Musink: Composing Music through Augmented Drawing
Theophanis Tsandilas, Catherine Letondal & Wendy E. Mackay
In|Situ, INRIA Saclay, Ile-de-France
LRI, Building 490, Univ. Paris-Sud, Orsay Cedex, France
email@example.com, firstname.lastname@example.org, email@example.com
We focus on the creative use of paper in the music compo-
sition process, particularly the interaction between paper
and end-user programming. When expressing musical
ideas, composers draw in a precise way, not just sketch.
Working in close collaboration with composers, we de-
signed Musink to provide them with a smooth transition
between paper drawings and OpenMusic, a flexible music
composition tool. Musink’s built-in recognizers handle
common needs, such as scoping and annotation. Users can
also define new gestures and associate them with their own
or pre-defined software functions. Musink supports semi-
structured, delayed interpretation and serves as a custom-
izable gesture browser, giving composers significant free-
dom to create their own, individualized composition lan-
guages and to experiment with music, on-paper and on-line.
Creativity, interactive paper, participatory design, musical
interfaces, end-user programming, gesture interfaces.
ACM Classification Keywords
H.1.2 [User/Machine Systems]: Human Factors, H.5.2
[User Interfaces]: Evaluation/methodology, Theory &
methods, Prototyping, User-centered design.
We are interested in developing tools that support the crea-
tive design process, in particular, the composition of origi-
nal music. Contemporary music composers are an interest-
ing user group because they combine a deep artistic sense
with, often, highly mathematical and technical skills. They
work with a standard notation, evolved over centuries, and
then invent new musical expressions to explore and repre-
sent new musical ideas. They often express these ideas on
paper, working out different aspects on different levels,
over time. Yet they also work extensively with the com-
puter, developing functions for new sounds, many of which
cannot be captured with traditional music notation, in a so-
phisticated form of end-user programming . A key chal-
lenge then is to create tools that support this creative proc-
ess and to provide composers with a rich, nuanced ex-
change between paper and the computer. We need to en-
hance, rather than replace, the composer’s ability to gener-
ate and explore musical ideas.
Although a number of researchers have developed paper-
based interfaces using Anoto1 technology, the usual empha-
sis is on improving productivity, enabling users to accom-
plish particular tasks by combining paper and electronic
documents. Our focus is slightly different: we are interested
in understanding and supporting the creative design process
itself. We have been working closely with contemporary
music composers, who use state-of-the-art music composi-
tion hardware and software, combined with extensive tech-
nical support. Even so, they continue to make use of paper
as a tool, from earliest sketches to the final printed score.
We decided to collaborate with them to better understand
both the role of interacting with paper in a creative context
and how to integrate paper and end-user programming tools
to enhance creativity. We build upon their existing, highly
personal methods of expressing ideas to support an open-
ended, customizable cycle of interaction between paper and
This article begins with a description of our initial study of
composers, including the insights that led to our initial de-
sign of Musink. Based on Anoto technology, Musink allows
users to express, annotate and interact with personally gen-
erated musical ideas, moving back and forth between paper
and OpenMusic , a state-of-the-art music composition
system. We then present our findings from a series of mini-
workshops with individual composers who experimented
with Musink and our rapid, iterative redesign in response to
their suggestions. We then offer insights into how creative
individuals find novel ways of interacting with paper and
the requirements for tools such as Musink, to support their
diverse needs. We conclude with a discussion and direc-
tions for future research.
Paper is a powerful medium, easily underestimated, that
provides users with a range of opportunities, from initial
sketches to definitive publication. We have long been inter-
1 The Anoto pen’s camera captures gestures on paper printed with
a computer-readable, human-invisible dot pattern www.anoto.com
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. To copy otherwise,
or republish, to post on servers or to redistribute to lists, requires prior
specific permission and/or a fee.
CHI 2009, April 4–9, 2009, Boston, Massachusetts, USA.
Copyright 2009 ACM 978-1-60558-246-7/09/04…$5.00
ested in the dilemma faced by people with excellent reasons
for using both paper and computers. Early work on the
Digital Desk  and Video Mosaic  explored the
benefits of combining paper and computers. Augmented
paper has been studied in several domains, e.g., annotated
engineering drawings , flight strips , text editing [9,
10] and hybrid laboratory notebooks . Anoto technology
made interactive paper practical and launched applications
for scientists [27, 35] while encouraging exploration of
multi-media indexing [4, 25], copy-paste between paper
and documents  and paper-mobile interfaces .
Several research groups have focused on the architecture of
paper interaction, notably PapierCraft , PaperProof
, and ModelCraft . Each approach proposes a spe-
cific pen-based set of gestures that are linked to pre-defined
computer functions. Users can use these gestures to perform
a command, e.g., copying a picture from one page to an-
other, replacing a word, or editing a physical model. Al-
though extensible, the gesture sets are defined by the appli-
cation designer, not the user.
