23) provides
a more interesting example of a magic rectangle. Viewing the opening four trichords
in terms of pitch range covered (left to right) and pitch class (top to bottom) pro-
duces interesting results.

B B D = 13
E G F = 13
G E F = 13
C C A = 13
e t 2
3 7 6
8 4 5
0 1 9
22 22 22

Webern follows this opening row statement using a similar process:

D B B = 13
F F E  = 13
F E G  = 13
A C C = 13
2 t e
6 7 3
5 4 8
9 1 0
22 22 22

This is the retrograde of the original. Ensuing trichords no longer follow this
process. Granted that Webern’s use of the original, inversion, retrograde, and retro-
grade inversion of these trichords tends to produce these kinds of relationships,
the correlations between the two sets of matching additions presents a quite con-
vincing case for Webern’s conscious or subconscious application of such number-
theory concepts to his musical thought processes. This work—and most others of
Webern’s mature style—also presents an example of music with very low informa-
tion content (see Chapter 2).
One could argue, of course, that this example represents but one isolated in-
stance of such a combinatorial magic-square occurrence. However, if we assume
this as fact, a musical magic rectangle would be an even more important notion,
clearly distinguishing this Webern work from the many others he composed using
similar serial techniques.
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 283

A LOOK TO THE FUTURE 283

11 24 7 20 3

17 5 13 21 9

23 6 19 2 15

4 12 25 8 16

10 18 1 14 22

FIGURE 7.3 A simple 5 × 5 magic square in which all horizontal ranks and vertical
columns sum to 65.

Webern clearly favored particular formalisms that resemble magic squares. His
favorite was:
S A T O R
A R E P O
T E N E T
O P E R A
R O T A S
The Latin here translates as “The sower Arepo holds the wheels with effort” (see
Bailey 1991, 21) and indicates Webern’s love for symmetry. This love is further
demonstrated musically in his use of combinatorial twelve-tone rows and integral
serialism of dynamics, rhythm, articulations, and tempos. However, whether or not
composers used such techniques in their compositions, if such patterned struc-
tures exist in their music, analysts should attempt to find and evaluate them. Com-
puters can make discovering these pattern contributions in music much easier.
There are many other ways to incorporate magic squares in music analysis. For
example, Figure 7.4 presents a magic square containing intervals that sum to 5 in
both top-to-bottom and left-to-right directions. Note that the numbers here repre-
sent intervals and not interval classes, hence the negative numbers indicating
downward motion. I have used this particular magic square in several of my com-
positions, assigning the horizontal and vertical combinations to six-pitch motives.
The resultant melodic line(s) obviously share the same tessitura—the perfect
fourth—but because each motive retains its own distinct interval content, each
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 284

284 HIDDEN STRUCTURE

w 1
w w
&w bw w
-2 0 7 9 -9

11 -7 -5 2 4
w w w
&w w bw -1 6 13 -10 -3

-8 -6 1 8 10
bw
w bw w
&w w 5 12 -11 -4 3

w
&w w w
#w w
w
w #w w nw
&w

w bw w
&w bw w

w
&w w w w w

#w w
w w w
&w

w w w w
&w #w

w w
&w bw w nw
FIGURE 7.4 A magic square containing intervals that equate (when added to-
gether) to 5 in top-to-bottom and left-to-right directions, along with
musical examples.
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 285

A LOOK TO THE FUTURE 285

melodic line maintains its own separate identity. Interestingly, the number of rows
and columns—including the diagonals—totals 12, and thus the entire collection can
be used as a row offering the same prime, retrograde, inversion, and retrograde-
inversion possibilities that any twelve-tone matrix normally would.
The music in Figure 7.5 from the second movement of my Concerto for Violon-
cello and Orchestra realizes the intervals appearing in the magic square shown
in Figure 7.4 as top-down horizontal continuities. Readers can follow the perfect-
fourth ambitus combined motion of each set, moving as they do every five inter-
vals from C to F, from F to A-sharp (B-flat), from A-sharp to E-flat, from E-flat to
A-flat, and finally from A-flat to C-sharp (D-flat). These combinations provide a level
of cohesion to this passage that it might otherwise lack.
Figure 7.6 presents a magic cube with both incremental numbers and music in-
tervals (with directions). Magic cubes require that the additions in each magic
square and each vertical column of successive layers of magic squares match, so
that straight lines in all three dimensions sum to the same single-digit number.
Entrepreneurs of magic squares have also created magic circles, spheres, and all
manner of other symmetrical and asymmetrical matrices, producing some of the
most beautiful and curious designs in the world of mathematics (Pickover 2002).
The Fibonacci sequence (0, 1, 1, 2, 3, 5, 8, 13 . . .) discussed in Chapter 4, in
which each new number in the sequence results from the addition of the two previ-
ous numbers, has provided a model for many composers, most notably Debussy
and Bartók (see Madden 2006). Dividing any number in the sequence by its prede-
cessor converges on a special number called the golden mean or golden section
(roughly 1.61803399 . . .), which has been used by painters, architects, and others as
a benchmark for structure. A good example of the golden mean used in music ap-
pears in a formal analysis of the first eighty-nine measures of Bartók’s Music for
Strings, Percussion and Celeste. Here, the two primary divisions separate into group-
ings of fifty-five and thirty-four measures, respectively (Lendvai 1983, 74). The first
fifty-five-measure grouping then contains two sub-sections of thirty-four and
twenty-one measures, with the first of these subsections further dividing into two
sections of twenty-one and thirteen measures. Even the second main section di-
vides into two groups of thirteen and twenty-one measures, numbers that represent
contiguous members of the Fibonacci sequence. Figure 7.7 presents all of these
subdivisions.
I often use two variants of the Fibonacci sequence in my own compositions.
These variants are formed by carrying Fibonacci’s principle one step further by
adding numbers further back in the sequence. For example, the sequence (0, 1, 2, 2,
3, 5, 7, 10, 15, 22, 32 . . . , seeded with 0, 1, 2) results from adding the second num-
ber previous to a current number and the current number to create a following
number. This sequence converges approximately on the number 1.47—what I call
the silver mean. The sequence (0, 1, 2, 3, 3, 4, 6, 9, 12, 16, 22, 31, 43, 59 . . . , seeded
with 0, 1, 2, 3) results from adding the third number previous to a current number
and the current number to create the following number and roughly converges on
the number 1.38—what I call the bronze mean.
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 286

286 HIDDEN STRUCTURE

bœ œ.
3 œ 3 bœ œ
&c ˙ j œ ˙ œ œ œ œ œj j # œ œj œ œ J œ ‰ j
œ bœ œ œ ˙
bœ. œ œ
3

FIGURE 7.5 The author, Concerto for Cello and Orchestra, 2nd movement (excerpt).

1 2

-2 0 7 9 -9 -8 -6 1 8 10

11 -7 -5 2 4 5 12 -11 -4 3

-1 6 13 -10 -3 -2 0 7 9 -9

-8 -6 1 8 10 11 -7 -5 2 4

5 12 -11 -4 3 3 -1 6 13 -10 -3

11 -7 -5 2 4

-1 6 13 -10 -3

-8 -6 1 8 10

5 12 -11 -4 3

-2 0 7 9 -9
4 5
5 12 -11 -4 3 -1 6 13 -10 -3

-2 0 7 9 -9 -8 -6 1 8 10

11 -7 -5 2 4 5 12 -11 -4 3

-1 6 13 -10 -3 -2 0 7 9 -9

-8 -6 1 8 10 11 -7 -5 2 4

FIGURE 7.6 A magic cube with both incremental numbers and music intervals
(with directions).
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 287

A LOOK TO THE FUTURE 287

55 34

34 21 13 21
89
21 13

FIGURE 7.7 A formal analysis of the first 89 measures of Bartók, Music for Strings,
Percussion and Celeste (Lendvai 1983, p. 74), based on the golden
mean and the Fibonacci series.

Inverting the process of division (i.e., dividing the previous number by the fol-
lowing number) produces approximately 0.62 (Fibonacci), 0.68 (silver), and 0.73
(bronze), with these latter successively larger numbers often representing both
phrase-length relationships and points of arrival in my music. I use particularly the
silver mean to gauge tension relationships within phrases (i.e., the grouping having
the most tension is placed at approximately 0.68 percent of the length of the phrase).
Another way to make such sequences musically viable is to apply modulo 12
(see Chapter 3) to each number in the sequence to create pitch classes. The result-
ing scales can then be compared with scales from actual music. As an example,
the Fibonacci sequence limited to numbers below fifty produces the sequence
[0,1,2,3,5,8,9,t]. Interestingly, when a modulo 12 scale is derived from a very large
sequence, it often includes all pitch classes but pitch class 6. However, the silver
variation, with very large sequences, produces the scale [0,1,2,3,4,5,6,7,9]. Interest-
ingly, the measures shown in Figure 7.8, the first fourteen measures of Varèse’s Den-
sity 21.5, produce the same prime-form pitch-class set as the silver variation, with
but one non-scale tone present.
General systems theory assumes that a system—defined as any analytic method
that operates on data in a continuous unfolding of syntactic relationships—
represents a dynamic collection of cooperating and collaborating parts. Hiller and
Levy note that
an analytic method here will be thought of as a system which describes an oper-
ational process. That is, it will describe how something was operated on to pro-
duce something. In other words, analysis is itself a system which is used to
describe some other system. (Hiller and Levy 1984, 297)
and
because a musical system is a process with changes of state constantly taking
place, it is also an aggregate of energy states resulting from the application of
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 288

288 HIDDEN STRUCTURE

q = 72 3
& c œ œ #œ ˙. œ #œ. #œ œ ˙ ˙ Œ nœ œ #œ
F f F p f F
4 3 3
3 3 3
j
& œ œ œ # œj œ # œ œ œ œ œ . œ # œ œ ‰ # œj œ- œ- œ b œ ˙ œ œ bœ
f p F p subito

8
j
& œ bœ. ˙ bœ œ œ ˙ œ ≈ œ bœ œ œ œ œ œ œ œ
J
f ƒ F subito
3 3

œ bœ. nœ. ˙ œ
. œ # œJ œ 3 # œj œ . # œ œj # œ .
11

& n˙ #œ œ J ‰ Œ
Ï 3
f 3
ƒ
FIGURE 7.8 The first 14 measures of Varèse, Density 21.5, with much the same
pitch classes as the silver variation.

some given class of syntactic decisions to some given class of information in or-
der to effect a change in another system—the listener. (Hiller and Levy 1984, 297)

In effect, general systems theory covers dynamically changing behavior that affects
a current state of data. An advocate of general systems theory might argue that
what currently passes for music analysis represents just another method of de-
scribing the superficial aspects of music and is little better than another type of
music notation. This same individual then might argue that general systems theory
attempts to discover the principles that make music itself a system. General sys-
tems theory would consider music not a static score, but an active process that un-
folds over time, with its various parts acting and interacting in dynamic ways.
Thus, what may be an accurate representation of musical analysis at one point dur-
ing a musical composition can change as future events occur; a single point in time
would then have many different interpretations depending on the unfolding of the
musical events that surround it.
The completion concept presented in Chapter 6 of this book could provide a
very simple verification process of a general systems theory analysis to a given
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 289

A LOOK TO THE FUTURE 289

point in a work. For example, if an extension of a phrase—an extension created by

using rules based on the current analysis—proved close enough to the actual mu-
sic the composer chose, it would help confirm the accuracy of the previous analy-
sis. If such an extension proved off course, then a different approach would be
required, one using different groupings, sets, scales, and so on. With each verifica-
tion point and its previous analysis saved and subsequently reviewed, this process
could provide a useful narrative of the way in which analyses of music change in the
minds of audiences, analysts, theorists, and even composers during composition.
General systems theory would seem a perfect match for computer music analy-
sis, though no one has yet successfully produced any significant analytical work us-
ing its processes. Aaron Copland remarks that “a composition is, after all, an
organism. It is a living, not a static, thing. That is why it is capable of being seen in
a different light and from different angles by various interpreters or even by the
same interpreter at different times” (Copland 1957, 269).
Other mathematical processes such as fractals have proven relevant to certain
types of musical analyses, particularly musical form (see particularly Madden 1999
and 2006). Fractals are self-imitating patterns occurring on different scales or di-
mensions. A work that begins, for example, with the pitches A–B–A and then uses
that template for beginning each phrase of that work, each section of that work,
and so on would present a good example of fractals used hierarchically. Several
movements of Tom Johnson’s Formulas for String Quartet (1994) are explicitly frac-
tal in origin, and the composer writes eloquently of these origins in the program
notes to this work.
Fuzzy logic offers analysts the ability to solve problems that have several vari-
ables. For example, chromatic note spelling, modulation, key identification, set rela-
tions, and so on all share the complexities created by numerous competing criteria.
Discovering best groupings in post-tonal music poses incredibly difficult problems
that fuzzy logic can often resolve. Fuzzy logic accepts such interactive variables
and derives results compatible with each (see Cope 2005, chapter 3, for more infor-
mation and examples).
One could continue the list of possible mathematical processes with subjects
such as chaos theory, cellular automata, and genetic algorithms, the list being lim-
ited only by one’s imagination. Any mathematical technique—whether designed for
music analysis or not—can potentially produce useful and insightful results (again,
see Cope 2001, chapter 3), and offer music analysts a rich arsenal of techniques
with which to analyze music. Most of these processes were available previous to
the advent of computers, but were seldom used owing to the exhaustive amount of
time required to implement them. Computers now make the application of these
techniques possible in a fraction of the time paper and pencil required. Thus, math-
ematics may now be used in as many ways as are conceivable, regardless of their
apparent relevancy or irrelevancy to music.
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 290

290 HIDDEN STRUCTURE

ARTIFICIAL INTELLIGENCE
Like mathematics, artificial intelligence offers numerous processes that can poten-
tially advantage those music analysts willing to take the risks necessary to study
and use these processes. Neural networks, for example, offer music analysts a vari-
ety of ways in which to study music. The term neural in “neural network” derives
from neuron, the primary functional unit of the central nervous system, including
the brain. Neural networks receive input that triggers neuron substitutes called hid-
den units, hidden because their values do not reveal much about their contributions
to the neural network’s resultant output. Hidden units trigger initially random re-
sults. Neural networks then cycle through a series of forward feeds and/or back
propagations (reversed-direction flow) that alter hidden unit values until the output
matches or approximates the desired result. This training process involves present-
ing the network with a series of examples of problems and sample solutions for
these problems. Figure 7.9 shows a simple model of a neural network. The number
of forward feeds and/or back propagations necessary for neural networks to pro-
duce appropriate output values based on training varies depending on the complex-
ity of the data involved and the type of network used. Many different types of neural
networks exist. However, hundreds and often thousands of forward feeds and/or
back propagations are typically required for successful output to be achieved.
Neural networks typically consist of many input and output nodes. As Dolson
puts it,
A neural network is basically an interconnected set of simple computational ele-
ments, each of which is typically functionally identical. Each element or unit re-
ceives inputs from other units and produces a single output, which is some
simple function of its inputs. The similarity of this structure to that of a biologi-
cal neuron has led some authors to prefer the terms “neuron” to that of “unit.”
(Dolson 1991, 4)
Each of these neurons or units then has
a precise rule for determining the output of a unit: add together all of the
weighted inputs and let the output be some nonlinear function of the resulting
sum. To state the rule mathematically, we can let each of the N units in the net-
work be identified with a unique number between 1 and N. The output x, of the
i th unit in the network is given by
N
xi = ƒ ( ⌺ wij xj )
j=1

where
wij
is the weight from the j th unit to the i th, and ƒ is some nonlinear function. (Dol-
son 1991, 4)
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 291

A LOOK TO THE FUTURE 291

Output

Back
Propagation

Hidden Units

Input

FIGURE 7.9 A simple model of a neural network.

Variable numbers of hidden units in a neural network have variable numbers of

connections between inputs and outputs making the training process somewhat
flexible in design, but usually inflexibly hardwired in both their hardware and soft-
ware incarnations. Successful output relies on such neural network design. Once
“taught,” a neural network can efficiently solve new problems of the same general
type on which the neural network was trained.
Robert Gjerdingen notes that neural networks
offer elegant ways of dealing with multidimensional complexity of the type found
in polyphonic, harmonically oriented music. One can, for example, define a musi-
cal event as an input pattern of activation that is then transformed by a network
of interconnected processing units into an output pattern of activation represent-
ing an interpretation of the event. (Gjerdingen 1991, 138)

Many music analysts have used neural networks (Loy and Todd 1991) with varying
degrees of success. Data representation and input can pose very difficult problems,
especially for highly contrapuntal music. Nonetheless, neural networks continue to
offer extraordinary potential for music analysis.
Association networks resemble neural networks in some respects (see Cope
2005, chapters 9–10). However, although neural networks compare output to input
values through interconnected “hidden” units, association networks do not make
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 292

292 HIDDEN STRUCTURE

such comparisons, nor do they have hidden units. The nodes in association net-
works can be accessed at any time, usefully revealing their weights for comparison
to the weights of other nodes. Although neural networks generally have fixed
numbers of nodes and connections, association networks can have virtually limit-
less numbers of interconnectable nodes, constrained only by their implementa-
tions. Although neural networks typically chain backward—back propagation—
association networks chain omnidirectionally.
Association networks can learn through a process not dissimilar to the Gradus
program (see Chapter 6; Cope 2005, chapter 5), though association networks use
very different mechanisms. Where Gradus utilizes a straightforward analysis algo-
rithm to create rules saved in variables that decrease mistakes, association net-
works invoke weighting shifts that slowly direct the program toward a particular
goal. The effect, however, is roughly the same—learning. Association networks also
offer opportunities for the integration of other important processes without the
need to write new programs or append other software. The natural negation and re-
inforcement approaches allowed by association networks further develop meaning-
ful links to inference and analogy (Cope 2005, chapter 5). Figure 7.10 presents a
very simple example of an association network. The circular nodes in this figure
may represent music or language, with output consisting of one or the other de-
pending on the current input type.
Association networks offer many opportunities for music analysis. In fact, the
creative processes by which association networks produce output depends on mu-
sical analysis. Because I discuss such analysis at great length in Cope 2005 (particu-
larly chapters 10 and 11), I will not repeat myself here, but merely point readers
toward this source. However, note that one of the features of data-driven programs
is that they always require some kind of analysis in order to operate. Thus, Experi-
ments in Musical Intelligence, Sara, Gradus, and Alice all have extensive analytic
components capable of producing (hopefully) interesting new music.
Genetic algorithms (GAs) represent another type of artificial intelligence process
with analytical capabilities. Genetic algorithms consist of virtual organisms that
evolve by procreating, sharing parental attributes (called crossover) in offspring
and responding to random mutations that affect the data and functions of these off-
spring. Genetic algorithms develop generationally in order to adapt over several
generations to achieve particular goals, tested en route by certain fitness criteria de-
scribed by the GA user. Genetic algorithms have proven very effective at pattern
recognition and have been successfully used in various industrial systems to analyze
problems and propose solutions to those problems. A number of composers have ex-
perimented with GAs to compose (see particularly Miranda 2001), but, unfortunately,
very little as of yet has been accomplished using GAs for music analysis.
Cellular automata (CAs) represent a subcategory of genetic algorithms. However,
CAs differ from genetic algorithms in that they do not have internal code. In one
sense, CAs represent a kind of visual counterpart to genetic algorithms. John Con-
way’s Game of Life (1970), for example, a very popular form of cellular automata
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 293

A LOOK TO THE FUTURE 293

FIGURE 7.10 A simple example of an association network.

