Contemporary Music Review

ISSN: 0749-4467 (Print) 1477-2256 (Online) Journal homepage: www.tandfonline.com/journals/gcmr20

Approaches to Using Rules as a Composition

Method

Örjan Sandred

To cite this article: Örjan Sandred (2009) Approaches to Using Rules as a Composition Method,
Contemporary Music Review, 28:2, 149-165, DOI: 10.1080/07494460903322430

To link to this article: https://doi.org/10.1080/07494460903322430

Published online: 10 Nov 2009.

Submit your article to this journal

Article views: 570

View related articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=gcmr20
Contemporary Music Review
Vol. 28, No. 2, April 2009, pp. 149–165

Approaches to Using Rules as a

Composition Method
Örjan Sandred

This paper will give an overview of the author’s experience in using rule-based computing
in music composition. After discussing both imposed and voluntary constraints in
existing music, a very short description of a rule-based computer system will be given.
The concept of musical dimensions will be discussed, and used to illustrate the complexity
of implementing musical structures into a computer system. A central section in the
article will focus on different approaches to the design of musical rules. This will be
followed by a discussion of the relationship between pitch and rhythm, and the role of
motifs and gestures. Finally, two examples from one of the author’s compositions will
illustrate how rules can be used to formalize music.

Keywords: Computer-assisted Composition; Rule-based Composition; Constraint-based

Composition; Constraint Programming; Music Analysis

Constraints and Tradition

Music can only exist within a framework of constraints. Composers voluntarily utilize
some of these constraints. Other constraints are forced by common practice in music.
The notation system commonly used in Western Art music imposes several
constraints for how composers write music. Two obvious restrictions are the use of
12 pitches per octave, and the concept and organization of whole and half tone steps
(the origin of this system for organizing pitch is older than the notation itself). It is,
however, possible to stretch how we use the current notation system by extending the
number of pitches per octave by using accidentals for microtonal pitches, and it is
possible to ignore the modal organization of pitch. By extending the system we are
not able to eliminate the constraints a notation imposes; we only redefine its limits.
The notation of musical time is also restricted by tradition: musical time is
commonly thought of as being proportional. Durations in a traditional score have a
very precise relationship to each other. They relate as fractions of a whole duration
unit (i.e. the whole note). By using proportional notation, a composer chooses a
concept for time representation that will force restrictions on the music.

ISSN 0749-4467 (print)/ISSN 1477-2256 (online) ª 2009 Taylor & Francis

DOI: 10.1080/07494460903322430
150 Ö. Sandred
Composers have used their creativity to stretch the framework of constraints.
According to Moles (1966, p. 105), ‘The evolution of music seems to be a methodical
violation of previously accepted rules.’ The fact that composers tend to stretch what is
acceptable within a framework does not mean that a framework is not desired. For
example, when composers abandoned tonality as a framework, it was soon replaced
by other systems with other rules (for example, serial or stochastic techniques).

Constraints as a Method
Anyone who has studied music theory has been exposed to the method of describing
a musical style in terms of rules. These rules are typically simple, and only describe a
single stylistic detail. A single rule will therefore have limited value; only the
combination of all rules gives a valid representation of the style.
Fux’s (1943) theoretical treatise Gradus ad Parnassum originally dating from 1725
is a good example of how a set of simple rules can explain complex polyphonic
relationships where pitches, durations and meter are interdependent. The combina-
tion of all Fux’s rules will give a good description of the sixteenth-century vocal
polyphonic style.
When computers were introduced in composition, composers were challenged
with how to formalize musical structures. Hiller’s early experiments in his Illiac Suite
applied rules inspired by Fux’s theoretical work (Hiller & Isaacson, 1957, 1958). A
later and musically more mature example of a rule-based formalization is Lindberg’s
piece Engine for orchestra (Lindberg, 1996). Lindberg’s method is also inspired by
Fux’s work in the way the individual rules are designed (see below). Just as Fux’s rules
give a complete description of a musical style, Lindberg’s rules do too (but in
Lindberg’s case the rules are only valid for one piece). Lindberg’s rules constrain how
pitches can be selected for events in a pre-composed rhythm score.