More generally, HCI researchers have begun exploring
creativity, not just productivity, with specialized confer-
ences such as Creativity & Cognition. Csíkszentmihályi’s
 work on ‘flow’ have influenced key HCI researchers
, who advocate tools that support idea generation and
sharing [1, 3]. This corresponds with an increasing interest
in design, both as a method and a focus of study.
Similarly, research in interfaces to support music composi-
tion, has grown significantly since Buxton’s early work 
on interaction techniques for drawing musical notation. The
NIME (New Interfaces for Musical Expression) conference
combines music and HCI research and is a fertile area for
end-user programming , because musicians and com-
posers often create their own instruments and electronically
generated sounds .
We are interested in how to create an interactive paper in-
terface for contemporary music composers that supports
individual creativity and bridges the gap between paper-
based and on-line expression of musical ideas. We began by
studying composers, to understand the current role of paper.
STUDY 1: INTERVIEWING COMPOSERS
Composers pose an intriguing user interface challenge:
How can they use the computer as tool, but still create art?
They do not want increases in efficiency, per se, but rather
support for reflection and exploration of ideas. Composers
would reject a system that automatically composes for
them; instead, they seek tools that provide maximum indi-
vidual freedom of expression but also maximum control
over the computer, throughout the creative process.
We interviewed 12 composers and musical assistants at
IRCAM, a world-famous center for contemporary music in
their offices or the labs where they composed music. Most
composers are proficient or skilled computer users; they can
also rely on music assistants, usually computer scientists
with musical training. We also met with an IRCAM re-
search team to discuss their longitudinal study of a single
composer and how he used paper while composing .
We found that, despite access to the latest computer-music
tools, these composers continue to use paper documents.
However, they were dissatisfied with the lack of connection
between their off-line scores and the on-line software that
generates the resulting music. Details of the study appear in
; here we focus on the results that affected the design of
Musink, specifically chronology and choice.
Chronology of paper and computer use
Why don’t composers of electronic music simply use a
computer? Clearly, it is not due to fear of computers, nor is
it particularly related to user interface problems, since most
of their tools are designed by and for musicians. Rather,
they choose the appropriate medium for the purpose at hand
. Composition progresses from an initial creative stage
to the final piece, with much iterative development in be-
tween. Design artifacts evolve over time, from quick early
sketches, through systematic explorations of alternatives, to
the definitive printed work.
The desired characteristics of the design medium evolve as
well. In the beginning, composers use paper because it is
flexible, easy to transport and less cumbersome than a sty-
lus on a graphics tablet. Most importantly, paper permits
free associations and provides a direct link between a hu-
man gesture and a musical idea. We were struck by the in-
novative ways these composers found to represent their
ideas. Figure 1a illustrates the relationships among different
elements of a symphony; Figure 1b shows where the com-
poser has added numbers under each instrument part to in-
dicate loudspeaker assignments to create a specialization
Figure 1. (a) Organization of the components of a symphony.
(b) Annotations on a hand-written score
In the middle of the process, paper and computers each of-
fer flexible, but different, modification capabilities and
power of expression. However, in the final stages, paper is
no longer valued for its flexibility, but rather for its perma-
nence as a reference point and archival artifact. One com-
poser reported that he even paid someone to rewrite his
scores by hand, from the ‘final’ version on the computer, so
as to create the true ‘original manuscript’.
Choosing between paper and computer
In many cases, composers move easily back and forth be-
tween paper and computer, with no conflicts. Figure 2a
shows a composer using the printed score to reference on-
line musical materials: a poem used as data for building
sounds, a rhythm series, drawings and code. He can browse
the printed score and uses folders containing paper of vari-
ous sizes to keep track of the structure of his composition.
Figure 2. (a) Navigating between paper score & on-line files.
(b) Extended musical notations for 1/4 and 1/8 tones.
However, some composers experience a conflict: neither
medium suffices by itself, nor they do not work well to-
gether, making it difficult to choose. In electronic music,
the computer is the instrument, but also a tool for creating
new instruments and exploring a musical space. However,
the preferred medium for imagination and writing remains
paper, because it is slow and static. Most composers only
use electronic music editors, such as Finale, when they
want to implement an idea that has been already expressed
Figure 3. (a) Score with added symbols to show actions to be
performed by the choir. (b) ‘Electronic’ is written vertically,
with links to various computer programs that are played in
addition to the more traditional score to the right.
In electronic music, the choice of the final format is more
complex than in classical music: with an electronic format,
the composer has to deal with non-conventional notations,
such as in Figure 3a, where the composer includes graphical
notations that indicate particular sounds for a choir. In con-
trast, Figure 3b shows a hand-written score with links to
various electronic documents and statements that trigger
particular computer programs while the piece is being
played. This hand-written paper score serves as the key
reference point, with links to the computer, but the final
work is fundamentally located both on-line and on paper.
Implications for design
These interviews provided us with useful insights about the
role that paper plays in creativity and suggested new ways
that interactive paper can support composer’s inventive
design process. Each composer has a unique creative proc-
ess, with personal strategies for expressing sounds,
rhythms, variations and structures. Each composer also has
a set of custom-made computer tools, e.g., created with
AudioSculpt  and manipulated with OpenMusic, a vis-
ual programming environment to support composition .