08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 294

294 HIDDEN STRUCTURE

found at numerous Internet sites, has user-established initial states with cells
controlled by the current states of adjacent cells. Pascal’s triangle, presented in
Figure 7.1, represents a kind of CA, with previous states added together to create
new states.
Agents represent another category of artificial intelligence techniques that may
prove useful for musical analysis. Agents are software objects (OOP; see Chapter 5)
programmed to have internal structure—code in terms of both functions and
data—that recognize and react to other agents as well as elements in their environ-
ment. Typically, agents do not procreate (like genetic algorithms) or share data or
functions. Unlike cellular automata, which do not have internal functions or data,
agents can adapt to their surroundings, collaborate, and even produce social struc-
tures. Agents and multi-agents—agents of different types coexisting in the same
environment—have useful analytical abilities (Reis 1999). Depending on their inter-
nal structure, agents can locate patterns or processes that otherwise may go unde-
tected. Composers and music analysts have just begun to explore the possibilities
that agents and multi-agents offer (again, see Reis 1999).
John Holland uses agents in what he calls “complex adaptive systems” to de-
scribe a phenomenon known as emergence. Complex adaptive systems begin as
agents that have the ability to adapt and achieve goals beyond the sum of their in-
dividual parts. As Holland describes it:
The human immune system is a community made up of large numbers of highly
mobile units called antibodies that continually repel or destroy an ever-changing
cast of invaders called antigens. The invaders—primarily biochemicals, bacteria,
and viruses—come in endless varieties, as different from one another as snow-
flakes. Because of this variety, and because new invaders are always appearing,
the immune system cannot simply list all possible invaders. It must change or
adapt its antibodies to new invaders as they appear, never settling to a fixed con-
figuration. Despite its nature, the immune system maintains an impressive coher-
ence. Indeed, your immune system is coherent enough to provide a satisfactory
scientific definition of your identity. It is so good at distinguishing you from the
rest of the world that it will reject cells from any other human. As a result, a skin
graft even from a sibling requires extraordinary measures. (Holland 1995, 2)

No current complex adaptive system yet exists for musical analysis. However, one
can imagine such a system developing extraordinary new ways to analyze music.
Interestingly, the Emily Howell program (described in detail in Cope 2005) repre-
sents a complex adaptive system capable of generating what I feel is creative be-
havior; it has produced several works that demonstrate the emergent properties
described by Holland. Emily Howell is an association network that accepts both
words and music as input and analyzes both in order to produce new output. Un-
fortunately, aside from the complex community of weightings associated with all in-
put and output, actual analysis in Emily Howell is relatively difficult to observe, so
complex is the process necessary to actually produce new output. Developing
more transparent methods to clarify these processes may help make both complex
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 295

A LOOK TO THE FUTURE 295

adaptive systems and association networks more adaptable to computer music

analysis.
Studying emergence can further inform analysts about the origins of certain phe-
nomena that may in turn reveal important information about their constitution.
We are everywhere confronted with emergence in complex adaptive systems—
ant colonies, networks of neurons, the immune system, the Internet, and the
global economy, to name a few—where the behavior of the whole is much more
complex than the behavior of the parts. (Holland 1998, 2)

Describing a kind of emergent phenomena, Douglas Hofstadter comments on

how the human brain consists primarily of unintelligent neurons, and yet intelli-
gence still emerges:
Here we come back to the mysterious collective behavior of ant colonies, which
can build huge and intricate nests, despite the fact that the roughly 100,000 neu-
rons of an ant brain almost certainly do not carry any information about nest
structure. How, then, does the nest get created? Where does the information re-
side? . . . Somehow, it must be spread about in the colony, in the caste distribu-
tion, the age distribution—and probably largely in the physical properties of the
ant-body itself. That is, the interaction between ants is determined just as much
by their six-leggedness and their size and so on, as by the information stored in
their brain. (Hofstadter 1979, 359)

Complex adaptive systems and emergence should provide an enormously impor-

tant resource for music analysis in the future. One might imagine that such
processes might even explain music’s ineffability and help us to actually describe
the indescribable.

MUSE
The computer program Muse that accompanies this book on CD-ROM includes as-
pects of every program thus far described in this book, as well as several ideas pre-
sented thus far in this chapter. However, before detailing how Muse operates, I will
present a short description of semiotics, alluded to but not mentioned explicitly to
this point. Semiotics, particularly musical semiotics, offers several analytical princi-
ples that I find enormously useful when using computers to analyze post-tonal music.
Semiotics, a term coined by philosopher Charles Sanders Peirce, is the scientific
study of signs and symbols, particularly with reference to the physical world or the
world of ideas. Peirce began his study of semiotics in the 1860s, but it was not until
the turn of the century and thereafter that he began to clearly separate the roles
of the sign, its object, and its interpretant, the three dimensions that I will describe
in more detail shortly. In music, semiotics has revealed that certain gestures in mu-
sic, particularly of the classical period in Western music history, can be traced to
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 296

296 HIDDEN STRUCTURE

previous sources and their composers (Agawu 1991; Gjerdingen 1988). Often such
semiosis takes the form of quotation or allusion, as discussed in Chapter 5. These
patterns, like musical signatures, typically integrate seamlessly into their immedi-
ate environment and take on both syntactic and semantic value (see Cope 1996).
They often hold little interest in themselves, but great interest for what they tell us
about what the music being studied might mean, or, at the least, where it originated.
Musical semiotics also poses many important criteria for the analysis and under-
standing of music (Camilleri 1987; Eco 1976; Karbusicky 1979; Tarasti 1994), not the
least of which is its separation of music into three distinct categories:
Poietic dimension—reverse engineering a work, from the creator’s point of
view
Neutral dimension—systematic analysis from an explicitly stated, abstract
point of view
Esthesic dimension—analysis from the perceiver’s point of view.
It is the second of these dimensions that I find particularly important, at least for
the Muse software, which I will describe shortly. The neutral dimension of musical
semiotics suggests a ruthless analysis of music, achieved by breaking it into vari-
ously sized groupings and analyzing these groupings in as many ways as possible,
regardless of their promise of success. Jean-Jacques Nattiez describes this neutral
level:
It should suffice to remember that a neutral level is a descriptive level containing
the most exhaustive inventory possible of all types of configurations conceivably
recognisable in a score. The level is neutral because its object is to show neither
the processes of production by which the work unfolds (poietics) not the
processes of perception (esthetics) to which it gives rise. In this sense it provi-
sionally neutralises the poietic and esthesic dimensions of the piece. (Nattiez
1982, 244–245)

Dunsby (1982) adds that

in his [Nattiez’s] studies both of Syrinx and Density 21.5 he developed relatively
fleshless forms of enquiry, not only seeking to express the most neutral kinds
of articulatory picture of a piece of music, but also investigating the uses of a de-
scriptive, distributional account of musical information. In contrast to the ascetic
in semiotic analysis, the “replete” is to be found everywhere, appropriately
enough, and the more replete it is, the more it threatens the possibility of a
methodological practice. (Dunsby 1982, 236)

Nattiez’s article “Varése’s ‘Density 21.5’ ” (Nattiez 1982), referred to above, repre-
sents an excellent example of how one might segment a work into groupings by
partitioning it into consistent but arbitrary-length segments and analyzing the mu-
sic in however many ways possible, regardless of how appropriate the approach
might seem to either music in general or the work under study. Nattiez’s analysis
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 297

A LOOK TO THE FUTURE 297

demonstrates not only the validity of this approach, but also the obvious need for
computational means to assist in maintaining an accurate and relatively unbiased
approach. As Nattiez remarks, “What is appropriate to the neutral level is to make
an inventory of all analytical possibilities” (Nattiez 1982, 364).
Tarasti (1994) adds that
the structuralist method is characterized by study of the smallest significant
units of a sign system. Especially during the first phase of musical semiotics, in
the 1960s, semiotics was dominated by structuralism and the direct borrowing of
linguistic methods. It was thought that in music also one might distinguish the
units of the first articulation (meaningful items, musical “words”) and the second
articulation (musical “phonemes,” meaningless items). Through a sort of ars com-
binatoria, musical semioticians tried to build units of signification from these
small atoms. (Tarasti 1994, 5)
A computer program that can analyze musical data from as many arbitrary
points of view as possible would be enormously valuable, even if it only confirmed
what we already know. Such a program would, of course, be even more valuable if
it could produce insights that its users had not even previously considered. I will
now describe such a program, limited only by the data in its database.
The Muse program attempts to analyze post-tonal music as thoroughly as is con-
ceivable. Understanding how Muse accomplishes this thoroughness requires a brief
but important discussion of grouping processes. No matter how many different
ways a program might analyze post-tonal music, these analyses will remain suspect
if the manners in which musical groupings are collected have not been carefully
sorted out.
Grouping music into logical collections of harmonic, melodic, and/or combina-
tions of harmonic and melodic pitches poses one of the more difficult computa-
tional problems encountered thus far in this book. Although humans can visually
and aurally separate seemingly appropriate collections of pitches into various-sized
sets with little difficulty, computer programs—unless they are provided very strict
constraints—cannot themselves determine a logical from an illogical grouping. To
avoid such problems, the program I will soon describe here collects all combina-
tions possible, leaving it to users to decide which analysis seem most logical. Inter-
estingly, this brute-force approach has its advantages in that human analysts can
for various reasons overlook groupings that may ultimately provide the best mate-
rial for analysis. Therefore, regardless of the potential overkill in combinatorially
collecting all possible groupings of pitches, the approach I present here is thor-
ough, analyzing every group thus collected.
The Muse grouping process incorporates combinations of both harmonic and
melodic collection. The grouping process follows a simple guideline: pitches must
be contiguous. Figure 7.11 presents an example of this grouping process. Grouping
for our purposes here is limited to trichords only to conserve space, and the music
contains but two beats of three whole-note voices each. Note that each added beat
will exponentially expand the number of possible groupings. The grouping process
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 298

298 HIDDEN STRUCTURE

ww b ww ww b ww
1. 10.

& & w bw
w bw

ww b ww ww b ww
2. 11.

& & w bw
w bw

ww b ww ww b ww
3. 12.

& & w bw
w bw

ww b ww ww b ww
4. 13.

& & w bw
w bw

ww b ww ww b ww
5. 14.

& & w bw
w bw

ww b ww ww b ww
6. 15.

& & w bw
w bw

ww b ww ww b ww
7. 16.

& & w bw
w bw

ww b ww ww b ww
8. 17.

& & w bw
w bw

ww b ww ww b ww
9. 18.

& & w bw
w bw

FIGURE 7.11 An example of Muse’s grouping process.

08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 299

A LOOK TO THE FUTURE 299

here collects only those pitches that sound either simultaneously or immediately
one after another.
In order to better understand the combinatorial possibilities described above,
Figure 7.12 presents all of the possible trichordal permutations of a six-pitch list of
non-repeating numbers, not including the original arrangement ([0,2,7] [1,5,8]) and
its retrograde ([1,5,8] [0,2,7]). Note, however, that the second group of nine permu-
tations here represent reversed, our-of-order duplicates of the first nine and thus
can be removed, producing a total of nine discrete trichords. The principle here,

1. [8,0,1] [2,5,7]
2. [5,8,0] [1,2,7]
3. [0,1,5] [7,8,2]

4. [7,8,1] [0,2,5]
5. [5,7,8] [0,1,2]
6. [1,5,7] [8,0,2]
7. [8,1,2] [0,5,7]
8. [2,5,8] [0,1,7]

9. [1,2,5] [7,8,0]
10. [0,5,7] [8,1,2]

11. [2,5,7] [8,0,1]

12. [0,2,5] [7,8,1]
13. [7,8,0] [1,2,5]

14. [7,8,2] [0,1,5]

15. [8,0,2] [1,5,7]

16. [0,1,7] [2,5,8]
17. [1,2,7] [5,8,0]

18. [0,1,2] [5,7,8]

FIGURE 7.12 All of the possible permutations of a six-note list of nonrepeating

numbers not including the original arrangement ([0,2,7] [1,5,8]) and
its retrograde ([1,5,8] [0,2,7]).
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 300

300 HIDDEN STRUCTURE

that forward gathering will collect all possible unordered pitch-class groupings for
trichordal sets from six pitches, applies to groupings of all sizes. Note that the
number of permutations for a tetrachordal collection from an eight-pitch list is
thirty-five. Considering the trichordal collections previously mentioned and the
126 pentachordal collections from a ten-length list, the relation to Pascal’s triangle
described at the beginning of this chapter becomes clear: this is the third through
fifth numbers of the first vertical offset series to the left and right of center (10, 35,
126) in Figure 7.1.
The general thoroughness with which the Muse program gathers contiguous
groupings ensures that every possible grouping will be analyzed and rated (a
process to be described shortly). I remind readers here that Muse is only as useful
as the accuracy and amount of data available allows. Like many data-driven pro-
grams, the code required for Muse’s operation remains simple and quite small in
comparison to the large database of music used for analysis.
Because collections of pitches will produce groupings larger than four or five
pitches (i.e., ten, fifteen, even twenty pitches), I have limited the set-collecting pro-
gram to first-order collections only. By “first-order” here, I mean that the initial set
will combine only those pitch classes allowable by the established set size. Subse-
quent sets, however, will consist of sets following the order of the grouping itself
less any pitch classes included in the initial set. As example, a trichordal grouping
of nine pitch classes (1 2 3 4 5 6 7 8 9) might collect trichordal sets, such as
(1 2 3)(4 5 6)(7 8 9), (1 2 4)(3 5 6)(7 8 9), (1 2 5)(3 4 6)(7 8 9), and so on, with initial
trichords collecting combinatorially and the other trichords remaining mutually ex-
clusive but ordered as closely as possible to the original grouping order. A first-
order collection process indicates that only the first grouping has a specific order-
ing. An nth-order collection—one that figures the ordering of all sets—of a ten-length
source would generate nearly four million (3,628,800 or 10!) different possibilities,
and this number times four (or 14,515,200) for the collections of trichords, tetra-
chords, pentachords, and hexachords. However, a first-order collection of the same
source produces only 792 collections. Although thoroughness suggests that an
nth-order process creates more complete and potentially neutral collections, a first-
order process produces a more practical number of set groupings, even using com-
putational processes.
The task of analyzing even a very short passage of music with the very large
number of possible groupings available in Muse’s grouping approach makes finding
best analyses very difficult. However, the number of groupings represents signifi-
cantly less than half of the truly large number of groupings created by using several
controllers (see Chapter 5 for definition) for pattern-matching purposes. Such con-
trollers, with plausible ranges as wide as thirty or forty increments, increase the
number of possible analyses significantly. In fact, these controllers can seriously
impact the number of possible analyses, even with one incremental change of a sin-
gle controller. The possible variable-grouping problem can thus grow to astronomi-
cal proportions. As I will now describe, however, the Muse program handles this
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 301

A LOOK TO THE FUTURE 301

difficulty by using only one setting of all controllers for each run through all of the
possible analyses of the grouping combinations. This process reduces output from
several billion trillion for a single run to several thousand or—depending on the
size of the music analyzed—several million outputs.
The Muse program analyzes music by mapping an analysis program over all
possible trichordal, tetrachordal, pentachordal, and hexachordal contiguous-pitch
collections. Because this mapping process produces so many possible correct
analyses, Muse presents the results separately for each output. In effect, each map-
ping produces one grouping analysis, consisting of a rating and an analysis in a
readable print format for the passage being analyzed. Such analyses then appear
every few seconds continuously for hours, days, and even weeks, in some cases, to
allow users to read them as they appear. Because the Lisp Listener window keeps a
record of its contents, users can scroll back and view any previous result or save
and print the output or any portion thereof.
Figure 7.13 presents a view of the listener window after the completion of one
analysis by Muse. The rating appears first here. Ratings result from the figuring of
the optimal value of each analysis divided by the number of analyses. For example,
the best-case scenario for set analysis, only one of the many analyses applied,
would be that all analyzed sets equate to the same prime form. The worst-case sce-
nario for set analysis would be that each set differs from the rest. There is, of
course, a range of possibilities between these extremes, and that range of possibili-
ties has a number—the number of sets—resulting in a value of the number of equal
sets divided by the number of sets overall. Thus, a ten-set phrase with four equal
sets results in a 40 percent rating for that portion of the analysis. This rating repre-
sents just one of the many shown that contribute to the optimal rating.
The manner in which the analyses align with one another is another factor in the
optimal rating. I have included this alignment statistic because the coincidence of

The rating for this analysis is: 7.13

The analyzed sets are: ((91 74 63) (60 90 75) (64 59))((91 74 63) (60 90 75) (64 59))
The pitch information content for these sets is: (1.0 1.0 0.0 1.0 1.0 0.0)
The dynamics information content for these sets is: (0.33 0.33 0.5 0.33 0.33 0.5)
The durations information content for these sets is: (0.33 0.33 0.5 0.33 0.33 0.5)
The texture information content for these sets is: (1.0 1.0 1.0 1.0 1.0 1.0)
The scales for these sets is: ((3 4 6 11) (0 2 3 4 6 7 11))
The roots of the current sets are: ((91 75 64) (91 75 64))
The SPEAC for these sets are: ((A C C) (A C C))
The functions for these sets are: ((7 3 4) (7 3 4))

FIGURE 7.13 A view of the listener window after the completion of one analysis by
Muse.
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 302