Rule-based Computing
If musical structures can be formalized by sets of rules, computers can generate them.
A programming paradigm called constraint programming can be used to find values
for sequences of variables that fulfil given rules. The variables can only be assigned a
value from a domain of possible candidates. For example, if a domain contains the
candidates (1 2 3), the solution to a sequence with 5 variables could be (2 2 1 3 1). In
a musical application, we can represent a sequence of pitches as a sequence of
variables. The domain will then be a set of all pitches that can exist inside the
sequence. For example, a domain that contains the pitches (D E F# G A B C#) will
result in a sequence of pitches within this mode (for example, a sequence with 6
variables could be D D A B G F#). The solution to a musical constraint problem can
in this way be a melody.
A rule can define a relationship to a fixed value; for example, ‘the melody has to
start on F’. Rules can also state relationships between variables; for example, ‘pitch
 Contemporary Music Review 151
classes are not allowed to be repeated’ or ‘melodic intervals are not allowed to appear
more than once’. The latter two rules are context dependent, and less trivial to solve.
We need to step through our sequence of pitches and step-by-step check that these
rules are fulfilled.
If we let a constraint program search for a sequence of 12 pitches with the above
three rules, it will generate 12-tone all-interval series (Figure 1).
A domain can represent other musical parameters than pitch. If the domain
contains durations, we would search for a sequence of durations (i.e. a rhythm).

Figure 1 Three different solutions from a constraint program programmed to generate

12tone all-interval series.

Musical Dimensions
The 12-tone example in Figure 1 is a simple problem for a computer to solve. It only
constrains a sequence of pitches. Rhythm is not part of the formalization, and there
are no other voices that influence the pitch order.
Musical structures are generally very complex. The main reason for this is that
there are interdependencies between different musical parameters. Typically the
choice of a pitch depends on its duration and its position within the metric hierarchy.
For example, in the study of traditional vocal polyphony one rule is that half notes on
unaccented beats may have dissonance, but only as passing notes. This rule explains a
detailed relationship between duration (half notes), metric position (unaccented
beats) and pitch (a dissonance as a passing note depends on the preceding and the
subsequent pitches as well as the simultaneous pitches in other voices).
In this article I will use the term musical dimensions to describe the different
parameters for musical events. A musical dimension (the way I use the term) can be
thought of as a sequence of values or quantities. A melody can be built of a sequence
of pitches and a sequence of durations. These two dimensions often also relate to
their position within the metric dimension. If there is more than one voice, the other
voices create additional dimensions. Two-part music will then have five dimensions:
the first voice has one pitch dimension and one rhythm dimension, the second voice
has one pitch dimension and one rhythm dimension, and they both share the same
metric dimension.
152 Ö. Sandred
The complex dependencies between musical dimensions make it hard to master
musical structures. Students of vocal polyphony who study the above-mentioned rule
for consonance and dissonance for half notes could considerably simplify an exercise
by only using half notes when writing their melodies. This is one of the pedagogical
ideas behind species counterpoint: by making two dimensions static (rhythm and
meter), the student can focus on mastering another dimension (pitch as consonance
and dissonance).
Musical structures are complex to assemble for humans, and they are also complex
for computers to process. Regardless of computational technique, it is hard to
generate two or more interdependent dimensions at a time algorithmically. As a
consequence, computers have often been used to generate pitch or chord sequences
(i.e. only one dimension) for which a composer later superimposes a rhythm. In
some cases (for example, in Lindberg’s Engine mentioned above) a rhythm structure
was first generated independently of pitches. As a second step, the computer selected
pitches for the rhythm structure.1
When composing my own music, I rarely follow a scheme where first one
dimension is composed and then another dimension is superimposed. Dependencies
are often more complex than that: a pitch might depend and adjust to a rhythm, but
just as often a rhythm depends on and will have to adjust to a pitch. Therefore the
method of creating one dimension before (and independently of) the others is not
optimal for my own work. I would prefer a more flexible system that allows for all
dependent dimensions to be created in parallel.