At the most basic level, it was clear that a future system
must provide a flexible way of linking composers’ drawings
to their music composition tools. However, we also wanted
to create a testbed for exploring creative design with inter-
active paper. One of our system’s guiding objectives was to
give users maximum control over the assignment of mean-
ing to their gestures, which required a trade-off between
openness and recognition. Offering a blank slate with com-
plete openness and perfect recognition is impossible. The
challenge was to create enough scaffolding to permit suffi-
cient recognition to be useful, while allowing users to in-
vent a wide variety of different representations of their mu-
sical ideas and add meaning over time, as required.
MUSINK: CYCLING BETWEEN PAPER MUSIC SCORES
AND COMPUTER COMPOSITION TOOLS
We designed Musink as an extensible gesture-based lan-
guage that uses a common musical structure, the 5-line staff
of a musical score and a small number of basic, recogniz-
able gestures. Users can define a personal interaction vo-
cabulary and associate it with computer tools such as
OpenMusic. Figure 4 illustrates a scenario in which a com-
poser expresses a musical idea on paper and manipulates it
using Musink’s Gesture Browser and OpenMusic.
Scenario: Leonard has invented a new type of crescendo
that vibrates according to a particular pattern he defined
in OpenMusic. He prints an earlier version of his composi-
tion onto Anoto paper and uses an Anoto pen to draw
angled lines over several series of notes. He uploads the
pen data and opens the Musink Gesture Browser, which
displays the hand-annotated version of his score. He se-
lects an instance of a crescendo and right-clicks to open a
gesture-definition dialog box. He then specifies the de-
tails, including a link to the relevant OpenMusic function.
Musink automatically recognizes most instances of the
crescendo gesture. For the rest, Leonard points to the
unrecognized gesture and uses a marking menu to assign
the crescendo function. Later, Leonard explores several
implementations of the new crescendo in OpenMusic’s
workspace and tests their outcome as different variations
of his piece.
The main goal of Musink was to provide a tool that would
stimulate our design explorations with composers. We be-
gan with a basic design goal, i.e. to optimize the trade-off
between openness and recognizability but also remained
open to new ideas that emerged as part of our participatory
design process. Given the prevalence of OpenMusic as a
tool at IRCAM, we decided to integrate several aspects of
OpenMusic’s design philosophy into our approach. Specifi-
cally, we treat gestures as functions that can take properties
of musical objects as arguments, e.g., their rhythm or pitch,
and generate new objects. We started in the middle of the
design process, when musical scores are already present,
and explored how to augment them, using Musink to create
new scores. We knew that most composers do not simply
write a finished section of a score, but rather work progres-
sively, adding layers of nuance over time. Thus, Musink is
designed to allow composers to reprint their scores in mul-
tiple cycles, annotated with new gestures and reprocessed
together with other evolving musical objects.
Figure 4. A Musink scenario
(a) Drawing on paper: expressing a new type of crescendo
(b) OpenMusic: defining the crescendo’s vibration pattern
(c) Musink Gesture Browser: defining the crescendo class
The challenge is how to provide the above functionality
while respecting the natural role of gestures on traditional
paper. Although paper-based gestures sometimes serve as
commands that perform specific software operations, they
are also declarative, with a representation designed to be
recognizable by people. We thus studied existing forms of
annotations of musical scores and integrated them into
Interacting with Musink on Paper
The basic Musink syntax supports the three elements identi-
fied by Winget : symbols drawn over or under individ-
ual notes or phrases, numbers, usually representing finger-
ing or tempos, and text. This is consistent with Chapuis et
al.’s classification  and enables us to handle both simple
(single-trace) gestures and complex gestures in which mul-
tiple traces are logically linked together. Musink gestures
thus include: a scope, to specify the elements of the score to
which a function is applied, a temporal range within the
score, and (3) a graphical representation. Figure 5 illustrates
several examples of complex Musink gestures.
Figure 5. Examples of operations expressed with Musink
Figure 6. Examples of Musink basic gestures
Figure 6 shows Musink’s basic, generic gestures:
• Pointers: describe specific locations within the score’s
timeline. Forms: vertical curves; arrows.
• Scoping gestures: define a range within the score, ei-
ther as a set of musical symbols or a temporal range.
Forms: closed curves; horizontal strokes under or over
a staff; parenthesized scopes (as in PapierCraft ).
• Text and Parameters: may be an annotation, an identi-
fier or a parameter. Forms: parentheses; any gesture
enclosed within a closed curve, e.g. a circle or a rec-
tangle. May be linked to pointers or scoping gestures if
the surrounding curve touches or is close to the gesture.