302 HIDDEN STRUCTURE

various points of analysis provides a very important measurement of analytic cohe-

sion. For example, if changes in root, tension, set, scale, and other such aspects oc-
cur simultaneously, the coincidence of this simultaneity creates a deeper focus
than that generated by any of the analyses registered separately.
Investigating the code that produces the output in Figure 7.13 produces a curi-
ous function call (see Figure 7.14). Buried in Muse is a function called sleep. This
function does precisely what its name implies—it forces the program to pause a
certain number of seconds. In Muse, sleep sleeps for four seconds before printing
subsequent analyses to give users time to read each output before progressing to
the next. Without the call to sleep, Muse would—even though it has fairly compli-
cated routines to run for each analysis—print nearly continuously, making its out-
put unreadable while it runs. Therefore, this curious bit of cosmetic code slows the
program from its analysis rate of microseconds to produce more human-intelligible
results.
Many of the processes that Muse uses to analyze music rely on basic pattern-
matching techniques. Creating an unbiased pattern matcher for Muse requires that
this matcher be as flexible as possible. Ensuring that such a matcher can function
effectively also requires that the data in the database it matches be stored as sim-
ply as possible. Neither the program nor the data should be organized in ways that
reflect a programmer’s or a user’s sense of what would be musically viable or logi-
cal. In fact, if this matcher is to provide undiscovered insights into music, it should
be designed as generically as possible in order to discover all of the matches in the
numerical data under analysis.
Muse pattern matches musical data on several levels: pitch, interval, rhythm and
meter, dynamic, channel, and combinations of these parameters. The Muse pattern
matcher uses only two controllers (see Chapter 5), which give it flexibility in terms of
(1) pitch variances that follow scale-degree matches rather than pitch-class matches
and (2) ignored intervening pitches, such as embellishments between members of a
pattern. Muse’s pattern matcher further collects and attempts to separately match
data grouped according to channel, duration, beat relation, and so on.
All of these groupings are then cross-referenced in order to discover as many dif-
ferent patterns as possible, no matter how unlikely one might be to hear, see, or
otherwise perceive the pattern. The thoroughness with which Muse collects pat-
terns is in many ways as important as its pattern-matching processes.
One program or concept from every previous chapter in this book contributes to
the output report of Muse. From Chapter 1 Muse gains its sets perspective. From
Chapter 2 Muse includes an AIT report (Principle 1). The registral emphasis from
Chapter 3 also appears here, along with scale analysis (Principle 2) from Chapter 4.
Functions (Principle 3) are analyzed according to the principles laid out in Chapter 5,
and, although extensions (Principle 4) of the music provided for analysis do not
explicitly occur, the rules-collection process described in Chapter 6 influences the
overall rating provided by Muse. Thus, the program interlocks the various ap-
proaches to analysis that each chapter provides, all of them contributing to the
final output in a meaningful way.
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 303

A LOOK TO THE FUTURE 303

( def un COLLATE- SETS- AND- PRI NT ( s et - 1 s et - 2 &opt i onal ( s et - 2- s av e s et - 2) )

" The f unc t i on t hat del i v er s t he pr i nt out s of t he v ar i ous anal y t i c al i nf or mat i on. "
( l et * ( ( f i r s t - s e t ( mapc ar #' ( l ambda ( x ) ( mapc ar #' s ec ond x ) ) ( f i r s t s et - 1) ) )
(second-set ( mapc ar #' ( l ambda ( x ) ( mapc ar #' s ec ond x ) ) ( f i r s t s et - 2) ) )
(tes t ( append ( mapc ar #' anal y z e- and- gr aph ( mapc ar #' s et - t o- z er o ( f i r s t s et - 1) ) )
( mapc ar #' anal y z e- and- gr aph ( mapc ar #' s et - t o- z er o ( f i r s t s et - 2) ) ) ) )
( pi t c h- i nf o ( mapc ar #' v er y - f i r s t t es t ) )
( dy nami c - i nf o ( mapc ar #' v er y - s ec ond t es t ) )
( dur at i on- i nf o ( mapc ar #' v er y - t hi r d t es t ) )
( t ex t ur e- i nf o ( mapc ar #' v er y - f our t h t es t ) )
( s c al es ( l i s t ( v er y - f i r s t ( c r eat e- s c al e- l i s t i ngs ( appl y #' append ( f i r s t s et - 1) ) ) )
( v er y - f i r s t ( c r eat e- s c al e- l i s t i ngs ( appl y #' append ( f i r s t s et - 2) ) ) ) ) )
( r oot s ( l i st
( mapcar #' ( l ambda ( x) ( get - r oot x) )
( mapcar #' ( l ambda ( y) ( mapcar #' second y) ) ( f i r st set - 1) ) )
( mapcar #' ( l ambda ( x) ( get - r oot x) )
( mapcar #' ( l ambda ( y) ( mapcar #' second y) ) ( f i r st set - 2) ) ) ) )
( s peac ( mapc ar #' ( l ambda ( z ) ( get - s peac z ) )
( mapcar #' ( l ambda ( x) ( mapcar #' ( l ambda ( y) ( mod y 12) ) x) ) r oot s) ) )
( f unc t i ons ( mapc ar #' ( l ambda ( z ) ( get - f unc t i ons ( appl y #' append s c al es ) z ) )
( mapcar #' ( l ambda ( x) ( mapcar #' ( l ambda ( y) ( mod y 12) ) x) ) r oot s) ) )
( r at i ng ( my - r ound ( ov er al l - r at i ng pi t c h- i nf o dy nami c - i nf o dur at i on- i nf o t ex t ur e- i nf o
( mapccar #' ( l ambda ( n) ( * . 1 n) )
( r emove- dupl i cat es ( appl y #' append
( mapc ar #' ( l ambda ( x ) ( mapc ar #' ( l ambda ( y ) ( mod y 12) ) x ) ) r oot s ) ) ) )
( mapcar #' ( l ambda ( n) ( * . 1 n) )
( r emove- dupl i cat es ( appl y #' append
( mapcar #' ( l ambda ( x) ( mapcar #' ( l ambda ( y) ( mod y 12) ) x) ) r oot s) ) ) )
( mapcar #' ( l ambda ( n) ( * . 1 n) )
( r emov e- dupl i c at es ( appl y #' append f unc t i ons ) ) ) ) ) ) )
( cond ( ( nul l set - 2)
( c ol l at e- s et s - and- pr i nt ( r es t s et - 1) s et - 2- s av e s et - 2- s av e) )
( ( nul l set - 1) t )
( t ( pr ogn ( f or mat t " ~A~A~&" " The r at i ng f or t hi s anal y s i s i s : " r at i ng)
( f or mat t " ~A~A~A~&" " The anal y z ed s et s ar e: " f i r s t - s et s ec ond- s et )
( f or mat t " ~A~A~&" " The pi t c h i nf or mat i on c ont ent f or t hes e s et s i s : " pi t c h- i nf o)
( f or mat t " ~A~A~&"
" The dynami cs i nf or mat i on cont ent f or t hese set s i s: " dynami c- i nf o)
( f or mat t " ~A~A~&"
" The dur at i ons i nf or mat i on cont ent f or t hese set s i s: " dur at i on- i nf o)
( f or mat t " ~A~A~&" " The t ex t ur e i nf or mat i on c ont ent f or t hes e s et s i s : " t ex t ur e- i nf o)
( f or mat t " ~A~A~&" " The s c al es f or t hes e s et s i s : " s c al es )
( f or mat t " ~A~A~&" " The r oot s of t he cur r ent set s ar e: " r oot s)
( f o rr mat t " ~A~A~&" " The SPEAC f or t hes e s et s ar e: " s peac )
( f or mat t " ~A~A~&" " The f unct i ons f or t hese set s ar e: " f unct i ons)
( f or mat t " ~A~&" " " )
( f or mat t " ~A~&" " " )
( f or mat t " ~A~&" " " )
( f or mat t " ~A~&" " " )
( sl eep 4)
( c ol l at e- s et s - and- pr i nt s et - 1 ( r es t s et - 2) s et - 2- s av e) ) ) ) ) )

FIGURE 7.14 Muse’s program includes a function called sleep.

08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 304

304 HIDDEN STRUCTURE

Although I have just made a case for integration of the ideas presented in this
book, I maintain that these ideas do not constitute a distinct analytical theory. Nor
does this integration suggest that this particular collection of analytic approaches
represents a unified whole. To the contrary, the analyses that I have presented in
this book identify only a few of the many possible methods of analyzing post-tonal
music. However, the types of analysis presented in this book have significant use,
and their overall alignment posits a valuable way to understand post-tonal music.
Muse’s approach to analysis—based as it is on the neutral dimension, a system-
atic analysis from an explicitly stated, relatively objective point of view—does not
completely remove bias. Neutrality, however admirable as a goal, is very difficult to
achieve. John Cage reminded me often of his desire to rid his music of his biases
and of his resultant failure to achieve this goal. Coding a program that analyzes mu-
sic from as many points of view as possible, using techniques that may or may not
have relevance to the music under analysis, and ruthlessly applying these strate-
gies to every conceivable grouping of the music regardless of the logic that such
strategies or groupings may imply, does not, unfortunately, negate the fact that this
program—despite my attempts to minimize them—contains biases.
Cage points out the contradictions that occur when attempting to analyze music
using unbiased approaches:
I always want to start from zero and make, if I can, a discovery . . . it’s very diffi-
cult, because we have a memory. There’s no doubt of it. And we’re not stupid.
We would be stupid if we didn’t have a memory. And yet it’s that memory that
one has to become free of, at the same time that you take advantage of it. So it’s
very paradoxical. (Cage 1980, 6)

One might argue that the paradoxes Cage refers to here can produce interesting as
well as problematic results. For example, acquired musical skills should enhance
the user’s potential of computer programs to succeed. Truly unique discoveries,
however, can be achieved only by following paths as yet not taken. Using previ-
ously defined skills, however admirable those skills may be, increases, rather than
decreases, the chance of producing biased results. Rather than discovering the
unexpected—the immensely valuable insights that result from thoroughness and
tireless devotion to neutrality—one often simply verifies the biased assumptions
with which one began.
Some analysts have taken to using off-the-shelf applications (e.g., Microsoft Ex-
cel) to achieve the neutral results they desire. Using such programs means that mu-
sic analysis need not require programming expertise, specialized equipment, or
even much knowledge of computer operation to be of use. Anthony Pople com-
ments that his
program is implemented in the form of an extensive Visual Basic “plug-in” for
Microsoft Excel, with the simplified score representation taking the form of a
spreadsheet: pitch data are supplemented by indications of metrical stress . . .
and by a “vertical” segmentation into harmonic areas. (Pople 2004, 150)
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 305

A LOOK TO THE FUTURE 305

Interestingly, using such programs merely replaces one set of biases with another.
Although the apparent irrelevancy of using such programs might itself produce in-
teresting results—results that might otherwise have gone unnoticed—more likely
these different programs will simply produce another set of biased outputs—Cage’s
paradox perpetuated. Although nonetheless not without biases, the Muse program
described here will hopefully, by applying relevant but thorough analytical processes
to the music in its database, reveal previously undetected and useful results.

MUSICAL EXAMPLES
As discussed in Chapter 6, analyzing complete works composed by computer pro-
grams offers interesting challenges and important opportunities for music analysts.
Discovering what might be lacking in a stylistic replication can reveal significant
aspects of the original music upon which the computer bases its composition.
Likewise, verifying the presence of certain techniques in computer-composed mu-
sic can confirm their contributions in defining the style of the original human-
composed music. This is particularly valuable in analyzing post-tonal music, where
stylistic traits are not necessarily shared among the works of a composer’s oeuvre,
but rather are particular to a work or group of works of that composer.
As an example, the Eine Kleine Stück arguably in the style of Arnold Schoenberg
by Experiments in Musical Intelligence shown in Figure 7.15 demonstrates many
attributes of Schoenberg’s post-tonal compositional processes. For example, the
overall style of this brief work resembles the opening of Schoenberg’s op. 11, no. 1,
shown in part in Figure 7.16. However, unlike op. 11, no. 1, the music in this computer-
composed example does not evolve or develop in any serious way, but rather sim-
ply states ideas in various forms and then abruptly ends.
Eine Kleine Stück has serial elements but does not contain pitch or duration
rows in any formal way. The recombinancy algorithm in Experiments in Musical In-
telligence cannot recreate forms such as canons, fugues, or serialism. For these lat-
ter forms and processes, I use special algorithms described in my books (see Cope
1996 and 2005 in particular). Such algorithms did not contribute to the creation of
Eine Kleine Stück, and thus the initial and subsequent quasi-twelve-tone rows ap-
pearing in this work surprised me as much as it might surprise readers. These coin-
cidences occur occasionally with recombinancy producing them. That is, post-tonal
music, containing as it does the seeds for serialism, generates music of like-minded
formalisms that occasionally arrange themselves into formations that seem pre-
meditated rather than serendipitous.
The final repeating C major chord over an E-flat dominant seventh–sounding
chord seems more polytonal—more Stravinsky-like—than serial. Yet this computer-
composed work holds together in surprising ways, with similar three-pitch motives
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 306

306 HIDDEN STRUCTURE

Molto rubato q = 70
6
&4 ˙
˙. bœ œ. œ ˙ œ j ˙
œœ
œ œ j‰
#œ #œ J bœ. b œ- -̇ œ-
P f P F
b ˙˙ œœ . n œ ˙ . bœ ˙
? 46 Ó Œ b œœœ n ˙ œ . b œœ ˙˙ . Œ #˙ nw b ˙˙ ‰ b œœ ˙˙ ‰ j
J # ˙˙ b n ww ˙ J œ
p P p P F
b b œœ ˙˙ n ˙˙˙ œœœ œ
5 j ˙ bœ ˙ œœ œ œb œ
& Œ ‰ b ˙˙ œœ b ˙˙˙ Œ .
b ˙ œ J
bœ œbœ œ œœ ˙
Œ
f ƒ f 3

˙. œœ b œ ˙ . œœ ‰ œj ˙ ‰ j b œ b œ
? œ œ œ bœ b˙ œ œ b ˙˙
3
& œ bœ
b˙ ?
b˙ œ œœ ˙˙
. œ œ ˙
b œ ˙ pœ
ƒ f P
9 3
˙˙˙ b ˙˙ ˙
3 œ
‰ J bœ
& b b ˙˙˙ b ˙ b n ˙˙ œ œ œ ˙ Ó n ww
œœ b n œœ œœ ˙˙ bw nœ
F f ƒ Ï F3 p P F
? ˙˙˙ b b b ˙˙˙ n ˙˙˙ b n ˙˙ œ bœ œ œ
˙
œœ b n œœ œœ ˙˙ ‰ j œ b œ Œ ‰
œ œ œ ˙ J ˙˙˙ .
bœ .
P
3 3

œœ ˙˙ œœ œœ ˙˙ .. n˙
#œ ˙ #œ œ ˙ bœ
œœbœbœ
12

& ‰œ œ ‰ J
j ≈
œœœœ œœ ˙ œ
P f P 6
F
j b˙ nœ
‰ . n n œœ
6
? ‰ nœ. ˙ &‰ œ ˙ œ ˙ ?≈
œ
b bn www œ ˙
bœ. ˙
œ ˙ œ œ ˙
b œ ˙ b œ œœ ˙ #œ œ œ œ R
F P π F
œ bœ
15
bœ œ b˙.
& n ˙˙ . ‰ r≈ ≈ r‰ ‰ j Œ
œœ œœ œœ ˙˙ ww .
4
œ. œ. œ. ˙ w.
ƒ
? ˙˙ .. bœ
‰ J œ bœ bœ w. w.
˙. b œœ ww .. ww ..
f
FIGURE 7.15 The Eine Kleine Stück arguably in the style of Arnold Schoenberg by
Experiments in Musical Intelligence.
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 307

A LOOK TO THE FUTURE 307

3 j j
& 4 Œ œ #œ nœ. nœ nœ n˙ b˙ nœ Œ
n œœ b œ n n œœ œ
Œ n˙ Œ œ
p J
Piano
j Œ.
? 43 n˙ n œ # œ n œ
Œ n˙ #œ
∑
b˙ Œ b˙ Œ Œ
# n œœ œ.

