Professional Documents
Culture Documents
Automated Composition in Retrospect
Automated Composition in Retrospect
Automated Composition in Retrospect
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=mitpress.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the
scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that
promotes the discovery and use of these resources. For more information about JSTOR, please contact support@jstor.org.
The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to Leonardo.
http://www.jstor.org
,
Charles Ames
Abstract-This article chronicles computer programs for automated musical composition over the last 30
years, providingmusical examples from 11 computer-composedworks. The author begins by focusing on
early American efforts to impose stylistic controls through random sampling and testing, then describes
Europeanstatistical approaches used until 1970. The rise of 'interactive'compositional utilities duringthe
1970s is linked to the growing proliferation of on-line computers.The currentdecade has been a period of
eclectic interests: the resurgence of statistical procedures, the introduction of 'top-down' recursive
grammars,the adaptationof problem-solvingtechniquesfrom Artificial Intelligence andthe continuationof
interactive efforts.
1. INTRODUCTION
This article chronicles efforts to harness oI
the computer as a tool for compositional
decision making, beginning in the mid-
POS4 8oi)od4 II I III
IIIII I
Be'RFfR)' IIIIII
t. j 1gF t e4< r
often virulent-resistance. Too often, 6,,. ct ^k 4y f Dm ^-4-
? S- 6r
~~~~~
~~ C' tIi M-, 1 ia G40r7~
however, the limitations of one or two 1t ijz }a
Ho-a I jaI 4*2" ; I I , a? II
t I-
individual approaches have been extra- CAL-CO - LAr- AJ, OAL- P-TA-riTJ' c4fcK - vJ/ird A
polated into conclusions that computers -^ ni .-.
are inherently unsuited to compositional
K
r-?9 r'K2m . :b
^ I 7-
i .:tn i sJI ) I
;X ^ ---I
' I^
I
I -
(A 4 I 17f --I c- " .
concerning how these composers have
r - -I J
,i 3.-,.1
? 1987 ISAST
Pergamon Journals Ltd.
Printed in Great Britain. LEONARDO, Vol. 20, No. 2, pp. 169-185, 1987
0024-094X/87 $3.00+0.00
progress in correcting this deficiency Ilisac Suite ExperimentsSummarized
through his 1970 comprehensive survey ExperimentOne: Monody, two-part, and four-part writing
of early efforts [8], he was able to provide
no musical examples and only the barest A limited selection of first-speciescounterpoint rules used for controlling the musical
details about how individual pieces were output
(a) Monody: cantus firmi 3 to 12 notes in length
made. The present article seeks not only
(b) Two-part cantus firmus settings 3 to 12 notes in length
to bring the history of automated (c) Four-part cantus firmus settings 3 to 12 notes in length
composition up to date, but also to clarify
(1) what has motivated composers over ExperimentTwo: Four-part first-species counterpoint
the last 30 years to delegate their creative Counterpointrules were added successively to random white-note music as follows:
decisions to a machine, and (2) how these (a) Random white-note music
composers have gone about programming (b) Skip-stepwiserule; no more than one successive repeat
the machine to give them what they (c) Opening C chord; cantus firmtis begins and ends on C; cadence on C: B-F
wanted. tritone only in VII; chord; tritone resolves to C-E
(d) Octave-rangerule
(e) Consonant harmonies only except for 6 chords
(f) Dissonant melodic intervals (seconds, sevenths, tritones) forbidden
II. "PUSH BUTTON BERTHA": 1956 (g) No parallel unisons, octaves, fifths
One of the earliest instances of auto- (h) No parallel fourths, no 6 chords, no repeat of climax in highest voice
mated composition was a program for
composing 'Tin Pan Alley' melodies, ExperimentThree: Experimental music
which was created by Martin Klein and Rhythm, dynamics, playing instructions, and simple chromatic writing
Douglas Bolitho of Burroughs, Inc. using (a) Basic rhythm, dynamics, and playing-instructionscode
a computer called DATATRON. An (b) Random chromatic music
anonymously written Burroughs publi- (c) Random chromatic music combined with modified rhythm, dynamics, and
