Haskins Laboratories Status Report on Speech Research

1992, SR-1111112, 227-260

Diversity and Commonality in Music Performance: An

Analysis of Timing Microstructure in Schumann's
uTraumerei"*
Bruno H. Repp

This study attempts to characterize the temporal commonalities and differences among
distinguished pianists' interpretations of a well-known piece, Robert Schumann's
"Triiumerei." Intertone onset intervals (lOIs) were measured in 28 recorded performances.
These data were subjected to a variety of statistical analyses, including principal
components analysis of longer stretches of music and curve fitting to series of lOIs within
brief melodic gestures. Global timing patterns reflected the hierarchical grouping
structure of the composition, with pronounced ritardandi at the ends of major sections and
frequent expressive lengthening of accented tones within melodic gestures. Analysis of
local timing patterns, particularly of within-gesture ritardandi, revealed that they often
followed a parabolic timing function. The major variation in these patterns can be
modelled by families of parabolas with a single degree of freedom. The grouping structure,
which prescribes the location of major tempo changes, and the parabolic timing function,
which represents a natural manner of executing such changes, seem to be the two major
constraints under which pianists are operating. Within these constraints, there is room for
much individual variation, and there are always exceptions to the rules. The striking
individuality of two legendary pianists, Alfred Cortot and Vladimir Horowitz, is objectively
demonstrated here, as is the relative eccentricity of several other artists.

INTRODUCTION
A. Diversity and commonality in music
performance
More than at any earlier period in musical
history, the contemporary scene in serious music
is dominated by the performer (see Lipman, 1990).
Music consumers thrive on a limited repertoire of
standard masterworks, primarily from the 19th
century, that are offered again and again in
different performances, both in live concerts and
on recordings. The great musical events of our
time are not the premieres of new compositions,
but the appearances and reappearances of
superstar conductors and instrumentalists.
This research was made possible by the generosity of
Haskins Laboratories (Carol A. Fowler, President). Additional
support came from NIH (BRSG) grant RR05596 to the
Laboratories. I am grateful to the Record Library, School of
Music, Yale University, for providing most of the recordings
analyzed here, and to its manager, Karl Schrom, for his help.
Thanks also to Pat Shove for many stimulating discussions.

227

Young musicians compete for career opportunities

by entering competitions in which their performances are compared and evaluated by juries (see
Cline, 1985; Horowitz, 1990). While ever new renditions of the standard repertoire vie for the attention of record buyers and concert audiences, a
spate of reissues of historical recordings on CDs is
offering stiff competition. For some of the more
popular works, the Schwann catalogue lists
dozens of performances; if deleted records, available in libraries and private collections, are
counted, they may run into the hundreds. There
have never been such ample opportunities to compare different performances of the same music.
Obviously, this remarkable diversity reflects not
only clever marketing and the promotion of superstars, but also the ability of music lovers to distinguish and appreciate different performances.
The more sophisticated among these listeners detect qualities in the performances of individual
artists that lead them to search out these performers' concerts and recordings. They may voice

228

Repp

OpInIOnS about the quality of particular performances by these and other artists, describing
them as "brilliant" or "noble" or "thoughtful."
Some of the more gifted professional critics excel
in characterizing different performances in terms
alternatingly scholarly and poetic.
While the attention of listeners and critics thus
is mainly drawn to the differences among performances, there are also strong commonalities, usually taken for granted and hence unnoticed.
Though there is a large variety of acceptable performances of a given piece of music, there is an
even larger variety of unacceptable performances,
which rarely make their way into the concert halls
or onto records. Unless they have the mark of inspired iconoclasm (as do some of the performances
by the late Glenn Gould; see, e.g., Lipman, 1984),
they quickly succumb to the fierce competition of
the musical marketplace. Music teachers, however, have to deal with them every day and try
their best to mold immature and wayward students into performers that can be listened to with
pleasure. Though teachers may differ considerably
in their methods and goals, and are rarely very
explicit about what these are, they are transmitting the unwritten rules (though see Lussy, 1882)
of a performance tradition that goes back to 19th
century central Europe, where most of the
standard repertoire originated. Despite various
changes in performance practices during the last
200 years, most of them of a narrowly technical
nature, there are generally accepted norms of musical performance, according to which the artist's
actions are largely subordinated to the musical
structure. The artist's primary task is the expression of the musical structure, so it can be grasped
and appreciated by the listener, and make an impression on him or her. (See Lussy, 1882;
Riemann, 1884; Stein, 1962.) This is presumably
done by conventional means that are adapted to
the hearers' perceptual and cognitive abilities.
However, the particular doses in which these
techniques are applied (unconsciously, for the
most part) vary from artist to artist and account
for individual differences in interpretation.
Thus there are two basic aspects of music performance: a normative aspect (i.e., commonality)
that represents what is expected from a competent
performer and is largely shared by different
artists, and an individual aspect (i.e., diversity)
that differentiates performers. The individual aspect may be conceived of as deviations from a single ideal norm; more profitably, however, it may
be thought of as individual settings of free parameters in the definition of the normative behavior.