FIGURE 7.16 Schoenberg, Three Piano Pieces, op. 11, no. 1 (opening).

abounding and consistently dissonant harmonies. Certain pitches evade appear-

ance: for example, pitch class 0 is missing from mm. 3 and 4, and pitch class 4 is
missing from mm. 7 and 8. Repeated tones appear at apparent cadence points.
Pitch-class set repetitions occur at times. For example, the pitch-class set [0,1,3]
begins in the right hand, followed by pitch-class set [0,1,3] in a left-hand chord.
These sets then move to pitch-class sets [0,1,4] and [0,2,5] before returning to
pitch-class set [0,1,3] in the chord near the middle of m. 2 in the left hand.
The description just provided represents a brief paper analysis of Eine Kleine
Stück. Using the various computer programs accompanying this book verifies many
of these findings. In fact, using Muse to analyze the two works presented here—
the Experiments in Musical Intelligence’s work and Schoenberg’s op. 11, no. 1—
produces very similar results. Unfortunately, the printouts of even just the more
likely successful attempts would require several dozen pages of this book to
demonstrate. I therefore invite readers to carry out such comparative analyses
themselves. Interestingly, the Experiments in Musical Intelligence Eine Kleine Stück
uses several small pieces by Schoenberg as a database, but not the music to which
the computer-generated work is compared here.
Analyzing computer-composed style replications for authenticity can prove ex-
tremely valuable. Even music theory students can make such analyses useful as
testimony to their understanding of this music. As computer composition in histor-
ical styles becomes increasingly sophisticated, both analysts and laypersons alike
will hopefully begin to better differentiate computer-composed style replications
from the originals on which they are based. Interestingly, however, these replica-
tions, controversial as they may be, are also originals.
Existing works revised by computer programs should also prove intriguing for mu-
sic analysts. A different version of, say, Beethoven’s Symphony no. 5 would provide
meaningful information about both Beethoven’s Symphony no. 5 and its computer re-
vision. The difference between these revisions and the music that Experiments in
Musical Intelligence otherwise composes is, of course, that the latter contains di-
rect quotes from only one work and develops that music in different ways from the
original. Figure 7.17 presents the first page of the score to a work created by Experi-
ments in Musical Intelligence in just the manner described here. Analyzing such
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 308

308 HIDDEN STRUCTURE

variations can be very helpful in more thoroughly understanding the original music
(see also the jigsaw test discussed in Chapter 6). Someday, computer-composed
variations may prove more interesting and musically satisfying than the human-
composed original, if one keeps an open mind to such possibilities.
Another example of post-tonal analysis continues the discussion of the third
movement of Stravinsky’s Three Pieces for String Quartet (1914) begun in Chapter 1
and continued in every chapter since. Pattern matching this work using Muse re-
veals that its opening section contains all pitch classes except pitch class 8. Bach’s
name (in the notes B-flat, A, C, and B-natural, which in German nomenclature spell
B–A–C–H) appears in the second violin’s first two notes and the violoncello’s third
and fourth notes and repeats several times thereafter. The principally stepwise
melodic material in each voice derives from variations of the Dies irae, a medieval
Latin sequence and the first words of the Requiem Mass (shown in Figure 7.18).
Like the Dies irae, Stravinsky’s music consists of mostly downward motion with
stepwise voice-leading. In their parallel motion, the top two voices also resemble
fourteenth-century organum.
Interestingly, thirteen measures before the interruption of the Mystic Circle of
the Young Girls in Stravinsky’s Rite of Spring, the music lapses into what later be-
came a signature chorale technique of Stravinsky’s: repeating fragments set against
polytonal harmonies with severely limited ranges (see Figure 7.19). This Rite of
Spring passage resembles the Cantique movement of the Three Pieces, with its lim-
ited leaps, repetitions, and falling chromatic motions. Were it not for the D-natural
in the third measure as part of a descending chromatic line in voice 2 here, this
orchestral passage would also contain only eleven notes, and it would suggest
F-sharp major and B major as possible contributing keys.
Another passage in Stravinsky’s Cantique resembles the Symphonies of Wind In-
struments, as shown in Figure 7.20. The three polychordal statements followed by a
more dissonant cluster resolving outward to the initial statement occur in both ex-
amples here—six years apart in composition, yet very close in concept.
My initial reaction to Stravinsky’s Three Pieces when I first heard them in my
teens—to label it the ugliest music I had ever heard—was matched by my even
more powerful attraction to the music’s dissonance and apparent lack of structure. I
listened to this music over and over again, hoping to reconcile the polar opposites
that confused me. I soon found that Stravinsky’s work and style had influenced my
own youthful compositional work and style, as shown in Figure 7.21, where the poly-
tonality and fragmentary repetitions resemble Stravinsky’s own, at least in certain
superficial ways. Using the analytical approaches discussed thus far in this book
has helped me to isolate the various contributors and better understand the com-
binations of the apparently disjunct materials in my own music.
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 309

A LOOK TO THE FUTURE 309

Flutes 1, 2
À& bbb 42 ∑
U
∑ ∑
U
∑ ∑

b U U
Oboes 1, 2 & b b 42 ∑ ∑ ∑ ∑ ∑

j U j U
Clarinets 1, 2 & b 42 ‰ œ œ œ ˙ ‰ œ œ œ ∑
ƒ ˙
in B b

Ã? bbb 42 U U
Bassoons 1, 2 ∑ ∑ ∑ ∑ ∑

À& bb 42 ∑
U
∑ ∑
U
∑ ∑
Horns
in F
b U U
& b 42 ∑ ∑ ∑ ∑ ∑

à U U
Trumpets 1, 2 & b 42 ∑ ∑ ∑ ∑ ∑
in B b
U U
? b 2 ∑ ∑ ∑ ∑ ∑
Timpani bb4

Violin 1
À& bbb 42 ‰ œ
j
œ œ
U
˙ ‰ œ
j
œ œ
U
∑
ƒ ˙
b 2 j U j U
&bb 4 ‰ œ œ œ ‰ œ œ œ ‰ œ œ œ
Violin 2 ˙ J
ƒ ˙ p
j U U
B b b b 42 ‰ œ œ œ ‰ j ∑
Viola ˙ œ œ œ
ƒ ˙
U U
? b 2 ‰ œ œ œ ˙ ‰ œ œ œ ∑
Cello bb4 J J ˙
ƒ
à U U
? b b 42 ‰ œ œ œ ˙ ‰ œ œ œ ∑
Contrabass b J J ˙
ƒ

FIGURE 7.17 The first page of the score to Beethoven, Symphony no. 5, revision.
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 310

310 HIDDEN STRUCTURE

Fl. 1, 2
À& bbb
6
∑ ∑ ∑ ∑ ∑

b
Ob. 1, 2 &bb ∑ ∑ ∑ ∑ ∑

Clar. 1, 2 &b ∑ ∑ ∑ ∑ ∑
in B b

Ã? bbb ˙ ˙ nœ œ œ ˙
Bn. 1, 2 ∑ ‰ J
p
À& bb ∑ ∑ ∑ ∑ ∑
Hns. 1, 2
in F
b
&b ∑ ∑ ∑ ∑ ∑

Tpt. 1, 2
in B b
Ã& b ∑ ∑ ∑ ∑ ∑

? b ∑ ∑ ∑ ∑ ∑
Timp. bb

À& bbb ‰ j
Vn. 1
pœ œ œ ˙ ˙ ˙ ˙

b j œ œ œ
Vn. 2 &bb ˙ ˙ ‰ œ œ œ ˙ œ

B bbb ∑ ‰
j
nœ œ œ ˙ ˙ ˙
Va.
p
? b ˙ ˙ nœ œ œ ˙
Vc. bb ∑ ‰ J
p
Cb. Ã? bbb ∑ ∑ ∑ ∑ ∑

FIGURE 7.17 continued

08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 311

A LOOK TO THE FUTURE 311

&œ œœ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ œ
œ œ œœ œ œ œ œ œ
Di– es i– rae di– es il– la Sol– vet saac– lum in fa– vil la les– te Da– vid cum Si– bil– la.

FIGURE 7.18 Dies irae, a medieval Latin sequence and the first words of the Re-
quiem Mass.

THE FUTURE
That computers will play ever-increasing roles in the analysis of music is, I believe,
undeniable. The burning question that remains, however, is exactly what these
roles will be. This book would not be complete without exploring those roles—
aside from the various analytical techniques described in Chapters 2 through 6 of
this book, and the preceding sections of this chapter—that I feel will most likely
develop over the next few decades.
Surely one of the most basic and immediate future developments in computer
music analysis will be improvements in existing music notation programs. For ex-
ample, having capabilities comparable to those of word processors would be an im-
portant enhancement. Although most computer notation programs can locate and
count measures, none of the most popular programs can return individual counts
of pitch classes, a feature that almost all word processors possess, typically reveal-
ing letter, word, line, and paragraph counts.
Music notation programs also capable of searching for particular patterns would
prove of immense value to composers, copyists, musicologists, and analysts alike.
Discovering pitch patterns—in transposition—both vertically and horizontally
would greatly enhance the ability of analysts and composers to find motives, other-
wise nearly impossible to discover by eye given the single-page-size screens of
most computers and our inability to see multiple pages in easily readable format.
Although our ears can often locate these patterns, carefully listening and counting
recognizable patterns can pose serious problems. Presenting discovered word pat-
terns in text processing programs has traditionally been considered a prerequisite
for a program, not an enhancement to one. The same ability to find musical pat-
terns should also be true for music notation programs.
Although most computer music notation programs offer some sort of algorith-
mic attempt to spell notes correctly, few offer the ability to locate questionable
pitch spellings and provide users an opportunity to make necessary changes.
Again, word processors not only find potentially misspelled words, but offer numer-
ous alternative spellings as well. Neither pitch spelling nor any of the other sugges-
tions above would require major overhauls of software on the part of commercial
music notation programs.
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 312

312 HIDDEN STRUCTURE

À & 46
a.

œ . œj œ # œ # œ œ œ œ . œj œ # œ # œ œ œ œ . œj œ # œ # œ œœ œ
Violin 1
#œ

Violin 2 & 46 # ˙ #˙ #˙ #˙ #˙ #˙ #œ nœ #œ nœ #œ nœ

Viola B 46 œ ˙ œ #œ œ œ ˙ œ #˙ ˙ œ œ #˙

#˙. nœ #˙. œ œ
Cello
à ? 46 # ˙ #˙ n˙ #˙ #œ

b.
5
& 4 b˙ 46 œ b œ 45
b˙ œœ œœ œœ œœ b b œœ ˙˙ œœ œbœ œœ ˙˙ œœ
π -

? 45 b ˙˙ b œœ n œœ b œœ b œœ œœ b ˙˙ n œœ 46 b œœ œœ b œœ ˙˙ n œœ 45
-
π
3
≤ ≤ ≤ ≤
œœ≤
4

& 45 œ b œ C œœ œœ œœ œœ ww
œ bœ œœ œœ œœ ˙˙ .. ˙˙ p
- F
≤ œ≤ œ≤ œ≤ b œœ≤ w
3

? 5 b œœ b œœ b œœ n œœ b œ œ
4 œ- b ˙˙ . ˙˙ C #œ œ œ œ #w
F p
FIGURE 7.19 (a) A passage from the Mystic Circle of the Young Girls in Stravinsky,
The Rite of Spring ; (b) a passage from Stravinsky, Three Pieces for
String Quartet.
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 313

A LOOK TO THE FUTURE 313

Cantique

a. Stravinsky
3

& 32 ˙˙˙ ˙˙˙ ˙˙˙ # ˙˙˙˙ www .

.
3
?3 ˙ ˙ ˙ #˙ w.
2 #˙
# ˙ ˙˙ ˙˙ # ˙˙ .
# ww .
#

b.

& 42 œœœ œœœ 34 ˙˙˙ b œœœ 42 ˙˙˙

œ- œ- -̇ œ- ˙

? 42 œ œ 34 ˙ œ 42 b ˙˙
b œœ œœ b ˙˙ œœ ˙
- - - -

FIGURE 7.20 (a) A passage from the 3rd movement of Stravinsky, Three Pieces for
String Quartet; (b) a similar passage in the Symphonies of Wind
Instruments.

e = 60
2 œ œ nœ
& 4 œœ œœ œœ œœœ œœœ n n œœœ œœœ 38 œœœ ‰ ‰ 24 œœœ œœœ n œœœ œœœ n n œœœ œœœ œœœ
J
Pœ
œ bœ œ œ œ bœ
œœ b b œœ œœ œœ b bb œœœ œœ œœ
? 42 œœ œœ b b œœ œœ œ bœ œ 38 œœœ ‰ ‰ 24 œœ œœ b b œœ œ œ œ
J

FIGURE 7.21 A post-tonal work by the author resembling Stravinsky’s style.

08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 314

314 HIDDEN STRUCTURE

In short, computer notation programs should model their engines on common

word processing program design. In broadening the scope of these tools, music no-
tation programs could produce, for example, counters that count particular chords,
finders that find particular chords, and spellers that appropriately spell particular
chords. Indeed, it would seem natural that such programs serve as a centerpiece
for musicians of all specialties, but certainly for music analysts.
On another front, most music analytical processes consist of three broadly de-
fined steps: reduction, representation, and functionalization. Reduction involves
reducing important notes from doublings, disparate registers, and so on. Represen-
tation provides a way to signify the ordered content of these reductions (e.g., the 56
arabic representations of intervals above the bass note in baroque figured bass).
These representations then qualify for certain musical functions defined according
to a local context such as key, mode, and so on (e.g., I or tonic). Formal and struc-
tural analyses typically follow these initial three steps, and understandably so, be-
cause these processes only make sense when we are attempting to relate musical
works to a model or to one another. Most of the analytical methods espoused in
this book follow similar techniques.
Unfortunately, reductional, representational, and functional analyses fail on sev-
eral counts that may be crucial for truly understanding music. For one thing, they
fail to identify the unique qualities of individual works of music. This should not
surprise us: reducing, representing, and functionalizing seek to discover commonal-
ities between works of music so that we might understand not only one work, but
a continuum of works. Although reductive, representational, and functional analy-
ses occasionally do reveal unique chords, forms, and structures in music, such
processes, by definition, eliminate much foreground material from analysis and
point their processes directly away from the unique. Searching for what makes par-
ticular works special requires approaches that reducing, representing, and func-
tionalizing typically fail to provide.
Discovering the differences between works of music often requires more dedica-
tion than finding the similarities between works, especially when such differences
occur subtly in orchestration, form, tempo, dynamics, and so on. Making such dis-
coveries possible while retaining current models should be one priority for future
computer music analysis. However, even for computer programs the problems of
detecting uniqueness remain complex. For example, to determine that a fragment
of music is unique requires that that fragment be compared with an enormous
number of other fragments to ensure that it has not occurred before. Otherwise, an
element that may initially appear unique—a turn of phrase, a sudden dynamic shift,
a particular rhythm, novel texture, or harmony—can trick us into believing it is
original, whereas in fact it has been used previously.
Despite the difficulties involved with analyzing the unique attributes of works of
music, the benefits of actually proving novel first instances in music far outweigh
the problems incurred. No longer will conjecture and assumption be required to de-
termine that certain chords (Tristan, Prometheus, etc.), scales (octatonic, whole
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 315

A LOOK TO THE FUTURE 315

tone, etc.), and even styles (those of Gesualdo, Berlioz, et al.) are unique. More
important, the status of certain works that have become renowned for their indi-
viduality (Debussy’s Prélude à L’après-midi d’un faune, Stravinsky’s Le sacre du print-
emps, etc.) can be verified.
Computers can also make possible what I term complementary analysis. Comple-
mentary analysis consists of discovering the musical ideas that a work does not
possess. Initially such analyses may seem frivolous. However, explicitly knowing
what a work does not contain can reveal as much of importance about that work as
discovering what it does contain. Imagine, for example, discovering that a classical-
period keyboard work of substantial duration does not contain the popular Vien-
nese simultaneous tonic/dominant cadence. And the importance of discovering a
late work by Bartók that lacks any references whatsoever to some known Hungar-
ian folk song could almost rival the significance of the discovery of all those works
of his that do contain such references.
At present, the only way to determine such omissions, even when using comput-
ers, requires locating every conceivable pattern in a work and then analyzing these
data for expected patterns. Heretofore, such processes have relied too heavily on
an analyst’s personal knowledge of musical patterns. Armed with appropriate data
and software capable of capturing all possible patterns from a large database of
music, a computer program should be able to fully analyze a work for the patterns
it does not include, possibly even the patterns it does not include but that one
would expect it to include.
One of the most important future goals for computer music analysts should also
be the creation of a universal database, one consisting of a standard data type and
interface. Like both Common Lisp and MIDI, such a universal database should be
the work of a committee charged with making the data as complete as possible, as
adaptable to as many existing and future programs as possible, and freely available
to all interested in using that data. As mentioned in Chapter 1 of this book, the
Digital Alternative Representation of Music Scores (DARMS), developed in 1963 by
Stefan Bauer-Mengelberg (see Bauer-Mengelberg 1970), continues today as an im-
portant computer representation program. DARMS code is extremely accurate and
currently appears in several dialects (Schüler 2000; Selfridge-Field 1997a). The In-
ternational Society for Music Information Retrieval (ISMIR) currently offers enor-
mous potential for the establishment of a common protocol for music storage and
retrieval for future computer music analysis.
A universal database should also have a universal pattern matcher. The Hum-
drum Toolkit developed by David Huron (1995 and 1999), a powerful Unix-based
music information retrieval computer program, searches for particular motives,
compares voice-leadings in various repertories, defines and catalogues harmonic
progressions, analyzes dissonance in relation to metric position, among other func-
tions. Expanding this toolkit to include searches that do not require an input
pattern (i.e., so that it can compare data with itself and return the patterns found in
common) would make this pattern matcher more effective.
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 316

316 HIDDEN STRUCTURE

A universal computer analysis program might seem oxymoronic in that its goal
of adaptability would necessarily make it more a computer language than an analy-
sis program. Nonetheless, creating an easy-to-use program/language to access both
the universal database and a pattern matcher would represent an extraordinary ad-
vance for the field of music analysis. To this point, however, too many well-meaning
individuals are still creating independent databases and programs whose life ex-
pectancy is often very short.
Another difficult but fairly achievable goal for the future would be the musical—
not waveform—analysis of sound files. The ability to analyze recordings for the
types of musical information discussed in this book using print-represented notes
would present an extraordinary opportunity for musical analysts to study patterns,
scales, function, and so on in the raw world of sound. Such processes could pro-
duce more accurate analyses of electronic and computer music, music of oral/aural
traditions, and other kinds of music not typically found in notated forms. Further-
more, the study of musical performance—a topic I have not treated in this book be-
cause of its obvious difficulty and complexity—would greatly benefit from such
possibilities.
Several other computer music analysis techniques seem natural for future explo-
ration. Space does not permit a complete explanation of these approaches, yet I
present them in brief here in hopes that others may pursue them further:

1. An analytical process for discovering the role of spectralism (compositional deci-

sions based on timbral structure) in music, a conscious or subconscious process
used in post-tonal composition.
2. Gestalt analysis—analysis that treats all musical parameters as interleaved rather
than as separate entities—likely paralleling the manner in which we perceive mu-
sic and thus providing a more natural approach to computer music analysis.
3. Transformational analysis of invariants—the computational study of how musical
transforms retain certain ideas while developing others over time.
4. Hermeneutic analysis—the computational study of the roles that nonmusical
ideas such as language, physical representations, cultural artifacts, personal
idiosyncrasies, and so on play in music.
5. Computer analysis of the roles that symmetry plays, particularly in post-tonal
music.
6. A computational search for formal commonalities in diverse works including be-
ginnings, endings, points, of arrival, and so on.
7. Creation of a musical first-order logic system ( predicate calculus) contained in
certain music or music as a whole made possible by computational processes
and a universal database.
8. Geometries of music, where structures acquire balance or imbalance according
to geometrical patterns.
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 317

A LOOK TO THE FUTURE 317

9. Computational discovery of narrativity versus stasis in particular post-tonal

works and styles.
10. Interactive analysis using computers in which analysts can test alternate strate-
gies for any parameter of music on the fly to observe new implications and,
possibly, why composers chose the solutions they did.
Each of these approaches to analysis, as imposing as some of them may seem, of-
fers potential as an extraordinary tool for the future understanding of music.