cation outlines the operation of the I playing-instructionscode
program as follows: (d) Chromatic music controlled by an octave-range rule, a tritone-resolutionrule,
and a skip-stepwiserule
TheoperatorinspiresDATATRONby (e) Controlled chromatic music combined with modified rhythm, dynamics, and
first keying in a 10-digit random playing-instructionscode
number.This causes the machineto (/) Interval rows, tone rows, and restricted tone rows
generate and store 1000 single digits,
each representingone of the eight Experiment Four: Markoff chain music
diatonic notes in the scale with two (a) Variation of zeroth-order harmonic probability function from complete tonal
allowable accidentals. The program
then motivatesDATATRONto pick restriction to "average"distribution
successive notes at random, testing (b) Variation of zeroth-order harmonic probability function from random to
eachformelodicacceptability as it goes "average" distribution
along[9]. (c) Zeroth-orderharmonic and proximity probability functions and functions com-
bined additively
One result of this process was the melody (d) First-order harmonic and proximity probability functions and functions com-
"Push Button Bertha" (see Fig. 1), which bined additively
was first aired on July 15, 1956 [10]. (e) Zeroth-order harmonic and proximity functions on strong and weak beats,
respectively, and vice-versa
(f) First-order harmonic and proximity functions on strong and weak beats, re-
spectively, and vice-versa
(g) ith-order harmonic function on strong beats, first-orderproximity function on
III. THE URBANA SCHOOL: weak beats; extended cadence; simple closed form
1957 TO 1966
Klein and Bolitho's method of random Fig. 2. Illlac SuiteExperiments
Summarized.
Reproduced fromHillerandIsaacson[5]. Copyright
sampling and testing provided what 1959McGraw-Hill of thepublisher.
BookCompany.Usedby permission
seemed at the time a viable emulation by
computer of traditional compositional Hiller, Isaacson and Baker experiments were collected into an Illiac
decision making. In fact, the same Suite for string quartet. Figure 2 details
During the same years in which Klein
method had been developedindependently and Bolitho were programming DATA- Hiller and Isaacson's procedures, which
by Lejaren Hiller and Leonard Isaacson TRON, Hiller and Isaacson were under- utilized two basic approaches:
at the University of Illinois, 1. Random selection constrained
taking a series of compositional "experi-
Urbana/Champaign. The publicity gen- ments"* using the ILLIAC computer, by lists of rules, an approach
erated by Hiller and Isaacson's Illiac which had been designed and built at resembling that of Klein and
Suite [11], supplemented by the establish- Urbana. Many of their results were Bolitho (Experiments 1-3).
ment under Hiller of one of the nation's 2. Markov chains, also random,
presented in a well-publicized concert
earliest electronic music studios, attracted in which the relative likelihood
which, coincidentally,also occurredduring
a number of individuals interested in of each option was conditioned
July 1956. Subsequently, these and later
merging the new music with the new by one or more immediately
technology. Along with Hiller and preceding choices (Experiment
Isaacson, these people included Robert *Enclosure of terms in double quotes indicates coined 4).
Baker [12], James Tenney [13], Herbert or otherwise idosyncratic terminology drawn from An excerpt from the Illiac Suite appears
specific sources.
Brun [14-17] and John Myhill [18-20]. in Fig. 3.
Musicis generallycalled'human'when
it considers temporary or inherent
tendenciesof the mind,of partor all of
a composer'spersonality.Suchmusicis
based on feelingand since it turnsits
back,in a sense,on pureknowledge,it
mightratherbe called'inhuman',for it
celebrateswhat we have in common
with all the animalsratherthan with
what is individualto man:his reason.
Algorithmicmusic is thus 'inhuman'
only in one sense of the word, it is
'human'in as muchas it is the product
of rationalbeings[36].
I
I
I
Fig. 6. Herbert Brun, Non-Sequitur VI, 1966, measures 1-8. Copyright 1966 Herbert Briin. Used by permissionof the composer.