• Connectors: are supplementary strokes that group
elementary gestures. Forms: line segments that visually
connect the traces of two gestures; marks indicating a
group of traces with a series of small line segments,
(useful when traces are distant, e.g., they appear on dif-
Semi-structured delayed interpretation
Users do not need to have a formal semantic definition of a
Musink gesture as it is being drawn on paper. It may act
solely as a structural element in the score or as a symbol
that represents an abstract idea. The user can revisit it later
to, for example, assign it semantic meaning or link it to
another gesture. We refer to this as semi-structured delayed
interpretation. Musink uses identifiers to define semantics.
A pointer or scoping gesture may use its own graphical rep-
resentation as an identifier. For example, the zigzag shape
of a horizontal line may act as the identifier for a “tremolo”
gesture, distinguishing it from other horizontal lines. Alter-
natively, a text ‘tag’ may act as an identifier when attached
to any pointer or scoping gesture. Any identifier can repre-
sent a computer function, e.g., an OpenMusic patch. The
function can take any of the following as arguments: score
positions, musical symbols, temporal ranges, text and nu-
meric parameters associated with the identifier, either di-
rectly or through connectors.
Interacting with Musink on the Computer
Figure 4c shows how users can assign semantics to gestures
via Musink’s Gesture Browser. The interface is imple-
mented in Java 6 using the PaperToolkit framework  to
load pen data. The main pane of the Gesture Browser dis-
plays the PDF of the printed musical score augmented with
an interactive layer that shows strokes drawn on paper. A
smaller pane lists the pen data available to load. The toolbar
to the left includes tools for updating the recognition of
gestures, zooming in and out, visualizing the underlying
score model, adapting the sensitivity of recognition, and
directing recognized function calls to OpenMusic. The three
panes to the right provide: a list of user-defined operations
or functions that can be linked to gestures; a list of defined
gesture identifiers; and a list of textual/numerical elements,
intended for use as function parameters. Thumbnails for the
classes in the two lists are automatically generated from the
associated gestures, as they are first defined.
Interaction with Gestures
The Gesture Browser lets users define new gestures and
refine gesture recognition results. Users can right-click on
the gesture’s trace to display a marking menu  (Figure
7a). Users can remove a basic gesture, revise its recognized
scope, define a new gesture class (Figure 7b) and associate
(or disassociate) the gesture with a previously defined class.
Gestures classes can function either as gesture identifiers or
as parameters. They can also be linked with user-defined
OpenMusic functions as they are defined. Figure 7c shows
how the user can modify the scope of a gesture and Figure
7d shows how the user can review the assignment of func-
tion arguments on complex gestures.
Users can define new functions on the fly, by assigning a
unique function name and specifying its arguments. Sup-
ported argument types correspond to the data types that a
gesture can represent: sets of musical symbols, pointers in
the score, temporal ranges, and textual/numerical values.
The actual implementation of functions is beyond the scope
of the Gesture Browser. Instead, it sends the function name
and arguments of the recognized gestures to OpenMusic
from which the user can create a function or patch to proc-
ess a call.
Figure 7. Interacting with Musink to classify a gesture
(a) right-click on a gesture to activate a marking menu
(b) choose ‘classify as’ and specify crescendo
(c) modify the scope of the crescendo gesture
(d) left-click on identifiers to view the assignment of function
arguments (numbers indicate recognized arguments)
We use the Open Sound Control (OSC) protocol  to
establish the connection with OpenMusic. OSC is a plat-
form-independent communication protocol used to share
data in real time between musical instruments, multimedia
devices and computers. Note that this system architecture
allows for connecting the Gesture Browser with other music
applications that support OSC, e.g., Max/MSP, but we have
only tested it with OpenMusic.
Technical Details about Gesture Recognition
Musink separates recognition into multiple steps, to sim-
plify both recognition and customization. Recognizing ele-
mentary gestures is relatively easy, since it involves only a
few fixed gestures. However, this small set can produce
multiple alternative representations for a given function.
Musink recognizes gestures in three steps:
• identify elementary strokes: pointers, scoping gestures,
connectors, and textual elements
• recognize identifiers and parameters
• match gestures grouped under recognized function
identifiers using arguments of their associated func-
The first recognition step takes the score structure into ac-
count and uses several heuristics, including the $1 recog-
nizer . The latter performs particularly well when dis-
tinguishing among open and closed curves. We use Ru-
bine’s algorithm  as implemented by the iGesture
framework  to recognize identifiers and parameters.
This fit to our design goals better than the $1 recognizer
, because it provides a reliable mechanism for rejecting
gestures that do not belong to a defined gesture set.
The third recognition step applies only to gestures that have
been recognized as function identifiers. The goal is to
match function arguments with compatible data elements
represented by pointers, scoping gestures, and textual pa-
rameters. The algorithm follows connections within a group
of gestures, starting from the identifier gesture and moving
to other connected gestures based on simple heuristics.