Multi-dimensional Rule-based Computing: PWMC

To be able to compute musical structures with more than one dimension, I developed
the PatchWork Musical Constraint system (PWMC) in 2004. PWMC is an extension
of PatchWork GL (PWGL), a visual programming language specialized in Computer
Assisted Composition developed at the Sibelius Academy (Laurson, Kuuskankare &
Norilo, 2009).
PWMC is a system that is dedicated to rule-based computing. It is based on
conventional constraint programming techniques, and extends these with the
knowledge of musical dimensions. A conventional constraint solver would only be
able to handle one musical dimension. PWMC makes it possible to search for
polyphonic musical structures where pitches, rhythm and meter are all unknown.
PWMC uses the same basic structure as a conventional constraint solver: there is a
domain of allowed values (in the case of PWMC the domain is a mix of pitches,
durations and metric units), and there is a set of rules that state the relationship
between the variables (i.e. between the durations, the pitches and their metric
positions).
Finding solutions to multi-dimensional constraint problems is a complex task. The
constraint solver that the current version of PWMC uses builds a solution step-by-
step. The exact order for how the pitches and their durations are added to the
 Contemporary Music Review 153
solution is context-dependent. The user can interact to find the optimal strategy for
this order. A bad strategy can result in that it takes a very long time to find a solution.
All examples in this article were generated using the PWMC system, but it is
beyond the scope of this article to fully explain and describe the PWMC system.

Rule Design in Musical Constraint Programming

An important strength of working with constraints in computer-assisted composition
(CAC) is the modular nature that rules have: a complex musical description can be
broken down into many sub-descriptions with simple rules that are easier for the
human mind to grasp.
There are several approaches to the way rules can be designed. All rules below are
discussed from the perspective that they are implemented in a rule-based computer
system (as described above, in Rule-based computing). Only the first group of rules is
commonly used in constraint-based music programming today (although they can be
found in other programming techniques).

Rule Design 1
The rules used in Fux’s description of sixteenth-century vocal polyphony restrict
details within a musical structure. Fux typically focuses on the individual notes: What
melodic interval can I choose to start from a quarter note? How can I position a
dotted half note within a bar? The network of rules illustrates how the treatment of
consonance and dissonance depends on an event’s metric position and on its
duration. The rules also outline how melodic intervals relate to their metric and
rhythmic context (Figure 2). Fux’s rules are, however, sparse on the description of
form and overall progression. In order to be able to create a good stylistic imitation,
students need to develop a sense for qualities outside of Fux’s rule system. If the rules
are implemented by a constraints program, the result would be more successful if
there would be some additional information regarding the form of the sequence; for
example, if the rhythm would be given and only the pitches need to be found.

Figure 2 In this example, only one rule-type is used to restrict how pitches and durations
can be combined. The rule is a typical ‘Fux-style’ rule: it restricts what melodic intervals
that are allowed to start from notes of certain durations. The rule in this example restricts
melodic intervals from quarter notes to only be octaves, from eighth notes and dotted
eighth notes to only be fourths, from sixteenth notes to only be minor or major seconds
and from eighth note triplets to only be minor or major thirds. There is also a rule
restricting syncopations and the subdivision of metric beats. Beside these restrictions,
pitches and durations were picked at random.
154 Ö. Sandred
Rule Design 2
A contrasting rule design to Fux’s approach is to look at the overall qualities of a
sequence. We might have a preference for certain characteristics. For example, if we
want a rhythmically active music, we might prefer sixteenth notes to quarter notes. We
can formalize the distribution of durations and estimate that of all the durations in our
piece, we would like 50% to be sixteenth notes, but only 10% to be quarter notes. This
type of rule gives many options for the individual note lengths, but forces the overall
rhythmical character in a certain direction. As a consequence, the rule will not give a
detailed stylistic control, but it can help in shaping the character of a longer sequence.
There are several ways to design a rule for the distribution of events. One way
would be to always keep the sequence as close as possible to the ideal distribution.
Another way would be to estimate the maximum number of each of the note lengths
that is allowed in the sequence, and not to allow them anymore when their maximum
numbers are reached. In Figure 3, the pitches follow a distribution rule: the low C is
the most common pitch and the high B is the least common (see Figure 3 for the
exact distribution of pitches). The distribution rule is designed according to the first
method described above. The rule is set to accept +2% deviation from the given
distribution. The result is a sequence that gravitates towards the lower pitches; for
more information, please see Sandred et al. (2009).

Figure 3 A distribution rule assures that the distribution of the pitches is close to a given
probability table: 15.4% of all pitches should be C, 14.1% should be C#, 12.8% D, 11.5%
Eb, 10.3% E, 9.0% F, 7.7% F#, 6.4% G, 5.1% Ab, 3.8% A, 2.6% Bb and 1.3% B. The rule is
set to accept a deviation of maximum +2%.