That way, it is possible for different artists to

meet the norm equally and yet be discriminably
different. For example, two pianists may be
equally adept in expressing a particular musical
structure in their performances, but one may
choose a slow tempo and large tempo changes,
whereas the other may prefer a faster tempo and
smaller deviations from the rhythmic beat. The
former artist may then be characterized as
"romantic" or "exuberant," while the latter will
evoke epithets such as "restrained" and "noble,"
though both may receive equal acclaim from a
sensitive audience.
The commonality-diversity distinction seems
obvious enough, but there is little tangible evi~
dence to substantiate it. Although volumes have
been written about different performers and their
characteristics, these discussions rarely go beyond
generalities, and the vocabulary used (such as the
adjectives quoted above) are not specifically linked
to particular performance properties; they also
may be used differently by different writers. There
is no consistent terminology, nor a well-developed
and generally accepted theory of performance description and evaluation, that would make possible an objective characterization of performance
commonalities and differences. Music criticism is
an art rather than a science, and the critic's impressions, accurate as they may be, are filtered
through an idiosyncratic web of personal experiences, expectations, preferences, and semantic associations. In fact, the belief is widespread that
performance differences cannot be characterized
objectively.
Such a negative conclusion, however, can only
be justified if objective performance analysis has
been attempted and has failed. A total failure is
highly unlikely, however, for there are many
physical properties of a musical performance that,
without any question, can be measured objectively. Whether objective analysis can capture a
performance exhaustively may remain in doubt;
the proper question is how much can be learned
from it. Surely, even a partial characterization in
terms of verifiable and replicable observations can
make a valuable contribution to our understanding of performance commonalities and differences, .
which remain mostly a mystery to this day. For
example, most music lovers would agree that
Artur Rubinstein and Vladimir Horowitz were two
very different pianists, whose performances of
similar repertoire (e.g., Chopin) are instantly distinguishable. But what is it, really, that makes
them so different and individual? And do they
have anything in common at all? It is all too easy

Diversitv and Commonalitv in Music Performance: An ,4na/vsis of Timmg Microstructure

to couch the answers to such questions in terms of

"artistic personality," which do not enlighten us at
all about the nature of the differences and commonalities. Objective performance analysis has a
contribution to make here.
B. Objective performance analysis

The concept of objective performance analysis

goes back to Carl Seashore (1936, 1938, 1947) and
his collaborators, who pioneered the use of
acoustic analysis techniques to derive
"performance scores" that show the exact
variations of pitch, timing, and intensity produced
by an artist on some instrument. A considerable
amount of data was collected in Seashore's
laboratory, but their analyses remained
rudimentary and focused primarily on technical
aspects such as pitch accuracy and vibrato in
singing and string playing, and chord synchrony
on the piano. Although some studies compared
different performances of the same music, no
statistical characterization of commonalities and
differences was attempted, nor were the results
interpreted with more than a passing reference to
musical structure. These limitations reflect the
behavioristic approach of American psychology at
the time, as well as the unavailability of
psychological performance models and advanced
statistical methods. Nevertheless, Seashore (1947,
p. 77) was able to reach conclusions that reinforce
the premises of the present study:
... there is a common stock of principles which
competent artists tend to observe; ... We should
not, of course, assume that there is only one way
of phrasing a given selection, but, even with such
freedom, two artists will reveal many common
principles of artistic deviation. Furthermore,
insofar as there are consistent differences in their
phrasing, these differences may reveal elements
of musical individuality.