Preliminary Evaluation of Gesture Recognition
We ran a small user study to test the accuracy of recogni-
tion of Musink’s basic vocabulary and assess the technical
viability of our approach. We recruited six participants with
basic knowledge of classical musical notation. In each 20-
30 minute session, we asked the participant to use an Anoto
pen to interact with a pre-printed musical score. They tested
various examples of Musink’s elementary gestures in series
of five controlled annotation tasks.
elements to be recognized correct total accuracy
1. closed scopes 48 50 96%
2. parenthesized scopes 60 60 100%
3. text/parameter elements 168 179 94%
4. direct line connections 45 49 92%
5. connections with small lines 40 48 83%
6. connections with parameters 145 179 81%
Table 1. Average recognition accuracies for elementary ges-
tures tested in the study
39 61 64%
Table 1 summarizes the results. With the exception of ar-
rows, other basic symbols were recognized at 80% or better
accuracy, and most at over 90%. Many of the errors that we
observed were due to various unpredictable ways that some
participants drew on paper. For example, we detected line
connectors that were drawn by repeatedly moving the pen
from one point to another. Also, some closed curves were
drawn in two steps, creating two strokes rather than one,
which was not anticipated by the recognizer. Although we
can expect that users would eventually learn how to draw
gestures ‘properly’ to avoid such errors, such situations are
not always preventable. The study helped us assess fixable
limitations of our recognizer and improve our heuristics
accordingly. We also believe that the introduction of digital
pens with direct feedback, such as audio, currently available
in Livescribe® pens, could minimize the problem. How-
ever, as 100% accuracy may not be feasible, our continuing
goal is to support powerful online interactions that give
users the option of post-hoc specification of gestures.
EXPLORATORY EVALUATION WITH COMPOSERS
After releasing the first version of Musink, we conducted a
series of mini-workshops with individual composers. Our
first goal was to get their reactions on Musink and Musink‘s
Gesture Browser. We also wanted to stimulate a design
exploration and reflect on the potential of augmented paper
in music composition.
Participants: We met with five composers: MM, DC, JH,
MS, and PL. Two had participated in the first study. Four
were senior composers with long experience in music com-
position. The fifth was just completing his Ph.D.
Participatory design: Prior to each session, we collected
artifacts, including scores, research articles and analyses of
their work . We had a second meeting with four out of
the five composers. In the interim, we modified Musink to
provide novel functionality that they had suggested. This
allowed us to better capture the needs of each individual
composer and explore several design alternatives. We asked
them to bring drafts of their current compositions on paper
and their laptops, including OpenMusic patches. We printed
several pages of their compositions printed on Anoto paper.
We also scanned the score of a composer and removed his
inked gestures, so we could explore how he would use
Musink to re-annotate his work. We gave the composers a
NOKIA digital pen, and, in the final four interviews, we
also brought a Livescribe® pen to observe how composers
would react to the audio feedback, and reflect on new pos-
sibilities that this pen would permit.
We encouraged composers to use the digital pen and Anoto
paper, and let them interact with paper as they normally did.
Our goal was not to enforce them to adapt to our approach,
but rather adapt our approach to their current work prac-
tices. We demonstrated the Musink Gesture Browser and
discussed the benefits and the constraints it posed.
Figure 8. Directions for faders on a music console
Gestures to control electronics during a performance:
a layout and semantic issue
MS raised a problem he faces, i.e. to control the levels of
faders on a mixing console during a performance. Cur-
rently, he draws these by hand. He demonstrated the poten-
tial use of Musink on a page of its own score together with
a folded Anoto paper (Figure 8). He drew directions and
linked them to the corresponding parts of the score. This led
him to reflect on how to balance layout and visualization
issues, including editing, performing, printing and publish-
ing perspectives. His reflections suggest that such design
decisions should be left flexible enough to accommodate
diverse types of use.
Composing with words
Figure 9 shows how PL would use interactive paper to
specify profile data in a patch for generating notes. The
composer uses the profile of the shape of letters in a word
to parameterize the generation of notes from a chord.
Figure 9. Words as musical parameters.
Left: The word “verre” is parameter for the Gesture Browser.
Right: OpenMusic patch: the shape of the letter P serves to
Figure 10. Representation and implementation of tremolos
Left: Gestures representing tremolos drawn on paper.
Right: OpenMusic patch that generates a sound from the en-
velope of a tremolo gesture.
Composing music though drawing and programming
Figure 10 shows a tremolo patch, controlled by an ampli-
tude curve. MM showed how he would specify this with a
gesture he had just drawn on paper.
MS encouraged us to create special Musink graph paper, so
he could draw extremely precise curves. These were not
sketches, but rather precise definitions used to generate
families of curves in OpenMusic, which could then be used
as parameters for a variety of functions (Figure 11a).
DC added his own gesture vocabulary, rather than inventing
a new one. He explained that OpenMusic did not meet his
needs, but started to explore new ideas when he realized
that he could use the Gesture Browser “as if” it were Open
Music (Figure 11b). He emphasized his need for “expres-
sive” gestures in his score, as opposed to “representative”
ones. He explored using them as drawings and suggested
that OpenMusic would now be useful at the analysis stage,
for example, to interpolate between drawing variants.