The distribution of events is just one out of many possible techniques for
controlling the overall character of a sequence. Other stochastic techniques such as
Markov processes can be transformed to suit rule-based computing. There are also
techniques that are not based on stochastic principles that fall into this category. One
example is profiles. A profile can outline the melodic or rhythmic progression. A rule
that forces a dimension to follow a given profile will influence the musical direction.
An example of the latter is a method proposed by Schilingi:2 a melodic profile derived
from a musical example is imposed heuristically on a sequence. Other rules (that for
example restrict melodic intervals or harmonic relations to other voices) are imposed
as strict rules on top of the profile rule. The result is a sequence that has a profile that
resembles the original musical model, but also fulfils additional restrictions.

Rule Design 3
A musical structure is often constrained by hierarchical relationships between events
within the structure itself. Music is full of hierarchies. The nature of the metric
 Contemporary Music Review 155
dimension is to define rhythmic hierarchies, where some events are more important
and therefore often more restricted than others (many of Fux’s rules above refer to
the metric hierarchy). A composed metric dimension is not necessarily perceived; a
composer needs to be aware of how meter is expressed through pitches and rhythm.
The notation system allows very complex metric structures to be notated, and rules
are well suited to limit the types of metric subdivisions and syncopations that are
acceptable in a music style (Sandred, 2000).
We can also create other types of hierarchies than metrical ones. Letting one voice
restrict another can create hierarchies between voices. In Figure 4, the two top voices
(played on two drums) are both constructed as rhythmic patterns: after four
durations the rhythms repeat themselves (the two patterns have different lengths).
The two percussion voices restrict the rhythm in the bass line: every starting point for
a note in any of the two percussion voices will force a new note to start at the same
time point in the bass line. The bass line is built of eighth notes and sixteenth notes.
As a consequence of the hierarchies, eighth notes are rare in the bass line. An
additional rule restricts eighth notes to only be performed as a low C (other pitches
are picked at random).

Figure 4 Hierarchical rules between the percussion voices and the base line restrict the
rhythm in the base line.

Rule Design 4
Existing music examples can serve as a model for new music sequences. The examples
can be generalized through music analysis (for example, an analysis of the harmonic
progression). Rules can refer to the analysis and apply its characteristics to the new
music sequence.
It is important to understand that all methods for music analysis have their
limitations: only a limited number of characteristics can be captured through
analysis. An analysis might contain very limited information of a musical style. Often
it is necessary to combine a rule that is generated from analysis with other types of
rules to make musical sense.
There are several analysis techniques that are suitable for computers, and therefore
can be used in rule-based computing. These include analysis of the application of
traditional counterpoint rules, and analysis of the distribution of events as discussed
156 Ö. Sandred
under Rule Designs 1 and 2 above. For example, a computer can count how many
times given rules fail in an existing score and in this way give an analysis of whether
or not they are valid. An analysis of the event distribution is simply achieved by
calculating the probability table for the events in a sequence.
A recent development in music analysis is the morphological descriptions of
melodic profiles (Schilingi & Voisin, 1997). Two examples of techniques for
morphological descriptions are the new/old-analysis and the energy profile analysis
(Aralla, 2003). The new/old-analysis relates to information theory. In information
theory, the predictability of an event is a measurement of its information content. If
an event is predictable, it does not carry any new information. An event that is
repeated over and over again will be perceived as more and more predictable. The
opposite situation is equally predictable: if every new event in a sequence is different
and unique, it will be of no surprise if the next event also is unique. In the context of
everything in a sequence is unique, an added, unique event does not carry any new
information.
The measurement given by the new/old-analysis gives can be understood as the
predictability of events in relation to the context in which they appear. The quantity
that is measured is labelled ‘newness’ (i.e. how new an event will be perceived). A
sequence with a repeated value will result in a new/old-analysis where the third event
has zero ‘newness’. The opposite case (where every new event is different from all
existing events) will result in a new/old-analysis with a gradually falling ‘newness’.
In most melodies, unique pitches are mixed with re-occurring pitches. The new/
old-analysis is a tool to describe the impact the individual pitches will have in their
context. The necessary algorithmic processing of a sequence in a new/old-analysis will
not be discussed in this article; for more information, please see Aralla (2003).
An energy profile is another way to display the outcome of a new/old-analysis: it
describes the degree of difference in ‘newness’ between neighbouring events (Figure 5).
A sudden change in the new/old-analysis will be visible as an energy peak in the energy
profile. The energy profile is often able to point out differences in melodic behaviour. It
can also point out where a melody modulates or where new pitches are introduced. The
energy profile has proven to be useful both for tonal and atonal music.
An energy profile can be used in rule design. The rule would then force the
characteristics of an existing energy profile on a new sequence. The energy profile
influences the event ordering. It does not carry any information regarding tonality,
dissonance/consonance or metric hierarchy. It is technically possible to apply an
analysis of the energy profile to the rhythmic dimension; however, the musical
meaning of this is less explored.
Rules that have a weak definition on their own can provide valuable additional
directions to other rules. Figure 5 illustrates this: the rule for the energy profile only
has an impact on the order of events. Most stylistic characteristics, such as tonality
and voice leading, are not inherent in this rule. Only by supplementing the rule for
the energy profile with two simple rules regarding pitch (Figure 5), is it possible to
rebuild the Bach theme that the energy profile is based on.
 Contemporary Music Review 157