CONCLUSIONS
Day by day, the machines are gaining ground upon us. Day by day we are becom-
ing more subservient to them; more men are daily bound down as slaves to tend
them; more men are daily devoting the energies of their whole lives to the devel-
opment of mechanical life. The upshot is simply a question of time, but the time
will come when the machines will hold the real supremacy over the world and its
inhabitants is what no person of a truly philosophic mind can for a moment
question.
Samuel Butler, as quoted in Heinrich Schwarz, Art and Photography (1985), 97

Butler was not referring to the chess-playing computer program Deep Blue, Experi-
ments in Musical Intelligence, or even computers. He made this observation in 1863
about the invention of the camera. Phobias about using mechanical tools would
appear to predate computers by several decades, if not centuries, if one were to
freely interpret ravings from the distant past.
Interestingly, computer music analysis and photography share many of the same
characteristics. For example, the squeezing of a camera shutter easily parallels the
tap on the mouse or the key on the keyboard that many feel represents the only hu-
man activity associated with computationally analyzing music. Clearly, however,
both photography and computer music analysis represent legitimate human en-
deavors whose time has come.
Computers can relieve analysts of the drudgery of detail and allow them to make
more meaningful decisions about the musical universe in which they live. Every
new step in computational technology represents a potential new step for under-
standing music. Using computers to analyze music offers an incredible opportunity
for more deeply understanding how music moves us so profoundly and thus allows
us to more deeply appreciate its subtleties.
Throughout this book I have referred to the four principles first presented in
Chapter 1. I list them again here to allow readers to gauge whether or not the rele-
vant text and figures in the intervening chapters have adequately demonstrated
their veracity:
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 318

318 HIDDEN STRUCTURE

1. All music consists of patterns.

2. All pitch patterns can be reduced to scales.
3. All elements of scales have different functions.
4. All patterns, scales, and functions in music are best understood by modeling
their processes.
Obviously readers can and will decide the validity of these principles for analyz-
ing post-tonal music for themselves. However, regardless of the decisions they
reach, these four principles will at least have provoked new considerations of musi-
cal analysis and the music these analyses represent.
It took roughly half a millennium—from the invention of polyphony to Rameau—
for analysts to discover what they considered a more or less complete analytical
approach to tonal music. It then took approximately another two hundred years for
analysts to delve deeper into that so-called complete theory—from Rameau to
Schenker—to create another complete theory. However, we may have just begun to
understand tonal music, as indicated by recent discoveries in such areas as music
cognition and empirical musicology (Clarke and Cook 2004; Deutsch 1999; Krum-
hansl 1990; McAdams and Bigand 1993). Tonal music will hopefully continue to be
one of the foci of future analyses and theories.
As for post-tonal music, most analysts would agree that we have yet to seriously
scratch its surface. Computers do not themselves offer a better understanding of
post-tonal music; only humans can accomplish this. However, we can learn to bet-
ter analyze and appreciate this music from our attempts with this extraordinary
tool.
In this book I have presented a number of ways computers offer to enhance the
analysis of music. Many of these types of analysis would require so much time by
hand as to make them virtually impossible. With the increase of computer memory
size, computational speed, and accuracy, however, we can apply techniques to mu-
sic that hitherto might have seemed irrational or irrelevant in order to discover
more of its secrets. Every work of music deserves as complete an analysis as possi-
ble, for even the simplest music contains a wealth of important information that we
should not ignore.
I have always felt that at least part of the music we experience will never suc-
cumb to analysis, simply because of its personal, unique, and illusory qualities. Yet,
I also feel that we must continue to attempt to understand as much as we can
about music. Furthermore, we must truly value making the trip as much as we do
achieving the goal. After all, even incomplete knowledge enhances our appreciation
of music. In fact, this paradox of striving but never fully achieving makes music, at
least for me, the most engaging of all the arts.
A few months ago, I received the following e-mail message from a young man,
probably one still in high school:
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 319

A LOOK TO THE FUTURE 319

I am a young Catholic and science enthusiast, and recently I have heard a lot of
discussion in the media saying that science and religion are irreconcilable. Since
you are a famous scientist, I wanted to ask you a question. My physics teacher
told me that lots of scientists have been Catholic, like Pasteur, Mendel, von Neu-
mann, etc, and that they saw no apparent contradiction. Is it okay for me to be
Catholic and a scientist too, or do I need to renounce my faith to continue to
work in science? Thank you for your time.

I decided not to argue with his description of me as a “famous scientist,” because

I believe music to be a science as well as an art. I reply to most e-mail, so I responded:
Thank you for your note. Science is great for answering all of those questions
that can be answered. Religion is great for all of those questions that cannot be
answered by science. And there will always be both kinds of questions. I do not
view science and religion as contradictory, but as complementary.

This young man was probably disappointed by my response in that I did not di-
rectly deal with his apparent crises—what to do when science and religion both
provide answers, but those answers are contradictory. However, my response re-
flects my belief that for some questions, no answers actually exist. I do not at-
tribute this to human shortcomings, complexity, or any other simple explanation,
but to the universe itself.
To a certain degree, we all analyze music as we listen, even if our analysis con-
sists only of remembering and recognizing ideas as they vary and return in a work.
If we did not perform this kind of analysis, music would simply contribute one
more ambient sound to our sonic environment that we might then ignore. Taking
advantage of the resources that computers provide extends our abilities to compre-
hend and ultimately better appreciate the music that fills our lives in so many im-
portant ways. I believe that music analysts stand on a threshold for which new
computational tools will provide revelations not previously thought possible. We
should eagerly welcome any such tool that allows new insights into why music
finds such deep resonance within the human spirit.
08_DAS_Chap7_pp275-320 1/29/09 1:37 PM Page 320
09_DAS_Bibliog_pp321-330 1/29/09 1:38 PM Page 321

Bibliography

Agawu, V. Kofi. 1991. Playing with Signs. Princeton, NJ: Princeton University Press.
Agmon, E. 1989. “A Mathematical Model of the Diatonic System.” Journal of Music Theory
33/1: 1–25.
———. 1990. “Statistics and Compositional Balance.” Perspectives of New Music 28/1: 80–111.
———. 1992. “Quantifying Musical Merit.” Interface 21/1: 53–93.
Alphonce, Bo. 1980. “Music Analysis by Computer: A Field for Theory Formation.” Computer
Music Journal 4/2: 26–35.
Ames, Charles. 1989. “The Markov Process as a Compositional Model: A Survey and Tutor-
ial.” Leonardo 22/2: 175–87.
Aristoxenus. 1902. The Harmonics of Aristoxenus, translated by Henry S. Macran. New York:
Oxford University Press.
Avdeev, L. V., and P. B. Ivanov. 1993. “A Mathematical Model of Scale Perception.” Journal of
Moscow Physical Society 3: 331–53.
Babbage, Charles. 1864. Passages from the Life of a Philosopher. London: Longman, Green,
Longman, Roberts and Green.
Babbitt, Milton. 1955. “Some Aspects of Twelve-tone Composition.” The Score and I.M.A. Mag-
azine 12: 53–61.
———. 1960. “Twelve-Tone Invariants as Compositional Determinants.” Musical Quarterly
46/2: 108–21.
———. 1961. “Set Structure as a Compositional Determinant.” Journal of Music Theory 5/1:
129–47.
———. 1965. “The Use of Computers in Musicological Research.” Perspectives of New Music
3/2: 74–83.
Backus, J. 1969. The Acoustical Foundations of Music. New York: Norton.
Bailey, Kathryn. 1991. The Twelve-tone Music of Anton Webern. Cambridge: Cambridge Univer-
sity Press.
Balaban, Mira. 1992. “Music Structures: Interleaving the Temporal and Hierarchical Aspects
in Music.” In Understanding Music with AI: Perspectives on Music Cognition, edited by Mira
Balaban, Kemal Ebcioglu, and Otto Laske. Cambridge, MA: MIT Press: 110–38.
Bauer-Mengelberg, Stefan. 1970. “The Ford-Columbia Input Language.” In Musicology and the
Computer: Three Symposia, edited by Barry Brook. New York: City University of New York
Press: 48–52.

321
09_DAS_Bibliog_pp321-330 1/29/09 1:38 PM Page 322

322 HIDDEN STRUCTURE

Bailey, Kathryn. 1991. The Twelve-Tone Music of Anton Webern. Cambridge: Cambridge Uni-
versity Press.
Benade, A. H. 1976. Foundations of Musical Acoustics. London: Oxford University Press.
Benson, David J. 2006. Music: A Mathematical Offering. Cambridge: Cambridge University
Press.
Bent, Ian. 2002. “Steps to Parnassus: Contrapuntal Theory in 1725.” In The Cambridge History
of Western Music Theory, edited by Thomas Christensen. Cambridge: Cambridge Univer-
sity Press: 554–602.
Berger, Anna Maria Busse. 2002. “The Evolution of Rhythmic Notation.” In The Cambridge
History of Western Music Theory, edited by Thomas Christensen. Cambridge: Cambridge
University Press: 628–56.
Bernard, Jonathan W. 1977. A Theory of Pitch and Register for the Music of Edgard Varèse. New
Haven, CT: Yale University Press.
Bharucha, Jamshed. 1993. “MUSACT: A Connectionist Model of Musical Harmony.” In Ma-
chine Models of Music, edited by Stephan Schwanauer and David Levitt. Cambridge, MA:
MIT Press: 498–510.
Blombach, Ann. 1981. “An Introductory Course in Computer-Assisted Analysis: The Com-
puter and the Bach Chorales.” Journal of Computer-Based Instruction 7: 70–77.
———. 1982. “Harmony vs. Counterpoint in the Bach Chorales.” In Computing in the Humani-
ties, edited by Richard Bailey. Amsterdam: North-Holland: 79–88.
Boethius, A. M. T. S. 1967. De institutione musica. Introduction and translation by Calvin
Bower. Ann Arbor: University Microfilms.
Bower, Calvin. 2002. “The Transmission of Ancient Music Theory into the Middle Ages.” In
The Cambridge History of Western Music Theory, edited by Thomas Christensen. Cam-
bridge: Cambridge University Press: 136–67.
Böker-Heil, N. 1972. “Musikalische Stilanalyse und Computer: einige grundsätzliche Erwägun-
gen.” IMSCR 1972: 45–50.
Brook, Barry. 1965. “The Simplified ‘Plaine and Easy Code System.’ ” Fontes Artis Musicae 12:
156–60.
Brooks, Frederick, A. Hopkins, P. Newmann, and W. Wright. 1957. “An Experiment in Musical
Composition.” IRE Transactions on Electronic Computers 6: 175–82.
Browne, R. 1981. “Tonal Implications of the Diatonic Set.” In Theory Only 5/1–2: 3–21.
Bruner, Cheryl. 1984. “The Perception of Contemporary Pitch Structures.” Music Perception
2/1: 25–39.
Byrd, Donald. 1970. “Music Notation by Computer.” Computers and the Humanities 5/2: 111.
Cage, John. 1980. “An Interview with John Cage.” Composer 10/11: 6–22.
Camilleri, Lelio. 1987. “Towards a Computational Theory of Music.” In The Semiotic Web,
1986: A Yearbook of Semiotics, edited by Thomas A. Sebeok and Jean Umiker-Sebeok.
Berlin: Mouton.
Caplin, William. 2002. “Theories of Musical Rhythm in the Eighteenth and Nineteenth Cen-
turies.” In The Cambridge History of Western Music Theory, edited by Thomas Christensen.
Cambridge: Cambridge University Press: 657–94.
09_DAS_Bibliog_pp321-330 1/29/09 1:38 PM Page 323

BIBLIOGRAPHY 323

Carey, N., and K. Clampitt. 1989. “Aspects of Well-Formed Scales.” Music Theory Spectrum
11/2: 187–206.
Castine, Peter, Alexander Brinkman, and Craig Harris. 1990. “Contemporary Music Analysis
Package (CMAP) for Macintosh.” In Proceedings of the International Computer Music Confer-
ence, Glasgow, 1990. San Francisco, CA: Computer Music Association: 150–52.
Cazden, Norman. 1958. “Pythagoras and Aristoxenus Reconciled.” Journal of the American
Musicological Society 11/2–3: 97–105.
Chaitin, Gregory. 2001. Exploring Randomness. London: Springer-Verlag.
———. 2005. Meta Math! The Quest for Omega. New York: Vintage Books.
Clark, Eric, and Nicholas Cook, eds. 2004. Empirical Musicology: Aims, Methods, Prospects. Ox-
ford: Oxford University Press.
Clough, John. 1980. “Diatonic Interval Sets and Transformational Structures.” Perspectives of
New Music 18: 461–82.
Cohen, David. 2002. “Notes, Scales, and Modes in the Earlier Middle Ages.” In The Cambridge
History of Western Music Theory, edited by Thomas Christensen. Cambridge: Cambridge
University Press: 307–363.
Cook, Nicholas. 1987. A Guide to Musical Analysis. London: J. M. Dent and Sons.
———. 2004. “Computational and Comparative Musicology.” In Empirical Musicology, edited
by Eric Clarke and Nicholas Cook. Oxford: Oxford University Press: 103–26.
Cooper, Grosvenor, and Leonard Meyer. 1960. The Rhythmic Structure of Music. Chicago: Uni-
versity of Chicago Press.
Cope, David. 1990. “Pattern Matching as an Engine for the Simulation of Musical Style.” In
Proceedings of the International Computer Music Conference. San Francisco: Computer Mu-
sic Association.
———. 1991. Computers and Musical Style. Madison, WI: A-R Editions.
———. 1996. Experiments in Musical Intelligence. Madison, WI: A-R Editions.
———. 2000. The Algorithmic Composer. Madison, WI: A-R Editions.
———. 2001. Virtual Music. Cambridge, MA: MIT Press.
———. 2004. “A Musical Learning Algorithm.” Computer Music Journal 28/3:12–27.
———. 2005. Computer Models of Musical Creativity. Cambridge, MA: MIT Press.
———. 2006. “Capriccio for Cello and Orchestra. On Mozartballs, DVD, directed by Larry
Weinstein and performed by Steven Isserlis. New York: Decca 074 3153.
Copland, Aaron. 1957. What to Listen For in Music. New York: McGraw-Hill.
Crocker, Richard. 1958. “Musica Rhythmica and Musica Metrica in Antique and Medieval The-
ory.” Journal of Music Theory 2: 12–15.
———. 1962. “Discant, Counterpoint, and Harmony.” Journal of the American Musicological
Society 15/1: 1–21.
Desain, P., and H. Honing. 1999. “A Reply to S. W. Smoliar‘s ‘Modelling Musical Perception: A
Critical View.’ ” In Parallel Distributed Perception: Musical Networks and Performance,
edited by N. Griffith and P. Todd. Cambridge, MA: MIT Press: 111–16.
Deutsch, Diana, ed. 1999. The Psychology of Music. 2nd ed. San Diego: Academic Press.
09_DAS_Bibliog_pp321-330 1/29/09 1:38 PM Page 324

324 HIDDEN STRUCTURE

Devlin, K. 1993. The Joy of Sets: Fundamentals of Contemporary Set Theory. 2nd ed. New York:
Springer.
Dolson, Mark. 1991. “Machine Tongues XII: Neural Networks.” In Music and Connectionism,
edited by Peter Todd and Gareth Loy. Cambridge, MA: MIT Press.
Dunsby, Jonathan. 1982. “A Hitch Hiker‘s Guide to Semiotic Music Analysis.” Music Analysis
1/3: 235–41.
Ebcioglu, Kemal. 1992. “An Expert System for Harmonizing Chorales in the Style of J. S. Bach.”
In Understanding Music with AI: Perspectives on Music Cognition, edited by Mira Balaban, Ke-
mal Ebcioglu, and Otto Laske. Cambridge, MA: The AAAI Press/MIT Press: 295–333.
Eco, Umberto. 1976. A Theory of Semiotics. Bloomington: Indiana University Press.
Erickson, Raymond. 1968. “Music Analysis and the Computer.” Journal of Music Theory 12:
240–63.
———. 1969. “A General-Purpose System for Computer Aided Musical Studies.” Journal of
Music Theory 13: 276–94.
Ferreirós, J. 1999. Labyrinth of Thought: A History of Set Theory and Its Role in Modern Mathe-
matics. Basel: Birkhäuser.
Forte, Allen. 1973. The Structure of Atonal Music. New Haven: Yale University Press.
———. 1989. “Banquet Address: SMT, Rochester 1987.” Music Theory Spectrum 11/1: 95–99.
———. 1993. “A Program for the Analytic Reading of Scores.” In Machine Models of Music,
edited by Stephan Schwanauer and David Levitt. Cambridge, MA: MIT Press: 55–81.
Freund, John. 1973. Introduction to Probability. New York: Dover.
Gamer, C. 1967. “Some Combinational Resources of Equal-Tempered Systems.” Journal of Mu-
sic Theory 11: 32–59.
Gibson, Don. 1986. “The Aural Perception of Non-traditional Chords in Selected Theoretical
Relationships: A Computer Generated Experiment.” Journal of Research in Music Education
34/1: 5–23.
———. 1988. “The Aural Perception of Similarity in Non-traditional Chords Related by Octave
Equivalence.” Journal of Research in Music Education 36/1: 5–17.
Ghahramani, Saeed. 2005. Fundamentals of Probability. 3rd ed. Upper Saddle River, NJ: Pear-
son Prentice-Hall.
Gjerdingen, Robert. 1988. A Classic Turn of Phrase: Music and the Psychology of Convention.
Philadelphia, PA: University of Pennsylvania Press: 138–49.
———. 1991. “Using Connectionist Models to Explore Complex Musical Patterns.” In Music
and Connectionism, edited by Peter Todd and Gareth Loy. Cambridge, MA: MIT Press.
Gould, Murray, and George Logemann. 1970. “ALMA: Alphanumeric Language for Music
Analysis.” In Musicology and the Computer: Three Symposia, edited by Barry Brook. New
York: City University of New York Press: 57–90.
Graebner, Eric. 1974. “An Analysis of Schoenberg’s Klavierstück, Op. 33A.” Perspectives of
New Music 12 1/2: 128–40.
Graham, Paul. 1995. ANSI Common Lisp. Upper Saddle River, NJ: Prentice-Hall.
Gross, Dorothy. 1975. “A Set of Computer Programs to Aid in Music Analysis.” Ph.D. diss., In-
diana University.
09_DAS_Bibliog_pp321-330 1/29/09 1:38 PM Page 325