Musical notations as parameters for programming
MS explained that the electronic parts of a piece generally
need a semantic representation, such as a drawing of the
resulting sounds (Figure 11c). We compared a published
version of one of his earlier works  to his current nota-
tion. Today, he has to recopy these sounds manually, but he
wants to be able to move smoothly back and forth between
either form: electronic patches in OpenMusic and paper.
Figure 12. Thinking-aloud over a Livescribe notebook and a
score paper interface: exploring how to extend Musink with a
paper palette for gestures.
Brainstorming on interactive paper interfaces
We used a Livescribe® pen to explore with composers al-
ternative ways to interact with Musink. When we showed
the interactive pen and its audio feedback, MS quickly
sketched how to define gestures on paper rather than online,
creating a customizable paper panel of gestures, similar to
those in Livescribe notebooks (Figure 12).
Figure 11. (a) Linking the data points of a precisely drawn curve with OpenMusic. (b) Directions displayed on the Gesture Browser
on how to play the congo (percussion). Such directions could potentially become drawings (graphical data) for OpenMusic.
(c) Graphical representation of electronic parts in a published composition, Traiettoria.
Insights for the Composition Process
Role of Paper Drawings
We were surprised to learn that these composers empha-
sized drawing, rather than sketching, in the composition
process. Their disciplined gestures express concrete musical
ideas. We observed three types of drawings throughout the
• Directions to instruct either human performers or elec-
tronic equipment. Each has a distinctive graphical rep-
resentation and scope within the musical score.
• Symbolic representations of musical objects and ideas.
Several composers have two versions of each score:
performers see the notes; composers see their personal
representation of musical ideas.
• Graphs, rather than notes. Composers can specify mu-
sical input along various dimensions, such as nuances
in sound, rhythm, or instrument. Once on graph paper,
these drawings are translated into coordinates onto the
computer. This explains why they are more controlled
and precise than initial sketches: the composer is aware
of their meaning for future computer operations. By
drawing graphs on paper, composers add natural input
that mitigates the deterministic nature of computation.
Role of Computers
Composers viewed the computer as an important tool for
generating variation. Here, drawings and programs play a
complementary role, since tools like Musink and OpenMu-
sic let them program with their drawings. However, it is
important to emphasize that, while the computer may gen-
erate alternatives, it is always the composer who is respon-
sible for the final result.
Evolution of Musink
Extended Data Representations
We found that conventional forms of printed musical
scores, as supported by tools like OpenMusic, were unable
to support the rich data representations that composers
normally use. We extended Musink’s model to support rich
musical notation drawn on paper. This allows users to draw
arbitrary symbols along the timeline of a musical piece
(Figure 13). Users can interact with them using Musink
gestures as they do with printed symbols like notes and
rests. Even if not recognized, such symbols can be linked
with other parts of a musical score through their position in
the score’s timeline.
We also support the use of graphical gestures and parame-
ters, and graphs (Figure 14), treating them as a special type
of musical object. In Musink, graphical data are represented
through lists of x-y coordinates. This representation can be
transmitted to OpenMusic through Musink’s user interface.
OpenMusic provides advanced tools for editing graphical
data and processing them with other musical objects. Ac-
cordingly, Musink’s syntax was enhanced to enable interac-
tions with graph data and link it with other musical repre-
sentations. Users can use Musink to point to and select
graphical data, specify temporal ranges, and attach parame-
ters and function identifiers.
Mixed Paper Formats
In order to support the new data representations, we ex-
perimented with new paper formats, such as pages with
empty staffs, pages with simple parallel lines defining the
timeline of musical events (Figure 13), and graph paper for
drawing graphs (Figure 14). We also prototyped paper lay-
outs that mix multiple data representations, e.g., pages with
strips of graph paper lying over staffs with regular notation,
printed from the computer. These layouts and their online
models were created manually by the researchers although
several composers asked to be able to create their own.
Figure 13. A score created by MM during an interview is
shown on the Gesture Browser (notice that 5-line staffs have
been replaced by single-line timelines). Sophisticated musical
notation can stay unrecognized and coexist, through the
score’s timeline, with symbols defined as Musink gestures.
Figure 14. Support of graphical data in Musink’s Gesture
Browser. Scores can contain graphical parameters, and Mus-
ink gestures can be linked with detailed graphs. Musink’s syn-
tax has been extended to support functions over curves.
Differentiating between Different Roles of Gestures
The coexistence of different representations (notation,
graphs, directional Musink gestures) makes recognition
harder. We have explored how to differentiate between
them based on context-specific information, e.g., the size
and form of curved lines drawn by users or their position,
but this solution is not general. Other approaches [10, 13]
have addressed similar problems by introducing additional
writing modes. Switching between modes can be handled
by using a different pen, e.g., a pen with a different ink
color, pressing a pen button, or alternatively, ticking on
specialized paper areas. As reported earlier, we explored
with composers the potential of defining gestures by using
paper palettes and an interactive pen. Although this strategy
facilitates gesture recognition, switching modes has draw-
backs, and we are still exploring other alternatives.