Constant deviation from central pitch Increasing deviation from central pitch Decreasing deviation from central pitch
o
c
.2
0
A A
*
1-a)
a
4)
0
C]
Ti-
VV\J\j/ l
o ItA
JLVV
;T.im V v
Fig. 7. Melodic contoursin John Myhill's Scherzo a Tre Voce,1965. At its maximum(vertical) spread,each contourspans a one-and-one-halfoctave range.
(Density=5.3)
cil ~ ~ ~ ~ -- -
-~~~0
~~~~-0o 0
.tto
Fig. 8. Chartof orchestrationversussubjectivedensityfor lannis Xenakis's Fig. 9. lannis Xenakis, Morsima-Amorsima, 1962, measures 136-146.
Theindicatedtimbreclassesare:(1) piano,(2) arco
Morsima-Amorsima. Each 'JW' marksignifies the first complete measureof a section; densities
ponticello, (3) harmonics,(4) arco normale,(5) glissando, (6) tremolo arco calculated by the program are presented in their 'objective' forms. The
ponticello, (7) pizzicato and (8) frappe col legno. After a diagramprovided abbreviationsin the string parts are: asp-arco sul ponticello; an -arco
by the composer. position normale, and fc--frappe col legno. Copyright 1967 Boosey and
Hawkes Music PublishersLimited, 295 Regent Street, LondonW1R 8JH,
United Kingdom.Reprintedby permissionof the publisher.
I I
-
-7
-
r.
7,
Entry Delay ALEA from 0.24, 0.30, ALEA from 0.46, 0.58, 1 C# F
2 F# A
0.37, 0.46, 0.58, 0.72, 0.72, 0.89, 1.11, 1.38, 3 Bb C
and 0.89 seconds for and 1.72 seconds. 4 F F#
entire chord. 5 A Bb
6 C# F F#
7 C# F C
Duration ALEA from 0.10, 0.12, Same as entry delay A Bb
8 F#
0.15, 0.19, 0.24, 0.30, 9 F F# A
0.46, 0.58, and 0.72 10 C# Bb C
seconds for entire chord; 11 A Bb C
12 C# F F# A
duration may not exceed 13 C# F Bb C
entry delay. 14 F# A Bb C
15 F F# A Bb
Dynamics GROUP: elements by SERIES SERIES from pp, p, mp, 16 C# F F# C
17 C# A Bb C
from pp, p, mp, mf, f, and mf, f, and ff.
ff; repetitions by SERIES
from 3, 4, 5, 6, and 7. Fig. 11. Chordal table for Structure 8 of
Koenig's Ubungfur Klavier.
trumpet-like
~Iv
ff -tpl
*
-
-
v710-_
-
"Fl-
-I-
_ . -v-_ -- w
v_=v
Fig. 14. EmmanuelGhent, Lustrum,1974, measures208-212. Copyright1978 PersimmonPress, 131 Prince Street, New York, NY 10012, U.S.A. Used by
permissionof the publisher.