Figure 5 The graph shows the energy profile of the theme from Bach’s Fugue XVI from
Das Wohltemperirte Klavier. The main rule in this example forces the characteristics from
this energy profile on a melody with 22 pitches. The energy peak is also constrained to
have the pitch c sharp. Two more rules restrict the sequence: one rule forces the pitches
on accented metric points to be consonant to a given harmonic progression (gm/cm/D/
gm/gm/c/gm/gm/gm/D/gm/A/D), and one rule specifies that melodic intervals from
sixteenth notes can only be minor or major seconds. This information is enough to
recreate the pitches for the theme from Bach’s Fugue XVI from Das Wohltemperirte
Klavier. The rhythm was not generated but given as a restriction.

The four examples of approaches in rule design above are not an exhaustive list of
possibilities. They only point out that there are many more approaches in rule-based
computing than what has commonly been explored in musical constraint
programming.

The Symbiosis of Pitch and Rhythm

In rule-based composition, the constraints we impose on our musical structures must
have the potential to shape aesthetic qualities. For this reason, I occupied myself for
several years with the question of what a good rhythm is. It is not obvious how to
formulate rules for durations that result in a satisfying rhythm.
One of my conclusions was that rhythm should not be discussed as a phenomenon
in its own right. Rhythm comes very rarely in such a pure form that it is unaffected by
other musical dimensions. Even in pure percussion music, timbre plays a role. The
colour of the events will have an affect on how a rhythm is perceived. A good rhythm
must therefore be seen in the context of the pitches/timbres it will carry.
My belief is that rhythm in many cases has been composed as a result of choices
made in the pitch dimension. There are clearly exceptions; if rhythm is separately
generated from an algorithm, there is no conceptual link to the pitch dimension.
However, in the end, the rhythm will be performed with timbres or pitches, and the
listener’s perception of rhythm will be linked to these parameters. For example, in an
isorhythmic composition technique, rhythm is treated separately from pitch. An
important aspect of isorhythmic structures is that when the rhythm pattern (the
158 Ö. Sandred
talea) repeats, it will be collored by new pitches: each repetition of the rhythm will
therefore be perceived differently. By separating rhythm from pitches, isorhythmic
techniques explore the effect of recombining the two dimensions again.
To clarify what I mean with the symbiosis of pitch and rhythm, I will give an
example: In Figure 2, a single rule inspired by ‘Fux-style’ rules is all that is used to
generate a simple melody (there is an additional rule to avoid complex positions of
durations in relationship to the metric dimension). This seemingly simple rule creates
a surprisingly elastic melody. The melody has its obvious weaknesses (form being
one), but the single link between the pitch and rhythm dimensions already gives a
first fragment of a musical expression.
The example in Figure 2 illustrates how music is perceived as a combination of
musical dimensions. The melodic octave leaps give the quarter notes a characteristic
musical expression. In this example, the rhythmic qualities can only be understood in
the light of the melodic intervals they express. This is why the rhythm has its elastic
quality. The rhythm is here the result of decisions in the pitch dimension (and vice
versa).
Other types of links between pitch and rhythm give different types of expressions
(see for example the bass line in Figure 4).