Contemporary with Seashore's work, similar

research was going on in Germany. Hartmann
(1932) compared the timing patterns of two
famous pianists' performances of a Beethoven
sonata movement and provided detailed numerical
descriptions of the differences between them. The
small size of the sample, however, limits the
generality of his conclusions. Although one of the
two artists seemed quite eccentric, this impression
cannot be substantiated without more extensive
comparisons.
The three decades between roughly 1940 and
1970 were barren years for the objective study of
music performance. In the last two decades,

In

Schumann's "Triiumcrc;P

229

however, several researchers have taken up the

topic again. Their work, with few exceptions, has
taken a case study approach; their focus was not
on individual differences but on the instantiation
of certain principles in (hopefully, representative)
performance samples, mostly from pianists. Thus,
Shaffer (1980, 1981, 1984, 1989; Shaffer, Clarke,
& Todd, 1985) examined the timing of piano
performances from the perspective of motor
programming and control (see also Povel, 1977).
His erstwhile students, Eric Clarke and Neil
Todd, went on to make significant contributions of
their own. Clarke (1982, 1985) conducted case
studies of piano performances of Satie's music and
wrote at length on how musical structures are
expressed in timing variations (Clarke, 1983,
1988). Todd (1985, 1989; Shaffer & Todd, 1987)
developed a computational model of timing at the
phrase level, which has been revised and extended
to dynamics in his most recent publications (Todd,
1992a, 1992b). Bengtsson and Gabrielsson (1980;
Gabrielsson, 1974; Gabrielsson, Bengtsson, &
Gabrielsson, 1983) conducted extensive studies of
the performance of different rhythm patterns in
relatively simple musical contexts, but with
attention to individual differences. Gabrielsson
(1985, 1988) has written more generally about
timing in music performance. A provocative theory
of "composer's pulse" in performance timing has
been developed by Clynes (1983, 1987).
There are few studies in the literature that employed what one might consider a representative
sample of different performances. Gabrielsson
(1987) conducted a detailed comparison of five pianists' performances of the first eight measures of
Mozart's Piano Sonata in A major, K. 331; the
study included measurements of timing, intensities, and articulation (i.e., tone durations). Palmer
(1989) studied the same music as performed by six
pianists in "musical" and "unmusical" styles, and
she analyzed the timing patterns, note asynchronies, and articulation. Palmer also recorded
eight pianists playing the first 16 measures of a
Brahms Intermezzo and studied timing patterns
in relation to intended phrasing. She concluded
that "pianists share a common set of expressive
timing methods for translating musical intentions
into sounded performance" (p. 345). These studies
still used relatively small samples and obtained
only limited amounts of data from each performer.
The largest amount of performance data was
analyzed in Repp's (1990) study, which included
19 complete performances by famous pianists of a
Beethoven sonata movement. The analysis was
limited, however, in that it concerned primarily