We are interested in the role that paper plays in creativity
and how interactive paper technologies can support the
creative design process. We conducted a series of inter-
views and mini-workshops with music composers, which
revealed an astonishing variety of strategies for creating and
representing musical ideas. Each composer has a unique
creative process and a set of their own custom-made com-
puter tools, created with software like AudioSculpt and
manipulated with tools like Open Music. At a basic level,
we created Musink to link paper-based drawings to these
music composition tools.
However, our more ambitious goal was to create a testbed
for exploring creative design with interactive paper, giving
users complete control over the assignment of meaning to
their gestures. We began with a basic structure, the musical
staff, and a small set of gestures that Musink could recog-
nize without training. Musink provides the user with a
flexible set of scoping techniques for specifying groups of
musical objects, both with respect to each other and to the
underlying score. In addition, the user can draw any ges-
ture, which Musink will either recognize or ignore. At any
point in the future, the user can select a gesture and assign it
a meaning on the computer, either a custom-made function
or a pre-defined command. Musink tries to interpret similar
gestures in the way, but the user retains control over the
results. This notion of 'semi-structured delayed interpreta-
tion' gives users a powerful combination of freedom of ex-
pression when drawing musical ideas, while retaining the
possibility of adding computational power at any time.
Five composers donated examples of their personal compo-
sitions, which we printed on Anoto paper, and participated
in collaborative design sessions with us, to explore new
possibilities for Musink. During this period, Musink
evolved very quickly, to incorporate specific suggestions
from one composer to the next. Although each composer
was unique, tools to support one composer were often di-
rectly relevant to the others. We came to view Musink as an
interactive paper interface to Open Music and co-invented
new Musink functionality with the composers. For example,
we provided different paper structures (5-line staff, vari-
able-width graph paper, space devoted to drawing curves)
and began to explore how the print-edit-reprint cycle could
significantly affect the composition process.
We admit to a certain bias in our approach; after many
years of studying interactive paper in different domains, we
fully expected to find interesting uses of paper/computer
interaction. However, we were genuinely surprised by the
incredible diversity and cleverness these composers evi-
denced in their drawings, which we came to understand are
in fact, musical objects in their own right. Unlike in other
creative domains we have studied, including from architec-
ture to fashion design, these composers rarely 'sketch'. In-
stead, they make very precise drawings in which the physi-
cal characteristics of each gesture has meaning with respect
to the musical idea being expressed and the computation
that will result. This is an unusual user group, with skills as
end-user programmers, extreme creativity, and, at the same
time, highly disciplined and precise gestures. (Many of
these composers are musicians, which was evident as we
watched them move and make gestures in the air.)
Future work will assess Musink in real work environments.
Several composers have expressed interest in trying Musink
and we plan to follow at least one composer from the be-
ginning to the end of a new musical composition.
Our studies of composers have demonstrated the important
role that paper continues to play in the creative process,
even when composers are adept computer users. Paper re-
mains the optimal choice at the two extremes of the creative
process: drawing initial ideas for the first time and creating
an archival record of the finished work. Between these two
extremes, composers mix paper and computers in a variety
of ways, influenced by the particular input/output character-
istics of each.
We have explored how the use of Anoto technology ex-
pands the options available to composers in this interim
stage of creation and supports semi-structured delayed in-
terpretation. Musink allows composers to add innovative
notations to traditional music scores, a common theme in
contemporary music, a framework in which paper, printouts
and files can be linked together and support the process of
co-adaptation  in which users are encouraged to adapt
technology to meet their own unique needs.
We are grateful to Carlos Agon, Nicolas Donin, Aurélien
Tabard, as well as the composers and musical assistants
who participated in our interviews.
1. Abrams, S., Bellofatto, R., Fuhrer, R., Oppenheim, D.,
Wright, J., Boulanger, R., Leonard, N., Mash, D., Rend-
ish, M. & Smith, J. QSketcher: an environment for com-
posing music for film. In Proc. Creativity and Cognition,
ACM Press (2002), 157-164.
2. Agon, C., Assayag, G. & Bresson, J. OM Composer's
Book. 2006: Editions Delatour France, IRCAM.
3. Amitani, S. & Hori, K. Supporting musical composition
by externalizing the composer's mental space. In Proc.
Creativity and Cognition, ACM Press (2002), 165-172.
4. Arai, T., Aust, D. & Hudson S. PaperLink: a technique Download full-text
for hyperlinking from real paper to electronic content. In
Proc. CHI 1997, ACM Press (1997), 327-334.
5. Buxton, W., Reeves, W., Baecker, R. & Mezei, L. The
use of hierarchy and instance in a data structure for
computer music. Computer Music Journal, 1978. 2(4):
6. Chapuis, Y., Fober, D., Letz, S., Orlarey, Y. & Daudin,
C. Annotation de partitions musicales dynamiques. In
Proc. JIM (2007), 18-27.
7. Csíkszentmihalyi, M. Flow: the psychology of optimal
experience. 1990: New York: Harper and Row.