Musical Building Stones

Up until now, we have assumed that the only constraint in our rule-based
composition method, beside the notation system itself, has been the set of rules we
apply to pitches, durations and metric units. The secret of a successful musical
language would then exist in well-designed rules. Composers have, however, had
other approaches when structuring the musical material than using rule restrictions.
One possible strategy has been to view the musical building blocks on a higher level
than on the level of single notes. Higher-level groupings of pitches and durations
create motifs and gestures.
In musical styles where motifs and gestures are valid concepts, their use can
simplify the composition process: it is often easier for a composer to assemble a piece
of music out of gestures and motifs than to build the whole piece note-by-note. The
same approach could be possible in rule-based computing: rules often give a very
detailed, low-level control of the musical result. If motifs and gestures could be
outlined before the low-level control is applied, it would increase the possibility to
reach a desired result.
Gestures and motifs represent fragments of the musical language in a piece. Their
design will have a great impact on the final musical structure. The concept of gestures
and motifs can be applied by rules, but it is more efficient to implement musical
shapes already in the domain. The domain would then contain groupings of pitches
and durations instead of single pitch and duration values. A group in the domain
cannot be broken down into its individual elements. For example, if the pitch C3 only
exists in a group (C3 E3 G3) in the domain, then C3 will always be followed by E3
 Contemporary Music Review 159
and G3 in the final piece. The first rule-based computer system that implemented the
concept of motifs in the domain was the OpenMusic Rhythm Constraints library
(OMRC) by the author (Sandred, 2000, 2003).
It is possible to combine pitch and rhythm motifs with individual pitches and
durations in a domain. Pitch motifs may or may not specify exact pitches. A pitch
motif that does not specify the exact pitches only outlines its melodic contour. A
pitch motif can for example be defined as a minor third up followed by a pure fifth
down (i.e. a broken triad). If this shape is part of the domain, the melodic line can at
any point follow it (if the rules allow it). Examples will follow below.

Examples from Whirl of Leaves

Two music examples will illustrate how music can be composed using a rule-based
method. The two examples both come from my composition Whirl of Leaves for flute
and harp. The piece was premiered at the UltraSchall Festival in Berlin, January 2009.
Whirl of Leaves was entirely composed using rule-based computing.

Example 1
The first example is taken from the third section of the piece, starting at measure 50.
The meter is fixed to 4/4. The domain contains all the pitches and durations that can
exist in the flute and harp parts: the flute can have any pitch between F4 and Ab6, and
the harp can have any pitch between C2 and C6. These ranges give access to both low
and high registers in both instruments (avoiding the extreme ends). The durations in
the domain are grouped into rhythm motifs. These motifs can be seen in Figure 6: the
idea is to give the harp fast-moving gestures with different subdivisions of the beat
(all rhythm motifs have the total duration of a quarter note). The flute domain has
two motifs that will match the subdivision of the first and fourth motifs in the harp
domain. The domain for the flute also contains four different rests to choose from.
In addition to pitches and rhythmic motifs, the domain contains information
about chords. There are five chords with corresponding modes that constitute the
base for the harmonic structure (Figure 7). Three long durations will be the building
blocks for the harmonic rhythm (see Figure 6). The program will determine the order
of the durations and the chords for the harmonic progression.
There are 12 rules restricting the score. Two types of rules are used in this example:
strict rules and heuristic rules. Strict rules have to be followed without exception.
Heuristic rules only give preferences to certain characteristics of the music.
Hierarchical rules (described above) define relationships between the harmonic
rhythm, the rhythm in the flute part and the rhythm in the harp part.

Rule 1: Every chord change in the harmonic rhythm triggers a new event in the flute
part.
Rule 2: The events in the flute part trigger events in the harp part.
160 Ö. Sandred

Figure 6 The rhythm domain for the first example. The durations for the flute, harp and
the harmonic rhythm are treated separately in the domain.

Figure 7 The pitch domain for the harmonic structure. Each chord is linked to a mode:
the modes contain a few more pitch classes than their corresponding chord.

The two rules force every harmonic change to be expressed as new events in both the
flute and the harp. The second rule also forces the rhythm in the harp part to follow
the flute when the flute plays, but when it has rests the harp is unaffected by this rule.
The rhythm in the flute part is controlled by two rules balancing each other. Rule 4
is heuristic and can only be followed when rule 3 allows it.

Rule 3: At the points of chord changes in the harmonic progression, the flute has to
play a motif using the first rhythm in Figure 6.
Rule 4: Rests are always preferred in the flute.

The result is that the flute will accentuate every chord change by playing a sextuplet
gesture. Most of the other time the flute will have rests.
The rhythm in the harp has an additional heuristic rule to follow. This rule will
create a rhythm that (almost) always changes the subdivision of neighbouring beats.
A rhythmical flow with a non-mechanical quality is desired.