230

Repp

reflect an adult's recollection of childhood

timing patterns at the level of quarter-note beats.
(Brendel, 1981; Chissell, 1987).
Also, the focus of the study was a search for
"Traumerei" occupies a central position in the
Clynes' (1983) elusive "Beethoven pulse" in the
"Kinderszenen" suite, not only by its location but
timing patterns. Following the example of
by its duration and structural role. It serves as a
Bengtsson and Gabrielsson (1980), Repp applied
resting and turning point in the cycle, which
principal components (factor) analysis to these
shows so many intricate thematic connections that
timing data, to determine how many independent
it may be considered a set of free variations (Reti,
timing patterns were instantiated in this sizeable
1951; Traub, 1981). Its key signature, meter, and
sample of expert performances. Two factors
motivic content also single it out as the hub of the
emerged, the first representing primarily phrasesuite (Brendel, 1981; Traub, 1981). However, it is
final lengthening, while the second factor captured
also often performed by itself, both by classical
other types of expressive timing variation. Musical
artists (e.g., it was one of Horowitz's favorite
listeners' evaluations of the performances were
encores) and in numerous popular versions and
also obtained, and some relationships between the
arrangements. Indeed, "Traumerei" is perhaps the
measured timing patterns and listeners' judgmost popular Romantic piano piece. Its score is
ments were found.
shown in Figure 1.
The present study continued the general apThe melodic/rhythmic structure (or grouping
proach taken by Repp (1990), but without the aim
structure; see Lerdahl & Jackendoff, 1983) of
of testing Clynes's theory of composer's pulse. Its
"Traumerei" is depicted schematically in Figure 2,
main purpose was to assemble a large sample of
which also introduces terminology to be used
performances of a particular composition by outthroughout this paper. The layout of the figure
standing artists, and to analyze the timing patcorresponds to that of the score in Figure 1, and
terns in detail, using various statistical methods.
bars (measures) are numbered. The piece is
These methods, it was hoped, would make it poscomposed of three 8-bar periods (A, B, A'), the first
sible to separate commonalities from differences.
of which is obligatorily repeated. Each period is
The common patterns would reveal how most pianists transmit musical structure and expression
subdivided into two 4-bar phrases, which are
represented by staff systems in Figure 1 and by
through timing variations, and it was of interest
large rectangular boxes in Figure 2. (Actually, the
to determine whether there would be a single
beginning and end of each phrase extend slightly
common factor or several. The characterization of
individual differences was also of interest, espebeyond the four bars, overlapping with the
preceding and following phrases, respectively.)
cially with respect to some of the legendary pianists in the sample, who are known for their
There are two phrase types, a and b, each of which
individuality.
recurs three times with slight variations
The music chosen for this investigation was a
(indicated by subscripts in Figure 2). Phrase al
(period A) is repeated literally in period A'.
well-known piano piece from the Romantic period,
Phrases b2 and b3, which constitute period B, are
"Traumerei" by Robert Schumann. It was selected
because it is a higWy expressive piece that permits
structurally identical but differ in key, harmony,
much freedom in performance parameters, and . and some other details.
hence much room for individual. differences in
The melodic events are divided horizontally
interpretation. Also, there are numerous record(roughly, along the dimension of relative pitch)
ings available.
among four registers or voices (S = soprano, A =
alto, T = tenor, B = bass). Vertically (along the
I. THE MUSIC
dimension of metrical distance), the events within
"Traumerei" ("Reverie," "Dreaming") is the
phrases are grouped into melodic gestures, which
seventh of the 13 short pieces that constitute
are represented by filled boxes in Figure 2.
Robert Schumann's (1810-1856) "Kinderszenen"
Characteristically, they extend across bar lines
("Scenes from Childhood"), op. 15. This little suite,
(vertical solid and dashed lines in Figure 2). A
melodic gesture (MG) is an expressive unit comuniversally considered one of the masterpieces in
its genre, was composed by Schumann in 1838
posed of at least two and rarely more than seven
when he was secretly engaged to Clara Wieck. The
successive tones. It is defined here to begin with
pieces were selected from some 30 pieces
the onset of its first tone and to end with the onset
composed for Clara around that time, and their
of its last tone. This seems reasonable, for a pititles may have been added as an afterthought.
anist cannot influence the time course or dynamThey are not intended for children but rather
ics of the last tone once the key has been struck.

Oiversitll and Commonalitll in lvluS1C Performance: An Anaillsis of Timing Microstructure in ScJlllmann '5 "Tniumerel"

----

;::C

tJ

:'-1

[t

~I

Fr'

Jl'

Ll

..

('jJ

.,

,-

<>

~.

.,"

)_.
I

~I

rn

r--..l-r

l.--'

...... ,

~I

-=1
1 I [

(espr.)

rTil

~.1:

c:

"'<

"it., I

, '"

liLLO/
b.. _.~.

m~
I.
i.

- #

----.--:

....::

~1

~~

",.

n-n

~d

~
(esp'.)

---fiYL

JJ IJll

~h~~~:

(J.'

I......!.-I

[ill

231

I.

~~l r

'L1....W
p.

L...J

",:~-,...

rr

I'$.

l;_

23

r.

~= I

1'=
I I

nl

f.'\

io o .

,":=!.
"

I
~a

Figure 1. The piano score of "Traumerei," created with MusicProse software following the Clara Schumann (Breitkopf
& Harlel) edition (with some deviations in minor details due to software limitations). The layout of the score on the
page is intended to highlight the structure of the music.