8. Donin, N., Goldszmidt, S. & Theureau, J. De Voi(rex) à
Apocalypsis, fragments d'une genèse. Exploration mul-
timédia du travail de composition de Philippe Leroux in
L'inouï revue de l'Ircam, no 2 2006.
9. Heiner, J., Hudson, S. & Tanaka, K. Linking and mes-
saging from real paper in the Paper PDA. In Proc. UIST
1999, ACM Press (1999), 179-186.
10. Hinckley, K., Baudisch, P., Ramos, G. & Guimbretière,
F. Design and analysis of delimiters for selection-action
pen gesture phrases in scriboli. In Proc. CHI 2005,
ACM Press (2005), 451-460.
11. Kurtenbach, G. & Buxton, W. The limits of expert per-
formance using hierarchic marking menus. In Proc. In-
terCHI 1993, ACM Press (1993), 482-487.
12. Letondal, C., Mackay, W. & Donin, N. Paperoles et
musique. In Proc. IHM 2007, ACM Press (2007), 167-
13. Liao, C., Guimbretière, F. & Hinckley, K. PapierCraft:
a command system for interactive paper. In Proc. UIST
2005, ACM Press (2005), 241-244.
14. Lieberman, H., Paternò, F. & Wulf, V. End user devel-
opment (Human-Computer Interaction Series). 2006:
15. Mackay, W. Augmented reality: Linking real and virtual
worlds: a new paradigm for interacting with computers.
In Proc. AVI 1998, ACM Press (1998), 13-21.
16. Mackay, W., Responding to cognitive overload: Co-
adaptation between users and technology. Intellectica,
2000. 30(1): p. 177-193.
17. Mackay, W. & Pagani, D. Video Mosaic: Laying out
time in a physical space. In Proc. Multimedia, ACM
Press (1994), 165-172.
18. Mackay, W., Fayard, A., Frobert, L. & Médini, L. Rein-
venting the familiar: Exploring an augmented reality de-
sign space for air traffic control. In Proc. CHI 1998,
ACM Press (1998), 558-565.
19. Mackay, W., Pothier, G., Letondal, C., Bøegh, K. &
Sørensen, E. The missing link: augmenting biology
laboratory notebooks. In Proc. UIST 2002, ACM Press
20. Rubine, D. Specifying gestures by example. In Proc.
SIGGRAPH 1991, ACM Press (1991), 329-337.
21. Sellen, A. & Harper, R. The Myth of the Paperless Of-
fice. 2003: MIT Press.
22. Shneiderman, B. Creating creativity: user interfaces for
supporting innovation. ACM TOCHI, 2000. 7(1): p.
23. Signer, B., Kurmann, U. & Norrie, M. iGesture: A Gen-
eral Gesture Recognition Framework. In Proc. ICDAR
24. Song, H., Guimbretière, F., Hu, C. & Lipson, H. Mod-
elCraft: capturing freehand annotations and edits on
physical 3D models. In Proc. UIST 2006, ACM Press
25. Stifelman, L., Arons, B. & Schmandt, C. The audio
notebook: paper and pen interaction with structured
speech. In Proc. CHI 2001, ACM Press (2001), 182-189.
26. Stroppa, M. Un orchestre synthétique: Remarques sur
une notation personnelle, in Le timbre: Métaphores pour
la composition. 1991, Bourgois, Paris. p. 485-538.
27. Tabard, A., Mackay, W. & Eastmond, E. From Individ-
ual to Collaborative: The Evolution of Prism, a Hybrid
Laboratory Notebook. In Proc. CSCW 2008, ACM
Press (2008), 569-578.
28. Vinet, H. & Delalande, F. Interfaces homme-machine et
création musicale. 1999: Hermes Sciences Public.
29. Weibel, N., Signer, B., Ponti, P. & Norrie, M. Paper-
Proof: A paper-digital proof-editing system. In Proc.
CoPADD (2007), 9-21.
30. Wellner, P. Interacting with paper on the DigitalDesk.
Communications of the ACM, 1993. 36(7): p. 87-96.
31. Winget, M. Heroic frogs save the bow: Performing mu-
sician's annotation and interaction behaviour with writ-
ten music. In Proc. ISMIR (2006), 73-78.
32. Wobbrock, J., Wilson, A. & Li, Y. Gestures without
libraries, toolkits or training: a $1 recognizer for user in-
terface prototypes. In Proc. UIST 2007, ACM Press
33. Wright, M. & Freed, A. Open Sound Control: A new
protocol for communicating with sound synthesizers. In
Proc. ICMC (1997), 101-104.
34. Yeh, R., Paepcke, A. & Klemmer, S. Iterative design
and evaluation of an event architecture for pen-and-
paper interfaces. In Proc. UIST 2008, ACM Press
35. Yeh, R., Liao, C., Klemmer, S., Guimbretière, F., Lee,
B., Kakaradov, B., Stamberger, J. & Paepcke, A. Butter-
flyNet: A mobile capture and access system for field bi-
ology research. In Proc. CHI 2006, ACM Press (2006),