Rule 5: It is preferred that the rhythm motifs (Figure 6) are not immediately
repeated.
 Contemporary Music Review 161
Pitches in both instruments are linked to the harmonic progression.

Rule 6: Only pitch classes from the current mode in the harmonic progression
(Figure 7) can be used in the harp.

Every mode in Figure 7 is designed so that it is possible to play all its pitches
without changing the pedals on the harp.
Pitches on beats are more restricted in the harp.

Rule 7: Only pitches from the current chord in the harmonic progression can be used
on beats or in-between beats (i.e. events appearing on an eighth note grid) in
the harp.

Rules 6 and 7 create a harmonic structure that is expressed through the metric
structure. The chord progression will be clear thanks to these two rules.
Pitches in the flute are linked to pitches in the harp.

Rule 8: Simultaneous pitches in the flute and the harp have to be of the same pitch class.

Since the flute mainly will play the sixteenth note sextuplet at chord changes, this rule
will only have an effect at these points. The flute will thus accentuate chord changes
(Rules 3 and 4) by doubling the pitches in the harp (Rule 8). Both instruments will
play a sextuplet at these points (Rules 2 and 3).
A few rules restrict the melodic line in the harp.

Rule 9: Pitches are not allowed to be immediately repeated.

Rule 10: Two steps in the same direction have to be followed by a step in the contrary
direction.
Rule 11: Melodic intervals have to be smaller than a fifth unless the harp plays in
homophony with the flute. In the latter case the melodic intervals can be at
the maximum an octave þ fourth.

The last two rules create a melodic line that stays in a narrow register until it is
doubled by the flute. At that point it may jump to a new narrow register.
One additional rule restricts pitches in the flute.

Rule 12: Pitches are not allowed to be repeated immediately.

Since the pitches in the flute follow the harp (Rule 8), Rule 9 already has some effect
on the flute line. Rule 12 ensures that immediately repeated pitch classes will be put
in different octaves.
The final score for this example can be seen in Figure 8. The two sustained, low C
sharps in the flute were added by hand after the computer had generated the score.
162 Ö. Sandred

Figure 8 Measure 50–52 from the final score of the first example from the composition
Whirl of Leaves by the author. The harmonic progression is shown under the score.

Example 2
The second example is taken from the second section of the piece. It illustrates how a
domain may contain pitch motifs only specified by their profiles (i.e. they can be
transposed to any pitch).
The rhythm domains for both voices in this example are built of individual
durations (and one rest), shown in Figure 9. The pitch domain for the harp is close to
identical to the previous example (the same range of individual pitches), but the pitch
domain for the flute is different: it contains nine transposable pitch motifs
(Figure 10). All motifs have 4 pitches (¼3 melodic intervals) and the distance from
the highest to the lowest pitch is always an octave þ fourth. Transposable pitch motifs
restrict the melodic shapes. From the viewpoint of the domain, any pitch can exist in
the flute part as long as it fits into one of these shapes.

Figure 9 The rhythm domain for Example 2.

In addition to the design of the pitch motifs in the domain, four rules control the
flute melody.

Rule 1: Pitches cannot be lower than C4 (the lowest pitch on the flute) or higher than F6.
Rule 2: Every sixth event has to be a rest.
Rule 3: Pitches higher than G5 have to be at least a half note long.
Rule 4: Pitches G5 or lower have to be shorter than a half note.
 Contemporary Music Review 163

Figure 10 The pitch domain for the flute part. The motifs can be transposed to any pitch.

The first rule is necessary to assure that the pitch motifs do not walk out-of-range of
the flute. The second rule divides phrases. The third and fourth rules link the choice
of rhythm to pitches (as discussed under ‘The symbiosis between pitch and rhythm’
above): they will create melodies that ‘hang’ in a higher register.
There are two rules for the melodic line in the harp.

Rule 5: Pitches cannot be repeated immediately.

Rule 6: Melodic intervals have to be smaller than a major sixth.

The main melody will be in the flute: the pitch motifs in the domain are the most
important restriction for shaping the pitches in the flute. The melodic line in the harp
is less restricted. In addition to Rules 5 and 6, the main constraint for the pitches in
the harp can be found in Rule 8 below.
164 Ö. Sandred
A few rules restrict the relationship between the voices. There is a hierarchical rule
between the flute and the harp, similar to the one in the first example.