BIBLIOGRAPHY 325

———. 1980. “A Project in Computer-assisted Harmonic Analysis.” In Proceedings of the Inter-

national Computer Music Conference, edited by Hubert S. Howe. San Francisco: Computer
Music Association: 525–33.
———. 1984. “Computer Applications to Music Theory: A Retrospective.” Computer Music
Journal 8/4: 35–42.
Halmos, P. R. 1974. Naive Set Theory. New York: Springer-Verlag.
Harris, Craig, and Alexander Brinkman. 1989. “An Integrated Software System for Set-Theoretic
and Serial Analysis of Contemporary Music.” Journal of Computer-Based Instruction 16/2:
59–70.
Harrison, Daniel. 1994. Harmonic Function in Chromatic Music. Chicago: University of Chicago
Press.
Herik, H. J. van den. 2000. “From Cognition to Perception.” ICGA Journal 23/4: 201.
Hiller, Lejaren. 1964. “Informationstheorie und Computermusik.” Special issue, Darmstädter
Beiträge zur Neuen Musik 8.
Hiller, Lejaren, and Leonard Isaacson. 1959. Experimental Music: Composition with an Elec-
tronic Computer. New York: McGraw-Hill.
Hiller, Lejaren, and Burt Levy. 1984. “General System Theory as Applied to Music Analysis—
Part I.” In Musical Grammars and Computer Analysis, edited by Mario Baroni and Lelio Cal-
legari. Firenze, Italy: Musicologia A Cura Della Società Italiana de Musicologia: 295–316.
Hindemith, Paul. 1942. The Craft of Musical Composition. Book I: Theoretical Part. Translated
by Arthur Mendel. New York: Associated Music.
Hofstadter, Douglas. 1979. Gödel, Escher, Bach: An Eternal Golden Braid. New York: Basic Books.
———. 2001. “Staring Emmy Straight in the Eye—And Doing My Best Not to Flinch.” In David
Cope, Virtual Music. Cambridge, MA: MIT Press: 33–81.
Hofstetter, Fred. 1973. “Computer Applications to Music at OSU.” Computational Musicology
Newsletter 1/1: 2.
———. 1979. “The Nationalistic Fingerprint in Nineteenth-century Romantic Chamber Mu-
sic.” Computers and the Humanities 13 (1979): 105–19.
Holland, John. 1995. Hidden Order: How Adaption Builds Complexity. New York: Helix Books.
———. 1998. Emergence. Cambridge, MA: Perseus Books.
Honing, Henkjan. 1990. “POCO: An Environment for Analysing, Modifying, and Generating Ex-
pression in Music.” Proceedings of the 1990 International Computer Music Conference. San
Francisco: Computer Music Association: 364–68.
Howe, Hubert. 1965. “Some Combinational Properties of Pitch Structures.” Perspectives of
New Music 4/1: 45–61.
Huron, David. 1995. The Humdrum Toolkit: Reference Manual. Menlo Park, CA: Center for
Computer Assisted Research in the Humanities.
———. 1999. Music Research Using Humdrum: A User‘s Guide. Stanford, CA: Center for Com-
puter Assisted Research in the Humanities.
Isaacson, Eric J. 1996. “Issues in the Study of Similarity in Atonal Music.” Music Theory Online
2/7.
Johnson, Tom. 1994. Formulas for String Quartet. Paris: Editions 75.
09_DAS_Bibliog_pp321-330 1/29/09 1:38 PM Page 326

326 HIDDEN STRUCTURE

Karbusicky, Vladimir. 1979. Systematische Musikwissenschaft. Munich: Wilhelm Fink.

Klumpenhouwer, Henry. 2002. “Dualist Tonal Space and Transformation in Nineteenth-
century Musical Thought.” In The Cambridge History of Western Music Theory, edited by
Thomas Christensen. Cambridge: Cambridge University Press: 456–76.
Kostka, Stephan. 1969. “The Hindemith String Quartets: A Computer-assisted Study of Se-
lected Aspects of Style.” Ph.D. diss., University of Wisconsin.
———. 1971. “Recent Developments in Computer-assisted Musical Scholarship.” Computers
and the Humanities 6/1: 15–21.
Krumhansl, Carol. 1990. Cognitive Foundations of Musical Pitch. Oxford: Oxford University
Kuusi, Tuiro. 2003. “The Role of Set-class Identity in the Estimation of Chords.” Music Theory
Online 9/3.
LaRue, Jan. 1970. “New Directions for Style Analysis.” In Musicology and the Computer: Three
Symposia, edited by Barry S. Brook. New York: City University of New York Press: 194–97.
Laske, Otto. 1972. “On Musical Strategies With a View to a Generative Grammar for Music.”
Interface 1/2: 111–25.
———. 1973. “On the Methodology and Implementation of a Procedural Theory of Music.”
Computational Musicology Newsletter 1/1: 15–16.
———. 1974. “In Search of a Generative Grammar for Music.” Perspectives of New Music 8:
351–78.
———. 1992. “Artificial Intelligence and Music: A Cornerstone of Cognitive Musicology.” In
Understanding Music with AI: Perspectives on Music Cognition, edited by Mira Balaban, Ke-
mal Ebcioglu, and Otto Laske. Cambridge, MA: The AAAI Press/MIT Press: 3–28.
———. 1993. “In Search of a Generative Grammar for Music.” In Machine Models of Music,
edited by Stephan Schwanauer and David Levitt. Cambridge, MA: MIT Press: 215–40.
Lawson, Colin, and Robin Stowell, 1999. The Historical Performance of Music: An Introduction.
Cambridge: Cambridge University Press.
Lawvere, F. William, and Robert Rosebrugh. 2002. Sets for Mathematics. Cambridge: Cam-
bridge University Press.
Lefkoff, Gerald. 1970. “Automated Discovery of Similar Segments in the Forty-eight Permuta-
tions of a Twelve-tone Row.” In The Computer and Music, edited by Harry B. Lincoln.
Ithaca, NY: Cornell University Press: 147–53.
Lendvai, Ernő. 1983. The Workshop of Bartók and Kodály. Budapest: Editio Musica.
Lerdahl, Fred. 1989. “Atonal Prolongational Structure.” Contemporary Music Review 4: 65–87.
———. 2001. Tonal Pitch Space. New York: Oxford University Press.
Lerdahl, Fred, and Ray Jackendoff. 1983. A Generative Theory of Tonal Music. Cambridge, MA:
MIT Press.
———. 1993. “An Overview of Hierarchical Structure in Music.’ In Machine Models of Music,
edited by Stephan Schwanauer and David Levitt. Cambridge, MA: MIT Press: 290–312.
Levy, Azriel. 1979. Basic Set Theory. New York: Dover.
Lewin, David. 1987. Generalized Musical Intervals and Transformations. New Haven, CT: Yale
University Press.
09_DAS_Bibliog_pp321-330 1/29/09 1:38 PM Page 327

BIBLIOGRAPHY 327

Lieberman, Fredric. 1970. “Computer-aided Analysis of Javanese Music.” In The Computer and
Music, edited by Harry B. Lincoln. Ithaca, NY: Cornell University Press: 181–92.
Lincoln, Harry, ed. 1970. The Computer and Music. Ithaca, NY: Cornell University Press.
Lindley, Mark, and Ronald Turner-Smith. 1993. Mathematical Models of Musical Scales. Bonn:
Verlag für systematische Musikwissenschaft GmbH.
Lofstedt, John, and Ian Morton. 1970. “FORTRAN Music Programs Involving Numerically Re-
lated Tones.” In The Computer and Music, edited by Harry B. Lincoln. Ithaca, NY: Cornell
University Press: 154–62.
Lovelace, Ada. 1843. “Notes.” In Scientific Memoirs, Selected from the Transactions of Foreign
Academies and Learned Societies and from Foreign Journals, edited by Richard Taylor.
Vol. 3. London: art. 29, 666–731.
Loy, Gareth, and Peter Todd, eds. 1991. Music and Connectionism. Cambridge, MA: MIT Press.
Madden, Charles. 1999. Fractals in Music: Introductory Mathematics for Musical Analysis. Salt
Lake City, UT: High Art Press.
———. 2006. Fib and Phi in Music: The Golden Proportion in Musical Form. Salt Lake City, UT:
High Art Press.
Mathews, Max, Joan Miller, F. R. Moore, J. R. Pierce, and J. C. Risset. 1969. The Technology of
Computer Music. Cambridge, MA: MIT Press.
Maxwell, John. 1992. “An Expert System for Harmonizing Analysis of Tonal Music.” In Under-
standing Music with AI: Perspectives on Music Cognition, edited by Mira Balaban, Kemal
Ebcioglu, and Otto Laske. Cambridge, MA: The AAAI Press/MIT Press: 335–53.
Mazzola, Guerino. 2002. The Topos of Music. Berlin: Birkhäuser Verlag.
McAdams, Stephen, and Emmanuel Bigand, eds. 1993. Thinking in Sound: The Cognitive Psy-
chology of Human Audition. Oxford: Oxford University Press.
McCreless, Patrick. 2002. “Music and Rhetoric.” In The Cambridge History of Western Music
Theory, edited by Thomas Christensen. Cambridge: Cambridge University Press: 847–879.
McHose, Allen. 1947. Basic Principles of the Technique of 18th- and 19th-century Composition.
Englewood Cliffs, NJ: Prentice Hall.
———. 1951. The Contrapuntal Harmonic Technique of the 18th Century. Englewood Cliffs, NJ:
Prentice Hall.
Meyer, Leonard. 1956. Emotion and Meaning in Music. Chicago, IL: University of Chicago
Press.
———. 1989. Style and Music: Theory, History, and Ideology. Chicago, IL: University of Chicago
Press.
———. 2000. The Spheres of Music: A Gathering of Essays. Chicago, IL: University of Chicago
Press.
Miranda, Eduardo Reck. 2001. Composing Music with Computers. Oxford: Focal Press.
Morris, Robert. 1987. Composition with Pitch Classes. New Haven, CN: Yale University Press.
Nattiez, Jean–Jacques. 1982. “Varèse’s ‘Density 21.5’: A Study in Semiological Analysis.”
Translated by Anna Barry. Music Analysis 1/3: 243–340.
Newman, William. 1961. “Kirnberger‘s Method for Tossing Off Sonatas.” Musical Quarterly 23:
460–80.
09_DAS_Bibliog_pp321-330 1/29/09 1:38 PM Page 328

328 HIDDEN STRUCTURE

Palisca, Claude. 1980. “Theory, Theorists.” In The New Grove Dictionary of Music, edited by
Stanley Sadie. 20 vols. London and New York: Macmillan: 18:741–62.
Pascal, Blaise. 1963. Traité du triangle arithmétique. In Oeuvres complètes, edited by L. La-
fuma. Paris: Éditions du Seuil.
Patch, Howard. 1935. The Tradition of Boethius; A Study of His Importance in Medieval Culture.
New York: Oxford University Press.
Perle, George. 1990. The Listening Composer. Berkeley: University of California Press.
———. 1991. Serial Composition and Atonality. 6th ed. Berkeley: University of California Press.
———. 1996. Twelve-tone Tonality. 2nd ed. Berkeley: University of California Press.
Pickover, Clifford. 2002. The Zen of Magic Squares, Circles, and Stars. Princeton, NJ: Princeton
University Press.
Pierce, John. 1980. An Introduction to Information Theory: Symbols, Signals, and Noise. [A re-
vised version of Pierce’s Symbols, Signals, and Noise: The Nature and Process of Communi-
cation (Harper & Brothers, 1961).] New York: Dover.
Pople, Anthony. 2004. “Modeling Musical Structure.” In Empirical Musicology, edited by Eric
Clarke and Nicholas Cook. Oxford: Oxford University Press: 127–56.
Rahn, Jay. 1987. “Some Recurrent Features of Scales.” In Theory Only 2/11–12: 43–52.
Rahn, John. 1980. Basic Atonal Theory. New York: Longman.
———. 1989. “On Some Computational Models of Music Theory.” In The Music Machine,
edited by Curtis Roads. Cambridge, MA: MIT Press: 663–69.
Rameau, Jean-Philippe. 1971. Treatise on Harmony. Translated by Philip Gossett. New York:
Dover.
Ratner, Leonard. 1970. “Ars Combinatoria Chance and Choice in Eighteenth-century Music.”
In Studies in Eighteenth-Century Music: Essays Presented to Karl Geiringer on the Occasion of
His 70th Birthday, edited by H. C. Robbins Landon. New York: Oxford University Press.
Regener, Eric. 1967. “A Multiple-pass Transcription and a System for Music Analysis by Com-
puter.” In Elektronische Datenverarbeitung in der Musikwissenschaft, edited by Harald
Heckmann. Regensburg: Gustav Bosse: 89–102.
———. 1974. “On Allen Forte’s Theory of Chords.” Perspectives of New Music 13/1: 191–212.
Reis, Ben. 1999. “Simulating Music Learning with Autonomous Listening Agents: Entropy, Am-
biguity and Context.” Ph.D. diss. Cambridge: University of Cambridge.
Riemenschneider, Albert, collector and ed. 1941. 371 Harmonized Chorales and 69 Chorale
Melodies with figured Bass by J. S. Bach. New York: G. Schirmer.
Rink, John, ed. 1995. The Practice of Performance: Studies in Musical Interpretation. Cambridge:
Cambridge University Press.
Robison, Tobias. 1967. “IML-MIR: A Data-Processing System for the Analysis of Music.” In
Elektronische Datenverarbeitung in der Musikwissenschaft, edited by Harald Heckmann. Re-
gensburg: Gustav Bosse: 103–35.
Ruggiero, Charles, and James Colman. 1984. CASAP User’s Guide [software manual]. Okemos,
MI: Okemos Music Software.
Salzer, Felix. 1962. Structural Hearing: Tonal Coherence in Music. New York: Dover.
09_DAS_Bibliog_pp321-330 1/29/09 1:38 PM Page 329

BIBLIOGRAPHY 329

Schaffer, John. 1994. “Threader: A Computer Interface for the Graphic Entry, Encoding, and
Analysis of Musical Material.” Computer Music Journal 18/1: 21–29.
Schenker, Heinrich. 1979. Free Composition. Translated and edited by Ernst Oster. Originally
published as Der freie Satz (Vienna: Universal Edition, 1935). New York: Longman.
Schillinger, Joseph. 1946. The Schillinger System of Musical Composition. New York: Carl Fischer.
Schoenberg, Arnold. 1983. Theory of Harmony. Translated by Roy Carter. Berkeley: University
of California Press. Originally published as Harmonielehre, 3rd ed. (1922).
Schüler, Nico. 2000. “Methods of Computer-assisted Music Analysis: History, Classification,
and Evaluation.” Ph.D. diss., University of Michigan.
Schüler, Nico, and Dirk Uhrlandt. 1994. MUSANA 1.0 / 1.1: Ein Musikanalyseprogramm, Programm-
dokumentation. Peenemünde: Dietrich.
Schwanauer, Stephan. 1993. “A Learning Model for Tonal Composition.” In Machine Models of
Music, edited by Stephan Schwanauer and David Levitt. Cambridge, MA: MIT Press:
512–532.
Schwarz, Heinrich. 1985. Art and Photography: Forerunners and Influences. Layton, UT: Pere-
grine Smith Books.
Seife, Charles. 2006. Decoding the Universe. New York: Penguin Books.
Selfridge-Field, Eleanor. 1997a. “Humdrum and Kern: Selective Feature Encoding.” In Beyond
MIDI: The Handbook of Musical Codes, edited by Eleanor Selfridge-Field. Cambridge, MA:
MIT Press: 375–401.
———. 1997b. Beyond MIDI: The Handbook of Musical Codes. Cambridge, MA: MIT Press.
Shannon, Claude, and W. Weaver. 1949. The Mathematical Theory of Communication. Urbana:
University of Illinois Press.
Shirlaw, Matthew. 1955. “Claudius Ptolemy as Musical Theorist.” Music Review: 181–90.
Simon, Herbert and Richard Sumner. 1993. “Pattern in Music.” In Machine Models of Music,
edited by Stephan Schwanauer and David Levitt. Cambridge, MA: MIT Press: 83–110.
Simoni, Mary, ed. 2006. Analytical Methods of Electroacoustic Music. New York: Routledge.
Simonton, Dean. 1980. “Thematic Fame and Melodic Originality: A Multivariate Computer-
content Analysis.” Journal of Personality 48: 206–19.
Slonimsky, Nicolas. 1947. Thesaurus of Scales and Melodic Patterns. New York: Coleman-Ross.
Smoliar, Stephen. 1980. “A Computer Aid for Schenkerian Analysis.” Computer Music Journal
4/2: 41–59.
———. 1993. “Process Structuring and Music Theory.” In Machine Models of Music, edited by
Stephan Schwanauer and David Levitt. Cambridge, MA: MIT Press: 190–212.
———. 1999. “Modelling Music Perception: A Critical View.” In Parallel Distributed Perception:
Musical Networks and Performance, edited by N. Griffith and P. Todd. Cambridge, MA: MIT
Press: 97–110.
Solomon, Larry. 2003. Theory of Nonharmonic Tones. http://solomonsmusic.net/ Nonharm
.htm (accessed 31 December 2007).
———. 2005. Interval String Table for the Identification of Chords, Modes, Scales, and
Melodies. http://solomonsmusic.net/intstring.htm (accessed 31 December 2007).
09_DAS_Bibliog_pp321-330 1/29/09 1:38 PM Page 330

330 HIDDEN STRUCTURE

Standage, Tom. 2002. The Turk. New York: Berkley.