served the option to modify what the percent, or anywhere in between, by the 'new generation' of composers seeking
computer produced. turning a knob while listening to the creative assistance from computers are
A variety of comments by Spiegel music evolve [the position of the knob
at each moment would be recorded by
Larry Austin, Claudio Baffioni [87],
illustrateshow a composer might typically GROOVE for later playback]; or I Clarence Barlow [88-90], Tommaso
interact in real time with GROOVE. To could throw a switch to change over to Bolognesi [91], Lelio Camilleri, Joel
create her 1974 composition Patchwork, an entirely different set of rules [85]. Chadabe, Thomas DeLio, Charles Dodge
for example, Spiegel developed programs [92], David Feldman, Christopher Fry
that monitored GROOVE's input devices [93], Peter Gena [94], Francesco Guerra
Truax and Buxton
in order to "derive from [them] much [95], Christos Hatzis [96], Steven
more complex music than [Spiegel] Two Canadians who had studied under Holtzman [97,98], Kevin Jones [99], Gary
actually played". Spiegel has commented Koenig and Laske, Barry Truax and Kendall [100], Shirish Korde, Petr Kotik,
as follows: William Buxton, have since organized Ron Kuivila, LauraTedeschini Lalli [101],
computer music studios respectively in David Levitt [102], Denis Lorrain [103],
The programI wrote for [Patchwork] Vancouver and Toronto. In addition to Leonard Manzara [104], Larry Polansky,
had all Bach's favorite [motivic] developing real-time performance systems, CurtisRoads [105-108], David Rothenberg
manipulations-retrograde,inversion, both Truax and Buxton have developed [109], Andrew Schloss, William
augmentation,diminution,transposi- non-real-time environments which permit Schottstaedt [110], Sever Tipei [111],
tion-available on switches, knobs,
users to get by with little or no pro- Horacio Vaggione, Charles Wourinen,
pushbuttons,and keys, so that I could
manipulatethe 4 simplemelodicand 4 gramming. These environments enable David Zicarelliand myself [ 112-120].The
rhythmicpatterns with them in the users to monitor results closely as a extent of this new activity has been far too
samewaythata playerof aninstrument composition takes shape. They minimize great to continue with a detailed survey of
manipulatesindividualtones. (I did delays in non-real-time interaction by pieces, so I shall limit my remaining
edit it a lot, too.) [84]
providing a menu of precompiled "macros" examplesto significantworksthat have not
The melodic patterns mentioned by (Buxton's term) and by incorporating been describedpreviously in the literature.
Spiegel were representedby her programs facilities for rapidly evaluating musical Further details may be obtained from the
as "relative (intervallic) distances" from a products, specifically: real-time digital works cited in the References and Notes.
"pitch reference point" which Spiegel synthesis augmented visually by graphic
could control in real time from her displays. Laske [86] has documented how
he built up his 1980 composition Thomas DeLio
keyboard. Elsewhere Spiegel describes
some of the 'higher-level' methods of Terpsichore as a succession of increasingly Thomas DeLio's 1976 Serenade for
controlling the musical evolution of her elaborate "subscores", using macros solo piano is an elaborate work for which
1975MusicforDance through GROOVE's supplied by Buxton's SCED (SCore the form was composed manually while
input devices: EDiting) utility. the details were selected automatically.
The Serenade has three parts, each
BecauseI couldhearthe musicplayed consisting of 10sections;markedcontrasts
VI. NEW APPROACHES: 1976-1986 and occasional long silences between
while computationwas in progress,I
couldchoosewhichmusicaldecisionsI Recent efforts have shown a new many of these sections give the Serenade
wanted to make, leaving the rest to sophistication toward automated deci- an episodic quality congenial to its title.
follow the logic I had coded .... For
sion making along with increased machine The work is built up from a number of
example,I couldsettheprobabilitythat
a certain pitch would be played on participation in decisions affecting large- primary rhythmicpatterns and chromatic
strong beats at zero percent, at 100 scale compositional form. Included among cells. Part I introduces these ideas in
"embryonic states"; they emerge "fully Rhythm. Over the first 17 quarters, the one from one quarter to the next. This
formed" in Part II, which also serves a rhythm coalesces from a sparse, irregular process resulted in the following chain:
developmental purpose; Part III extra- texture to a consistent 7:4 pattern. DeLio
polates several of the ideas from earlier accomplished this evolution by treating 566777878899 10 10 10 10 10 10...
parts into "broad sweeping guestures". each quarter as a statistical frame whose
Figure 15 reproduces Part I, Section 10 elements are the 29 unequal rhythmic Once these numbers have been deter-
of the Serenade.The basic ideas employed units obtained by interfering septuplet mined, DeLio's program locates these
in this section are a 7:4 polyrhythm and thirty-secondsagainst duple sixty-fourths. attacks within each quarter by selecting
two chromatic cells: G# B C D and its A graphic depiction of the resulting without replacement from the units
reflection, G# A# B D. Registers are pattern appears in Fig. 16. In this section depicted in Fig. 16. Initially, each unit
distributed uniformly over the range of the Serenade,only one note may attack receives equal weight; as the rhythm
from middle C upward. The section at a time; the number of attacks per evolves, the off-beat thirty-seconds and
consists of two gradual evolutions, each quarter results from a "first-order sixty-fourths receive less and less weight
lasting 17 quarters. The following transition prQcess"(i.e. a Markov chain) until only the 7:4 pattern remains. This
breakdown of methods for handling in which the number of attacks may either pattern holds consistently through the
rhythm and pitch is based upon a letter stay the same, increase by one or (with remaining 17 quarters.