Rule 7: Every event in the flute triggers an event in the harp (the harp might have
more events in-between the events triggered by the flute).

A difference to the previous example is that the domain does not contain any
information regarding the harmony. Instead, the harmony is implemented through a
rule. This rule compares all harmonic intervals and chords that exist in the score to a
predefined list of chords.

Rule 8: Only harmonic intervals and chords that exist in the predefined list of allowed
chords can exist in the score.

The chords to which Rule 8 refers are fixed in pitch and cannot be transposed.
In the final score we can see the effect of the rules. The phrases in the flute are always
five notes long, and are all divided by an eighth note rest (the only rest available in the
domain). When the flute melody goes above G5, it becomes more sustained. The exact
shape of the melody is the result of linking pitch motifs from the domain: the melody
starts with the ninth motif (Figures 10 and 11), followed by the sixth motif, then the
ninth motif again, etc. The rhythm in the harp is linked to the flute. The pitches in the
harp are a result of harmonic restrictions (Rule 8) and melodic restrictions (Rules 5
and 6).
I feel that I have reached the musical result I intended for Whirl of Leaves. The
problem with the rule sets presented above is of another nature: there are very few
considerations regarding playability within the rule sets. The first example creates a
music that can be very demanding for the performers, while the second is in general
very easy to play.

Figure 11 Measures 23–30 from the final score for the second example from the
composition Whirl of Leaves by the author. The extra pitches in the chords in the harp
part are added by hand after the computer generated the score. These added pitches all fit
into the chords.
 Contemporary Music Review 165

Conclusion
A composer can use rule-based computing to generate a musical score. Rule-based
computing has the strength of breaking down a complex problem into simple sub-
problems.
A score can be assembled from musical elements such as pitches and durations, or
from higher-level groupings of these elements (i.e. motifs and gestures). A composer
can interact with the system by shaping the motifs and gestures, and by designing the
rules. A rule-based system allows the composer to experiment with the impact
individual rules, motifs and shapes have when they are combined into a complex
system of dependencies.

Notes
[1] As a composition student of Lindberg in 1996 (at the time he composed Engine), one of the
topics we studied was the Score-pmc constraint engine inside PatchWork (a visual music
programming language). The nature of Score-pmc is that it needs a complete rhythm structure
before the program can assign pitches to the predefined rhythm score. The piece Engine was
composed with (among other tools) the Score-pmc.
[2] This can be found in the JBS-constraints library inside PWGL.

References
Aralla P. (2003). Analisi morfologica. Un modello matematico per descrivere la relazione fra
struttura morfologica del messaggio, attività mnemonico-percettiva e risposta psichia. In
PRISMA 01 (pp. 51–75). Milano: Euresis Edizioni.
Fux J. (1943). Steps to Parnassus: The study of counterpoint. translated by Alfred Mann with the
collaboration of John St Edmunds. Original title: Gradus ad Parnassum (1725). New York:
W.W. Norton & Company.
Hiller, L. & Isaacson, L. (1957). Illiac Suite. Score, New York: Theodore Presser Co.
Hiller, L. & Isaacson, L. (1958). Musical composition with a high-speed digital computer. Journal of
Audio Engineering Society, 6, 154–160.
Laurson, M., Kuuskankare, M. & Norilo, V. (2009). An overview of PWGL, a visual programming
environment for music. Computer Music Journal, 33, 19–31.
Lindberg, M. (1996). Engine. Score, London: Chester Music Ltd.
Moles, A. (1966). Information theory and esthetic perception. Chicago, IL: University of Illinois Press.
Sandred, Ö. (2000). OMRC library tutorial. Version 1.1 (2nd ed.). Paris: IRCAM.
Sandred, Ö. (2003). Searching for a rhythmical language. In PRISMA 01 (pp. 117–127). Milano:
Euresis Edizioni.
Sandred, Ö. (2009). Whirl of Leaves for flute and harp. On Cracks and Corrosion, CD/CD-ROM,
audio recording and score, Boston: Navona Records.
Sandred, Ö., Laurson, M. & Kuuskankare, M. (2009). Revisiting the Illiac Suite—a rule-based
approach to stochastic processes. Sonic Ideas/Ideas Sonicas, 2, 42–46.
Schilingi, J. B. & Voisin, F. (1997). Morphology. Fonctions d’analyse, de reconnaissance, de
classification et de reconstitution de séquences symboliques et numériques (Reference). Paris:
IRCAM.