232

Repp

p s

p s
Melodic Gestures:

MG'

O[] MG2

D
[l]]] lJI]
Period

Phrase
Type

a,

MG2i
MG6 f

~~ MG3a
tzJ [23 MG4a
MGSa
~
MG6a
2

sa

p s
~~ MG3b
~!6d MG4b
~~ MGSb

[ill

o
S
A

T
B

MG6b

4
S
A
T
B

o
5

b,

T
B
S
A
T
B

S
A
T
B

20

a,

AI

S
A
T
B
S
A
T
B

Figure 2. Schematic representation of the melodiclrhythmic structure of "Traumerei." (See text for explanation.)

Oiversit1/ and Commonalit1/ in Music Performance: An Analllsis of Timmg Microstructure

(In Figure 2, each MG box extends one eighth-note

space beyond the metrical onset of the last tone.)
Blank spaces represent time spans devoid of MGs;
they may contain single tones, sustained tones, or
rests. Multivoiced chords having some gestural
quality are represented by vertical ellipses in
Figure 2. Arrows indicate continuity of a MG
across a line break or with a subsequent melodic
event. Thus, MG1 and MG6 continue from the end
of one line to the beginning of the next, and MG3b
and MG5b in the soprano voice are closely linked.
When an arrow points into blank space, it points
to the onset of a single-tone event that coheres
with the MG.
MGs are divided into primary (p) and secondary
(s) ones. The latter are usually shorter and
accompany primary MGs; they are represented by
boxes filled in lighter shades. An exception to this
classification is MG2i, a delayed imitation of MG2

in

Schumann's "Triiumerei"

233

divided between the tenor and alto voices (bars 10

and 14). In the following timing analyses, we will
essentially be concerned only with the primary
MGs, which represent the leading voice(s) in the
polyphonic quartet. As can be seen in Figure
2, in phrases of Type a the primary MGs
(including the final MG6f in bar 24) are all in the
soprano, except for MG6a which is in the bass and
overlaps both MG5a and MG 1. In Type b phrases,
during the second half, the primary MGs cascade
down through the four voices, overlapping each
other.

II. THE PERFORMANCES

The sample of performances analyzed here
includes 28 performances by 24 outstanding
pianists, selected according to ready availability
(see Table 1). Two artists, Cortot and Horowitz,
are represented by three different recordings each.

Table 1. The artists and their recordings.

Code

Artist

Recording

ARG
ARR
ASH
BRE
BUN
CAP
Cal

Martha Argerich (1941-)

Claudio Arrau (1903-1991)
Vladimir Ashkenazy (1937-)
Alfred Brendel (1931-)
Stanislav Bunina
Sylvia Capova a
Alfred Cortot (1877-1962)
Alfred Cortot
Alfred Cortot
Clifford Curzon (1907-1982)
Fanny Davies (1861-1934)
Jorg Demus (1928-)
Christoph Eschenbach (1940-)
Reine Gianoli (1915-1979)
Vladimir Horowitz (1904-1989)
Vladimir Horowitz
Vladimir Horowitz
Cyprien Katsaris (1951-)
Walter Klien (1928-1991)
Andre Krust a
Antonin Kubalek (1935-)
Benno Moiseiwitsch (1890-1963)
Elly Ney (1882-1968)
Guiomar Novaes (1895-1979)
Cristina Ortiz (1950-)
Artur Schnabel (1882-1951)
Howard Shelleya
Yakov Zak (1913-1976)

DG 410 653-2 (CD) [1983]

Philips 420-871-2 (CD) [1974]
London 421 290-2 (CD) [1987]
Philips 9500 964 [1980]
DG 427315-2 (CD) [1988]
Stradivari SMC-6020 (C) [T] [1987]
EMI3C 153-53793M [1935]
EMI3C 153-53794M [1947]
EMI3C 153-53795M [1953]
London LL-I009 [-1955]
Pearl GEMM CD 9291 (CD) [1929]b
MHS OR 400 [-1960s]
DG 2535 224 [1966]'
Ades 13.243-2 (CD) [1974]
RCA LD 7021 [T] [1947]
Columbia MS 6411 [1963]
Columbia MS 6765 [1965] (live) [T]
Telefunken 6.42479 AP [1980](live)[T]
Allegretto ACS 8023 (C) [T]C
MHS 1009 (orig. Erato) [-1960s]
Dorian DOR-90116 (CD) [1988]
Decca DL 710,048 [-1950s]
Electrola WDLP 561 [-1935]
Vox PL 11.l60 (orig. PL8540) [1954]
MCA Classics MCAC-25234 (C) [1988]
PatM COLH 85 [1947]
CHAN 8814 (CD) [1990]
Monitor MC 2039 [-1960]