from DeLio to the author: relatively small probability) decrease by
Pitch. At the beginning of Section 10,
j = 60n DeLio's programselectsdegreesuniformly
"N
r ~ 7 . -r
,
r3
7
_-
-nI
7 Z -- z
I
Grammars
The effectiveness of composing pro-
gramshas beenimprovedradicallythrough
the introduction in recent years of
recursive programming techniques; the
first major area of activity relying heavily
on such techniqueshas beenthe application
of Noam Chomsky's phrase-structure
grammarsto composition. This approach
enables a composer to describe musical
forms economically, by first providing a
general archetype (or axiom) of the form
and by further listing a set of productions
for deriving details from generalities. The
full power of such an approach can be
attained only if the productions are
capable of acting upon their own results
(i.e. capable of recursion).
Although Curtis Roads had already
described an elaborate left-to-right
composing program called PROCES-
S/ING as early as 1975 [121, 122], his
advocacyof top-downChomskianmethods
[123] has achieved the most influence.
I Roads'semphasisupon productionsclassi-
fied by Chromsky as context-free was
quickly extended to embrace other
Chomskian ideas such as context-sensitive
productions and simple transformationsin
Steven Holtzman's 1980 "A Generative
Grammar Definitional Language for
Music" [124]. In Kevin Jones's 1981
"stochasticweb grammars"[125], context-
Fig. 15. ThomasDeLio, Serenade, 1976, Part I, Section 10. Copyright1976 ThomasDeLio. Reprinted free productions are employed to carve
by permissionof the composer. up a region of musical space (described in
VII. CONCLUSIONS
For most of the composers discussed in
this article, composition has not been a
vehicle of 'expression' or 'affect' in the
usual senses of these words. Rather, these
; i P composers expect listeners to judge their
musical "artifacts", as Barbaud called
them, on the basis of more abstract
aesthetic qualities such as balance, clarity,
depth (in the hierarchic sense) and so
,y "~
,% ,r'T. forth. They have used computers to
create these artifacts because they believe
that such qualities are quantifiable-at
least within the scope of a particular
work-and they have concluded from
experience that there are many com-
wp positional problems that computers are
better equipped than humans to solve.
Considered individually, many of the
earlier efforts described in this article
have been quite simple. This simplicity
has been due partially to limits imposed
by technology and partially to the spirit
of the 1960s and 1970s, when economy of
musical process was itself considered a
8 P virtue. Today, a one-pass, self-contained
:c- a , D
79'o" composing program such as Ebcioglu's
Bach simulator [140] might involve as
many as 7,000 lines of highly concise
code. An alternative approach employed
St by DeLio, Barlow [141] and myself [142]
has been to divide the overall com-
P X- positional process into multiple phases,
8
each undertaken by a separate program.
Fig. 17. James Tenney,Excerptfrom Bridge, 1984, for two pianos.The elapsedtime fromthe beginning
of the performanceis indicated in minutes and seconds at the end of each system of staves. Notated Taken collectively, the compositions
pitches are modified by the tuning system shown in Fig. 18. The rhythmic notation shows relative surveyed here present a rich diversity of
attack-pointsby their horizontalplacementon the page. Unbeamed,solid noteheadsindicatestaccato; techniques for programming computers
beamed notes last either up to the next stem or to the end of the beam; open noteheads employ pedal. to compose music. Practically all the
Copyright 1985 Sonic Art Editions, 2617 GwynndaleAvenue,Baltimore, MD 21207, U.S.A. Used by
permissionof the publisher.