Steele, Guy. 1990. Common Lisp: The Language. 2nd ed. Bedford, MA: Digital Press.
Steinbeck, Wolfram. 1976. “The Use of the Computer in the Analysis of German Folk Songs.”
Computers and the Humanities 10: 287–96.
Straus, Joseph N. 2005. Introduction to Post-Tonal Theory. 2nd ed. Upper Saddle River, NJ:
Prentice Hall.
Suchoff, Benjamin. 1968. “Bartók, Ethnomusicology, and the Computer.” Institute for Computer
Research in the Humanities Newsletter 4: 3–6.
Tarasti, Eero. 1994. A Theory of Musical Semiotics. Bloomington: Indiana University Press.
Temperley, David. 2001. The Cognition of Basic Music Structures. Cambridge, MA: MIT Press.
———. 2007. Music and Probability. Cambridge, MA: MIT Press.
Temperley, David, and Daniel Sleator. 1999. “Modeling Meter and Harmony: A Preference-rule
Approach.” Computer Music Journal 23/1: 10–27.
Touretzky, David. 1990. Common LISP: A Gentle Introduction to Symbolic Computation. Red-
wood City, CA: Benjamin/Cummings.
Wason, Robert. 2002. “Musica Practica: Music Theory as Pedagogy.” In The Cambridge History
of Western Music Theory, edited by Thomas Christensen. Cambridge: Cambridge Univer-
sity Press: 46–77.
Weakland, Rembert. 1956. “Hucbald as Musician and Theorist.” Musical Quarterly 42: 66–84.
Werner, Eric. 1956. “The Mathematical Foundation of Philippe de Vitry’s Ars Nova.” Journal of
the American Musicological Society 9/2: 128–32.
Widmer, Gerhard. 1992. “The Importance of Basic Musical Knowledge for Effective Learning.”
In Understanding Music with AI: Perspectives on Music Cognition, edited by Mira Balaban,
Kemal Ebcioglu, and Otto Laske. Cambridge, MA: AAAI Press/MIT Press: 491–507.
Wilensky, Robert. 1986. Common LISPcraft. New York: W. W. Norton.
Williams, J. Kent. 1997. Theories and Analyses of Twentieth-Century Music. New York: Harcourt
Brace.
Winnington-Ingram, Reginald Pepys. 1929. “Ancient Greek Music: A Survey.” Music and Letters
10: 326–345.
———. 1932. “Aristoxenus and the Intervals of Greek Music.” Classical Quarterly 26: 195–208.
Wolfram, Stephen. 2002. A New Kind of Science. Champaign, IL: Wolfram Media.
Wright, Ernest. 1939. Gadsby. New York: Wetzel.
Xenakis, Iannis. 1971. Formalized Music. Bloomington: Indiana University Press.
Youngblood, Joseph. 1958. “Style as Information.” Journal of Music Theory 2: 24–35.
———. 1970. “Root Progressions and Composer Identification.” In The Computer and Music,
edited by Harry B. Lincoln. Ithaca, NY: Cornell University Press: 172–78.
10_DAS_Glossary_pp331-336 1/29/09 1:38 PM Page 331

Glossary

Acoustic root theory—See root theory.

Agogic accent—An accent that results from duration.
AIT—Acronym for algorithmic information theory, a branch of information theory
that concentrates less on the communication accuracy of information and more
on the precise amount of noncompressible information contained in a message.
Algorithm—A finite step-by-step process for solving a problem.
Algorithmic information theory—See AIT.
Ambitus—The distance between the highest and lowest pitch of a passage (range).
Argument—The value(s) passed to a computer language function as in (+ 1 2) in
Lisp where both 1 and 2 are arguments to “+.”
Artificial intelligence (AI)—The branch of computer science that attempts to digi-
tally duplicate or approximate human intelligence.
Association networks—Networks with virtually limitless numbers of universally in-
terconnectable nodes, constrained only by their implementation, and standing in
contrast to neural networks, which generally have fixed numbers of nodes and
connections.
Atonal—See post-tonal.
Back propagation—A method for training a neural network in which the initial sys-
tem output is compared to the desired output and the system is adjusted until
the difference between the two is minimized.
Bayes rule—Follows the relatively simple notion that adding new information to ex-
isting information impacts probabilities and their distributions. It can be ex-
pressed mathematically as the probability of A given B equals the probability of
B given A times the probability of A, divided by the probability of B.
Cellular automata—Collection of cells on a grid that evolve in some way through
discrete steps according to rules based on the states of neighboring cells.

331
10_DAS_Glossary_pp331-336 1/29/09 1:38 PM Page 332

332 HIDDEN STRUCTURE

Chromaticism—In conventional musical use, the pitches outside of a particular

scale in use.
Class—A means of describing the rules by which object instances behave.
CLOS—Acronym for Common Lisp Object System.
Combinatorics—The study of the possible orderings of finite collections of numbers.
Complexity theory—Related to the study of input size as relating to computer time
and memory requirements.
Cryptography—The mathematical study of information transmission.
DARMS—Acronym for digital alternative representation of music scores, developed
by Stefan Bauer-Mengelberg in 1963.
Diatonicism (diatonic)—Pitches within a particular scale in use.
DMAIT—See dynamic musical AIT.
Duodecimal—The base-12 numbering system.
Dynamic musical AIT—Systematically refigures musical AIT for several musical pa-
rameters simultaneously.
Emergence—The process by which complex systems and patterns arise out of a
multiplicity of relatively simple interactions.
Entropy—Related to information; measures the amount of improbability or unpre-
dictability in data.
Fibonacci sequence—A numerical pattern in which each new number is the sum of
the two preceding numbers.
Functional—In computer programming, command-based outcomes; e.g., typing
function names and arguments achieves the output. In music, a hierarchic struc-
ture of pitch or chord progressions, as found in the tonic-dominant relationships
of the Common Practice era. Musical function helps define syntax (expectation,
fulfillment, deception, etc.) and semantics (tension, release, voice-leading con-
straints, etc.) as a means to describe style.
Fuzzy logic—A type of logic in which propositions can be represented with degrees
of truthfulness and falsehood.
Game theory—A formal modeling approach to maximize results in game playing.
General systems theory—A system theory described by the nonlinearity of the sys-
tem’s interactions of its various components typically producing output that is
more than the sum of its parts.
Genetic algorithm—A virtual population that produces a number of trial solutions
to a problem, each of which is evaluated (fitness test) and a new generation cre-
10_DAS_Glossary_pp331-336 1/29/09 1:38 PM Page 333

GLOSSARY 333

ated from the better of them. The process continues through many generations
until an acceptable solution is achieved.
Generative theory of music—Follows the relationships between the syntactic struc-
ture of music and its semantic representation.
Global variable—A symbolic representation of a quantity or expression that does
not depend on current circumstances for its value.
Graph theory—The study of mathematical structures used to model relations be-
tween points in graphs.
Guidonian hand—A visualization for locating the semitones in the central part of
the medieval gamut (see Fig. 1.5). It serves as a kind of algorithm in itself, a sim-
ple organization of rules for memorization.
Hexachord—A six-note collection of pitches.
Hidden unit—Intermediate layers in a neural network that receive entire input pat-
terns modified by the passage through weighted connections; provides the inter-
nal representation of neural pathways.
Information—In information theory, the non-redundant portion of any string of
numbers.
Information theory—A branch of communications theory that deals with the
amount and accuracy of information when transmitted from a source through a
medium to a destination.
Interval—The number of half steps or scale degrees between two pitches.
IRCAM—Institut de Recherche et Coordination Acoustique/Musique; a long-
standing part of the musical community in Paris associated with modern music.
ISMIR—International Society for Music Information Retrieval; an organization that
currently offers enormous potential for the establishment of a common protocol
for music storage and retrieval for future computer music analysis.
Local variable—A value only within a current closed environment.
Logic—The study of formal systems relating to inference.
Magic square—A matrix of n rows and columns. The first n2 integers are arranged
in the cells of the matrix in such a way that the sum of any row or column (op-
tionally diagonal) is the same.
MAIT—See musical AIT.
Markov chain—Probabilistic processes often expressed in terms of orders. A zero-
order Markov chain, for example, makes random decisions, with no applicable
rules. A first-order Markov chain, however, bases new decisions on immediately
preceding choices, while a fifth-order Markov chain bases its decisions on the
previous five choices.
10_DAS_Glossary_pp331-336 1/29/09 1:38 PM Page 334

334 HIDDEN STRUCTURE

Mathematical sequence—An ordered list of objects, typically numbers, of finite or

infinite length.
Method—In CLOS, a class-defined function.
MIDI—Musical Instrument Digital Interface; code to create a common interface be-
tween various electronic music instruments, as well as links between electronic
performance and digital storage.
MIR—Musical Information Retrieval.
Modeling—The building of accurate replicas of objects or phenomena in order to
better understand their structure.
Music analysis—A term used to identify specific instances when the principles en-
countered in music theory apply to the evaluation of a specific work or body of
works.
Music theory—The basic tenets that govern all music. See also music analysis.
Musical—A term meaning that within the context of a particular piece of music, log-
ical, intuitive, and physical interpretations agree. Being logical infers the follow-
ing of explicit rules. Being intuitive infers the following of implicit rules. Being
physical infers the following of natural physical laws (referring here to human
performability). A musical passage is therefore one in which the user of the term
finds all of the above criteria acceptable and in coincidence.
Musical AIT—One of two general types of information. The first contains data that
occur repeatedly in various forms throughout the remainder of the work being
analyzed. The second contains data unique to their first appearance in a work.
Musical set theory—See set theory.
Neural network—Like the human brain, an entity made up of interconnected neu-
ron units that respond in parallel to a set of input signals. Consists of four main
parts: (1) processing units, (2) weighted interconnections between processing
units, (3) an activation rule, and (4) a learning rule that specifies how to adjust
the weights for a given input/output pair.
Number theory—The study of dicrete numbers in general and integers in particular.
OOP—Object-Oriented Programming.
Ordered PC set—A pitch-class set whose order follows the order of the music it
represents.
Pascal’s triangle—A triangle of numbers generated by affixing a 1 to either end of a
new row and then generating all numbers between those 1s by adding together
the two numbers to the left and right above each number of the new row.
Polymorphism—The concept that methods of the same name may have different
effects depending on the class with which they are associated.
10_DAS_Glossary_pp331-336 1/29/09 1:38 PM Page 335

GLOSSARY 335

Post-tonal—Music without a conventional tonal structure, including polytonal, octa-

tonal, and serial styles.
Predicate calculus—A system of symbolic logic.
Probability theory—The study of probability.
Recombinancy—In music, a method for producing new music based on the recom-
bination of elements of extant music.
Recursion—The act of a function calling itself during operation.
Register—The octaves in which pitches are used in musical compositions.
Root theory—Derived from the work of the composer Paul Hindemith and involving
(1) identifying all of the separate intervals present in the grouping, (2) locating
each interval’s lowest occurrence in the overtone series, (3) determining each
interval’s root, and (4) finding the strongest root among the roots present.
Rotational chromaticism—Includes pitches foreign to the current scale that repeat
or otherwise become consistent over time, alter their native scale versions,
and/or appear in strong metric, rhythmic, and/or agogic circumstances, and that
create a different scale but one of the same vector class.
SAM—System for Analysis of Music; designed by Eric Regener in the 1960s; in-
cludes an elaborate assembly-language program for the then state-of-the-art IBM
7090 computer.
Semiotics—The study of signs and symbols, what they mean, and how they are
used, particularly in relationships established through cultural convention.
Set theory—An approach to analysis that involves the use of pitch classes in de-
scribing musical structures, especially those that involve post-tonal works. When
applied to music, offers analysts opportunities to analyze post-tonal music by
grouping it into sets and then reducing these sets into various categories.
SPEAC—S (statement ) for stable, P (preparation) for weakly unstable, E (extension)
for fairly stable, A (antecedent) for very unstable, and C (consequent) for strongly
stable; pronounced “speak.” SPEAC identifiers thus follow an A–P–E–S–C stability
order, with the most unstable identifier on the left and the most stable identifier
on the right.
State transition matrix (STM)—A matrix of probabilities for moving from one state
to the next state in a Markov chain.
Structural analysis—An analytical method that, unlike formal analysis. which relies
on material repetition, variation, and/or contrast, attempts to reveal the archi-
tecture (harmonic and melodic superstructure) of music.
Tessitura—The idiomatic range of a set of pitches, as opposed to the potentially
wider range used.
10_DAS_Glossary_pp331-336 1/29/09 1:38 PM Page 336

336 HIDDEN STRUCTURE

Tetrachord—A four-note collection of pitches.

Transformational chromaticism—Includes pitches foreign to the current scale
that create—through repetition and in strong metric, rhythmic, and/or agogic
circumstances—a demonstrably new scale of a different vector class.
Transient chromaticism—Pitches foreign to the current scale that appear briefly,
typically resolve to their associated pitch, and/or appear in weak metric, rhyth-
mic, and/or agogic circumstances.
Transpositional chromaticism—Pitches foreign to the current scale that repeat or
otherwise become consistent over time, alter their native scale versions, and/or
appear in strong metric, rhythmic, and/or agogic circumstances, and that create
a transposed version of the scale currently in use.
Trichord—A three-note collection of pitches.
Unordered PC set—A pitch-class set whose order does not matter.
11_DAS_Index_pp337-344 1/29/09 1:39 PM Page 337

Index

A Bach, C. P. E., 17, 250

agents, 294 Bach, J. S., 4, 21, 23–24, 31, 32, 35, 36,
agogic, 156–57, 209, 222 65–66, 68, 95, 114, 210–11, 214, 231,
AI. See artificial intelligence 235–37, 244, 245, 247–49, 308
AIT. See algorithmic information theory Chorale no. 2, 38
algebra, 99 Chorale no. 42, 211
algorithm, 1, 3–7, 10, 12, 14, 16–18, 21, Suite no. 1 in G Major for Violoncello
22, 25, 32, 37, 40, 43, 57–58, 145, 148, Solo, 66
159, 213, 232, 234, 236, 260, 268, 305 Wenn wir in höchsten Noten sein, 23
algorithmic information theory (AIT), back propagation, 291
43, 45–97, 99, 231 See also neural networks
See also information theory background, 32, 222, 223
Algorithmic Interactive Composing Envi- See also Hintergrund
ronment (Alice), 37, 223, 263–64, 266, Balaban, Mira, 35
268–69, 273, 292 Bartók, Béla, 2, 35, 252, 258–62, 315
Alice. See Algorithmic Interactive Com- Mikrokosmos, no. 71, 255, 256
posing Environment Mikrokosmos, no. 77, 255, 257
allusion, 244–46, 296 Mikrokosmos, no. 80, 261
Alphonce, Bo, 31 Mikrokosmos, no. 81, 69, 74, 75, 269–
antecedent (A), 213–14 70
See also SPEAC Program Music for Strings, Percussion and
L’antica musica ridotta alla moderna Celeste, 285, 287
prattica, 14 Bauer-Mengelberg, Stefan, 26
Aristotle, 7–8 Bayes rule, 250–51
Aristoxenus, 7–9 See also probabilities
artificial intelligence (AI), 3, 25, 46, 97, Beethoven, Ludwig van, 16, 17, 29, 35,
214, 274, 275, 290, 292, 294 95, 244, 245
association network, 214, 291–95 Bagatelle, op. 119, no. 1, 246
atonality, 2 Piano Sonata in C Major, op. 53,
augmented sixth chord, 245 149–51, 157
Piano Sonata, op. 7, 157, 158
B Piano Sonata, op. 2, no. 1, 244
Babbage, Charles, 22–25 Symphony no. 5, 66, 68, 307, 309
Babbitt, Milton, 22, 29, 99, 102, 106, 277 Berg, Alban, 29

337
11_DAS_Index_pp337-344 1/29/09 1:39 PM Page 338

338 HIDDEN STRUCTURE

Bharucha, Jamshed, 35 CMAP. See Contemporary Music Analy-

binary numbers, 53, 60, 83, 88, 151, 152, sis Package
277 Colman, James, 32
Blombach, Ann, 31–32 combinatorics, 17, 19, 151, 250, 280,
Boethius, Anicius, 8, 10–11 282–83, 297, 299, 300
Böker-Heil, Norbert, 56 Common Lisp, 41, 43, 46–49, 60, 91, 93,183,
Boolean processes, 91 189, 190, 192, 196, 253, 273, 274, 315
bottom-up programming, 196–97 See also Lisp
Boulez, Pierre, 123, 125, 126, 131–32, Common Lisp Object System (CLOS), 190
134, 189 Common Music (CM), 37
Structures, 123, 125, 126 Comparison Program, 62
Second Sonata for Piano, 131–32, 134 complementary analysis, 315
Brahms, Johannes, 16, 66, 68, 72, 73, 95 complexity theory, 280
Symphony no. 1, 67 compression, 51–52, 57–58, 60, 62, 65–
Brinkman, Alexander, 32 66, 68, 71, 91, 93–95
bronze mean, 285, 287 See also Algorithmic Information
Brooks, Frederick, 27 Theory
Computer-Assisted Set Analysis Pro-
C gram (CASAP), 32
Cage, John, 4, 304–5 conditional variable, 250
calculus, 99 conditionals, 48, 204
Campion, Thomas, 14 consequent (C), 213–14
CASAP. See Computer-Assisted Set See also SPEAC Program
Analysis Program Contemporary Music Analysis Package
CCRMA. See Center for Research in (CMAP), 32
Music and Acoustics Cook, Nicholas, 2, 31, 318
cellular automata, 58, 60, 61, 232, 289, Cooper, Grosvenor, 36
292, 294 Cope, David, 66, 95
Center for Research in Music and Concerto for Cello and Orchestra,
Acoustics (CCRMA), 37 285, 286, 308
Chaitin, Gregory, 57, 60 Horizons for Orchestra, 27
See also Algorithmic Information Three Pieces for Solo Clarinet, 67
Theory Triplum for flute and piano, 133,
chaos theory, 289 135–38, 140, 228, 229
chaos, 45, 54 Copernicus, Nicolaus, 6
Chopin, Frederic, 31, 75–77, 83, 84, 86, Corelli, Archangelo, 35
89, 221 The Craft of Musical Composition, 21, 30
CHORAL Program, 35 Cryptography, 2–3, 52, 57, 280
chromatic scale, 72, 74, 114, 135, 146,
149, 151–53, 162, 164, 177, 179, 205, D
252, 253, 263, 269, 271 Dallapiccola, Luigi, 31
class (object), 190–97 DARMS. See Digital Alternative Repre-
CLOS. See Common Lisp Object System sentation of Music Scores
CM. See Common Music de Garlandia, Johannes, 12
11_DAS_Index_pp337-344 1/29/09 1:39 PM Page 339

INDEX 339

De harmonica institutione, 8, 10 F
De institutione musica, 8, 10 Fibonacci sequence, 146, 276, 285, 287
de Liège, Jacques, 12 foreground, 21, 32, 218, 221–22, 314
De mensurabili, 12 See also Vordergrund
de Vitry, Philippe12, 14 Forte, Allen, 22, 29, 32, 43, 99, 102, 106,
Debussy, Claude, 35, 113, 285, 315 112, 215–16, 277
La cathédrale engloutie, 113 forward feed, 290
Deep Blue, 317 See also neural networks
Desain, Peter, 35, 268 fractals, 277, 289
diatonicism, 14, 21, 114, 157, 187 function,
Dies irae, 308, 311 in music, 1, 3, 4, 7, 16, 21, 23, 24, 36,
Digital Alternative Representation of 37, 41, 105, 108–10, 144, 149, 155,
Music Scores (DARMS), 26, 315 169, 188–229, 245, 260, 290, 314,
discrete mathematics, 43, 99, 280–81 316, 318
distribution, 250 in probability, 250
DMAIT. See dynamic music algorithmic in programming, 46–49, 54, 62, 65, 90–
information theory 95, 109, 141, 147–48, 182–84, 236,
Dodecachordon, 14 255–59, 269, 272–74, 290, 294, 301–3
dodecaphonic, 114 fundamental bass, 16
duodecimal numbers, 99, 113–20, 127, Fux, Joseph Johann, 14, 16–17
133, 137, 139, 141–42, 144, 146 fuzzy logic, 289
dyads, 114, 123–24, 127
dynamic music algorithmic information G
theory (DMAIT), 45, 71, 72, 74, 75, 83, Gadsby, 54–55
84, 91, 96 Galilei, Galileo, 6
game theory, 280
E general systems theory, 287–89
Ebcioglu, Kemal, 35 generative theory of music, 31, 36
Elody, 37 genetic algorithms, 289, 292, 294
emergence, 294–95 gestalt analysis, 316
entropy, 28, 32, 45, 51, 57 Gjerdingen, Robert, 245
esthesic dimension, 296 Glarean, Henricus, 14
Experiments in Musical Intelligence, 21, global variable, 255, 272
75, 76, 83, 196, 221, 234–39, 244, 246, golden mean, 146, 285, 287
263, 292, 305, 307, 317 Gould, Murray, 26
Eine Kleine Stück, 305–7 Gradus Program, 16, 262–63, 273, 292
Rondo Capriccio for cello and orches- Gradus ad Parnassum, 14, 17
tra, 238–43 Grame, 37
Symphony no. 5 (revision), 307–10 graph theory, 280
See also recombinancy Greek modes, 8
Extend Program, 234, 259, 260, 262, 269, Gregorian Chant, 10–11
271, 272 Gross, Dorothy, 31
extension (E), 213–14, 223, 225 Guido of Arezzo, 11–12
See also SPEAC Program Guidonian hand, 12–13
11_DAS_Index_pp337-344 1/29/09 1:39 PM Page 340

340 HIDDEN STRUCTURE

H interval vector, 32, 109, 113, 119–21, 123,

Harris, Craig, 32 126, 139, 141
Hauptmann, Moritz, 19, 21 inversion, 4, 16–19, 23, 36, 37, 41, 63–64,
Haydn, Franz Joseph, 17, 19, 20, 31, 231, 90–93, 103–9, 114, 202, 206–7, 281–82,
250 285
hermeneutic analysis, 316 inversional symmetry, 155
hexachord, 11–12, 110, 184, 204, 300, 301 IRCAM. See Institut de Recherche et
hidden units, 290–92 Coordination Acoustique/ Musique
hierarchy, Isaacson, Leonard, 28–29, 45, 231, 249,
in music, 188, 189, 197–98, 214, 217 281
in programming, 190, 193–94, 197–98 ISMIR. See International Society for
Hiller, Lejaren, 28–29, 45, 231, 249, 281, Music Information Retrieval
287, 288 Le istitutioni harmoniche (Zarlino), 14
Hindemith, Paul, 21–22, 29–30, 148, 199,
201–2 J
Hintergrund, 21 jigsaw test, 247–48, 308
See also background Johnson, Tom, 289
Hofstadter, Douglas, 295
Hofstetter, Fred, 31 K
Holland, John, 294–95 KEITH Program, 31
Honing, Henkjan, 35 Kepler, Johannes, 6
Howell, Emily, 294 Kirnberger, Johann Philip, 17–18, 250
Hucbald, 8, 10–11 Kostka, Stephan, 30
Humdrum Toolkit, 35, 37, 315 Krenek, Ernst, 66–67, 95
Huron, David, 35, 315 Suite for Violoncello, 66–67

I L
Illiac Suite, 28–29 LaRue, Jan, 30
IML. See Intermediary Musical Language Laske, Otto, 31
information theory, 45–97, 99, 187, 231 layer analysis, 221
inheritance, 196 Lefkoff, Gerald, 30
instance, 191–96, 202, 204, 212 Lerdahl, Fred, 22, 31, 36, 217
Institut de Recherche et Coordination lexicons, 51, 60, 110, 235–36, 266
Acoustique/ Musique (IRCAM), 37 Liber de arte contrapuncti, 14
Intermediary Musical Language (IML), Lieberman, Fredric, 30
26 Lincoln, Harry, 30
International Society for Music Informa- linear interval vectors, 113, 119–23, 126–
tion Retrieval (ISMIR), 26, 315 27, 131, 139, 141, 148, 152–54, 158–60,
intersection, 100–1 64–65, 167–72, 177, 179, 181–83
interval summations, 113, 119–21, 126, Lisp, 41, 43, 49–97, 108, 109, 112, 147–48,
127, 130–33, 135, 139, 141 182–83, 189–90, 192, 196, 204, 253–55,
interval maps, 113, 119, 120, 126, 127, 271, 273, 274, 301, 315
130–35, 139 See also Common Lisp
11_DAS_Index_pp337-344 1/29/09 1:39 PM Page 341

INDEX 341

local variable, 184, 255 Mozart, Wolfgang Amadeus, 17, 19, 29,
logic, 19, 35, 99, 101, 280, 281, 316 35, 66, 95, 231, 232, 238–39, 244, 250
Lovelace, Ada, 23 Piano Sonata K. 279, 233–34
Piano Sonata K. 533, 233–34
M Piano Sonata K. 545, 221–22
magic cubes, 285 Symphony no. 40 in G Minor, K. 550,
magic squares, 281–85 66, 95
MAIT. See musical algorithmic informa- MTW. See Music Theory Workbench
tion theory Multigraph Program, 62, 74, 76, 90, 96,
Marenzio, Luca, 56 97
Markov chains, 249, 251–54, 258 multiple-octave scale, 186
Markov Program, 254, 255, 258, 259 MUSANA, 32–33, 35
Mathews, Max, 26 Muse Program, 295–305, 307, 308
matrix, 63–64, 252–55, 258, 281, 285 music cognition, 5, 31, 57, 71, 169, 268,
Mazzola, Guerino, 36 318
McCarthy, John, 46 MUSIC i–v, 26
McHose, Allen, 23–24 music perception, 5, 169, 268
Melisma Music Analyzer, 36 Music Theory Workbench (MTW), 36, 38
Mendelssohn, Felix, 28 Musica enchiriadis, 10
metaclass, 153–55, 164 musical algorithmic information theory
meta-patterns, 220–23, 238 (MAIT), 45, 65, 67–68, 71, 88
methods, 192–96, 204, 219 musical geometries, 281, 316
Meyer, Leonard, 22, 36, 189, 198–99, 268 Musical Information Retrieval (MIR), 26,
Micrologus, 11–12 35, 315
middleground, 21, 32, 221–22 Musikalisches Würfelspiele, 17–19, 250
See also Mittlegrund MUSTRAN, 26
MIDI, xxiv, xxvii, 26, 36, 37, 62, 76, 90,
116, 254, 315 N
minimalism, 67–68 Nattiez, Jean-Jacques, 296–97
MIR. See Musical Information Retrieval neural networks, 35, 290–92
Mittlegrund, 21 neutral dimension, 296–97, 300, 304
See also middleground A New Way of Making Fowre Parts in
modulation, 72, 149, 156, 164, 213, 217, Counter-point, 14
289 nonharmonic tones, 36, 37, 149, 162,
modulatory chromaticism, 157 213, 245
modulo, 102, 108, 111, 148, 277, 278, normal form, 37, 103–12, 117–19, 137,
287 141
Monte Carlo method, 28 number theory, 99, 280–81
Morley, Thomas, 14
Morris, Robert, 106, 142, 149, 277 O
Morse code, 53–54 object-oriented programming (OOP),
Morse, Samuel, 53 189–90, 196, 204, 221, 228
Morton, Ian, 30 octachord, 110
11_DAS_Index_pp337-344 1/29/09 1:39 PM Page 342

342 HIDDEN STRUCTURE

octatonal, 2, 147, 149, 151, 153, 155, 159, The Politics, 7

162, 182, 205, 314 polymorphism, 193
octatonic scale, 147, 149, 151, 153, 155, polynomial formula, 147–48
159, 162, 182, 205, 314 polytonality, 2, 40, 41, 131, 160, 305, 308
OM. See OpenMusic Pople, Anthony, 2, 36, 304
omega (⍀), 60, 153 Praetorius, Michael, 14
music and, 153–54 predicate, 48
mathematics and, 60 predicate calculus, 35, 99, 316
OOP. See object-oriented programming preparation (P), 213, 214, 223, 225
OpenMusic (OM), 37, 39 See also SPEAC Program
ordered set, 99, 102–8, 117–20, 127, 130, prime form, 31, 32, 63, 65, 91, 92, 105–12,
133, 135, 139, 195, 300, 314 117–21, 135, 141, 142, 203, 285, 287,
orders, 52–53, 252 301
Markov, 252 prime numbers, 99, 147, 276
Shannon, 52–53 probabilities, 28, 51–54, 231, 249–58, 272
organum, 10–12 probability theory, 249, 280, 281
overtone series, 16, 19, 21, 201–3 Proportionale musices, 14
pseudo-code, 182–83, 192, 204, 235, 269
P Ptolemy, Claudius, 8
Palestrina, Giovanni Pierluigi da, 56 Pythagoras, 7, 43
palindrome, 281
Pascal, Blaise, 275–78, 294, 300 Q
Pascal’s triangle, 275–78, 294, 300 Quintilianus, Aristides, 8
pattern matching, 1, 7–8, 28–30, 32, 35,
37, 45–94, 99, 110, 122, 126, 155, 156, R
164, 215, 218–23, 233, 236–38, 251, Rahn, John, 102, 106, 142, 149, 275, 277
258, 274, 281, 183, 289–302, 308, 311, Rameau, Jean-Philippe, 16–19
315–18 recombinancy, 83, 234–38, 244, 245, 249,
Peirce, Charles Sanders, 295 268, 280, 305
pentachord, 110, 135, 204, 300, 301 recursion, 48–49, 109, 273
pentatonic scale, 153, 263 Regener, Eric, 29, 30
pi (π), 50–51, 58 register, 25, 63, 72, 99–144, 184–86, 210,
Pierce, John, 28, 52, 90 213, 214, 222, 223, 235, 302, 314
pitch class, 63, 102–8, 111–20, 135, 137, The Republic (Plato), 7
139, 141, 145, 151–53, 156–64, 167, Réti, Rudolph, 22
169, 172, 177, 179, 182–83, 185, 189, retrograde, 63–65, 90–93, 282, 285, 299
198, 202, 204–5, 223, 225, 259, 263, retrograde-inversion, 63–65, 91–93, 282,
282, 287, 288, 300, 307–8, 311 285
A Plaine and Easie Introduction to Prati- Riemann, Hugo, 21, 216
call Musick, 14 Riepel, Joseph, 18–19
The Plaine and Easy Code, 26 root, 21, 30, 36, 37,102–4, 194–96, 198–
Plato, 7–8 207, 209–14, 216–17, 225, 227, 301–3
poietic dimension, 296 Root Program, 225, 227
11_DAS_Index_pp337-344 1/29/09 1:39 PM Page 343

INDEX 343

rotational chromaticism, 155–57, 169, Shannon, Claude, 28, 52–55

170 silver mean, 285, 287–78
RUBATO, 36 Simonton, Dean, 32
Ruggiero, Charles, 32 Sleator, Daniel, 36
Ruwet, Nicholas, 22 Smoliar, Stephen, 32
SPEAC Program
S statement (S), 213–14, 216, 223, 225
Saint Augustine, v, 45 See also SPEAC Program
scale classes, 151–55 statistics, 5, 23, 28, 30, 32, 35, 36, 52,
Schaffer, John, 35 249, 301
Schenker, Heinrich, 21–23, 32, 68, 91, Straus, Joseph, 2, 107, 112, 122, 127, 145,
213, 221–22, 318 149, 162, 223
Schillinger System of Musical Composi- Stravinsky, Igor, 2, 39, 305, 308
tion, 22 Le sacre du printemps, 308, 312, 315
Schillinger, Joseph, 22 Symphonies of Wind Instruments, 313
Schoenberg, Arnold, 2, 63, 126, 142, 148, Three Pieces for String Quartet, 39–
223, 224, 305 41, 74–76, 142, 181–82, 225–28, 269,
Das Buch der hängenden Gärten, 186 308, 312–13
Sechs kleine Klavierstücke, op. 19, 111, structural analysis, 32, 68, 70, 91, 142,
169, 170, 171, 172, 179, 210, 212, 214 169, 189–90, 200, 217, 222, 223, 225,
Suite for Piano, op. 25, 121–22 227–29, 237, 238, 251, 263, 297
Three Piano Pieces, op. 11, 217, 219, Structure Graph Program, 228
305, 307 Structure Map Program, 225, 227, 228
Schubert, Franz, 28 subclass, 194, 196
Schüler, Nico, xxiv, 26, 28–30, 32 subset, 31, 100–1, 244
Schumann, Robert, 28 superclass, 194, 196
secondary chromaticism, 72 Symbolic Composer, 37
secondary harmonies, 72, 149, 151, 155, Symmetry, 155, 156, 159, 281, 283, 285,
157 316
Selfridge-Field, Eleanor, 26 Syntagma musicum, 14
semiotics, 295–96
semiotics (musical), 245, 295–97 T
septachord, 110 Taube, Heinrich, 36, 38
serial music, 2, 4, 31, 63, 114, 122, 145, TDS, 21, 216
189, 262, 282, 283, 305 Temperley, David, 36, 251
Set Database Program, 110 tessitura, 25, 238, 283
Set Multiples Program, 141 tetrachord, 8, 110, 121–23, 126, 131–33,
set theory 142, 276, 300, 301
mathematics and, 99–102, 104, 105, Tinctoris, Johannes, 14–15
277, 280, 281 Tn form, 106
music and, 4, 29, 32, 43, 97, 102, 106, TnI form, 106
109–12, 148, 189, 197 t-normal form, 108, 110–12, 117, 119,
SetMath Program, 114, 141 141, 205
11_DAS_Index_pp337-344 1/29/09 1:39 PM Page 344

344 HIDDEN STRUCTURE

top-down programming, 196–97 verbing, 212

transformational analysis, 316 Vicentino, Nicola, 14
transformational chromaticism, 155–57, Visualize Program, 41
167, 169 Vivaldi, Antonio, 231
transient chromaticism, 155–56, 158, Vordergrund, 21
164–65, 169, 172, 179 See also foreground
transpositional chromaticism, 151, 155,
157, 165, 167 W
trichord, 110–11, 121, 123–24, 127, Webern, Anton von, 2, 65–66, 68, 95, 126
130–31, 133, 135, 139, 142, 203, 282, Concerto for 9 Instruments, op. 24,
297, 299–301 127–32, 282–83
Turing, Alan, 3, 25, 275 Symphony, op. 21, 29
twelve-tone music, 2, 30, 114, 122, 145, Variations for Piano, op. 27, 67,
151, 205, 283, 285, 305 122–24
See also serial music whole-tone scale, 149, 153, 155, 169, 179,
263
U Wolfram, Stephen, 2, 59, 232, 235
Uhrlandt, Dirk, 32 Wright, Ernest, 54–55
unification, 218–20
union, 100 X
unordered set, 99, 102–4, 106, 111, 117, Xenakis, Iannis, 1, 26, 251, 277
119, 167, 172, 300
Urlinie, 21 Y
Youngblood, Joseph, 28, 30
V
Varèse, Edgard, 2 Z
Density 21.5, 160–72, 179, 287, 288 Zahorka, Oliver, 36
Venn diagram, 101, 104 Zarlino, Gioseffo, 14–16
Venn, John, 101 Z-relations, 43
DAS23CoverCope 2/28/14 1:22 PM Page 1

VOLUME 23 THE COMPUTER MUSIC AND DIGITAL AUDIO SERIES

VOLUME 23 THE COMPUTER MUSIC AND DIGITAL AUDIO SERIES

Hidden Structure
Hidden Structure

Hidden Structure
Music Analysis
Using Computers Music Analysis
David Cope
Today’s computers provide music theorists with unprecedented opportunities to analyze
Using Computers
music more quickly and accurately than ever before. Where analysis once required several
weeks or even months to complete—often replete with human errors, computers now pro-
vide the means to accomplish these same analyses in a fraction of the time and with far

Music Analysis Using Computers

more accuracy. However, while such computer music analyses represent significant improve-
ments in the field, computational analyses using traditional approaches by themselves do
not constitute the true innovations in music theory that computers offer. In Hidden Structure:
Music Analysis Using Computers David Cope introduces a series of analytical processes
that—by virtue of their concept and design—can be better, and in some cases, only accom-
plished by computer programs, thereby presenting unique opportunities for music theorists to
understand more thoroughly the various kinds of music they study.
Following the introductory chapter that covers several important premises, Hidden Struc-
ture focuses on several unique approaches to music analysis offered by computer programs.
While these unique approaches do not represent an all-encompassing and integrated
global theory of music analysis, they do represent significantly more than a compilation of
loosely related computer program descriptions. For example, Chapter 5 on function in post-
tonal music, firmly depends on the scalar foundations presented in chapter 4. Likewise, chap-
ter 7 presents a multi-tiered approach to musical analysis that builds on the material found
in all of the preceding chapters. In short, Hidden Structure uniquely offers an integrated view
of computer music analysis for today’s musicians.

Of Related Interest
The Algorithmic Composer
By David Cope
DAS 16 ISBN 978-0-89579-454-3 (2000) xiii + 130 pp. (with CD-ROM) $49.95

Hyperimprovisation: Computer-Interactive Sound Improvisation

By Roger Dean
DAS 19 ISBN 978-0-89579-508-3 (2003) xxvi + 206 pp. (with CD-ROM) $49.95

New Digital Musical Instruments: Control and Interaction Beyond the Keyboard
By Eduardo Reck Miranda and Marcelo M. Wanderley
DAS 21 ISBN 978-0-89579-585-4 (2006) xxii + 286 pp. (with CD-ROM) $49.95

David Cope
I S B N 978-0-89579-640-0
A-R Editions, Inc. 90000
8551 Research Way, Suite 180
Middleton, WI 53562
800-736-0070
608-836-9000
http://www.areditions.com 9 780895 796400

Í David Cope