C02
C03
CUR
DAV
OEM
ESC
GIA
Hal
H02
H03
KAT

KLI
KRU
KUB
Mal
NEY
NOV

aRT
SCH
SHE
ZAK

abirthdate not known

boriginally Pearl CLA 1000
Cdate unknown
Abbreviations: CD =compact disc; C =cassette; < =date of liner notes (recording date probably the same or preceding year); estimated date; T = "Traumerei" only

Repp

234

Most of the recordings (17) are on long-playing

records, 3 are on cassettes, and 8 are on CDs
(including a transfer of a 1929 recording by Fanny
Davies, a one-time student of Clara Schumann,
from a very scratchy original). Most of the
performances are of the complete "Kinderszenen,"
but five are of "Traumerei" only. Two of the latter
are from live concerts; all others are studio
recordings. Table 1 includes actual or estimated
recording dates.
The 24 artists include some of the most
renowned pianists of this century as well as some
less well-known artists. They can be grouped
according to gender (6 female, 18 male), country of
origin (5 from Russia; 4 each from Austria and
France; 3 from England; 2 each from Germany
and Brazil; one each from Czechoslovakia,
Argentina, and Chile; one unknown), and
approximate age at the time of recording (about
equal numbers of young, middle-aged, and old). At
least 12 pianists are no longer alive.

III. ANALYSIS METHODS

A. Measurement procedure
All tone interonset intervals (lOIs) were handmeasured by the author using a waveform editing
program. Each performance was low-pass filtered
at 4.9 kHz and digitized at a 10 kHz sampling
rate. The digitized waveform was displayed on the
screen of a computer terminal, 2 s at a time. The
screen resolution for that display was about 2 ms.
A cursor was placed at each tone onset, and a
permanent "label" was attached to that point in
the waveform file. The differences between the
time points of successive labels yielded the lOIs,
which were noted down to the nearest millisecond.
After some practice, it took about 2 hours to
measure one complete performance.
Two problems had to be coped with. One was
that the onset of a soft tone was often difficult to
detect by eye, particularly in some of the older
recordings, which had much surface noise. (DAV
was the worst by far.) An auditory method was
used in that case: The cursor was moved back in
small steps, and the waveform segment up to the
cursor was played back until the onset of the tone
in question could no longer be heard. This procedure was time-consuming but usually resulted in
rather accurate location of tone onsets (cf.
Gabrielsson, 1987). The second problem concerned
onset asynchronies among simultaneous tones in
different voices. Unintended asynchronies, which
usually were too small to be detected by eye in the
waveform, had to be ignored. Larger asynchronies,
which in most cases must have been intended by

the artists, were noted (especially in performances

by COl-3, DAV, MOl, and SHE) and measured,
but the label from which the 101 was computed
was placed at the onset of the major melody tone
(usually the one with the highest pitch and the
latest onset). The same convention was followed in
the case of the written-out arpeggi in bars 2, 6,
and 18.
B. Data representation
The data were submitted to further analysis in
the form of eighth-note lOIs. That is, intervals
involving grace note onsets (in bars 2, 6, 8, 16, and
18) were omitted from the data and were
considered separately.l lOIs longer than a
nominal eighth-note were divided into eighth-note
intervals of equal duration. Thus, for example, the
half-note 101 in bar 2 was represented as four
equal eighth-note lOIs. A complete performance
thus yielded 254 eighth-note lOIs.

e. Measurement error
The measurement procedure would have been
prohibitively time-consuming if maximum accuracy had been aimed for (e.g., by displaying
shorter waveform segments on the computer terminal screen). A certain amount of speed-accuracy
tradeoff had to be taken into account. The magnitude of the resulting measurement error can be
estimated from three performances (BRE, C02,
and the first half of CUR) that, for various reasons, had to be remeasured. Two of them (BRE,
CUR) derived from good LPs with a moderate
amount of surface noise, whereas C02 was of
older vintage and had special problems related to
the pianist's tendency to play the left- and righthand parts asynchronously. There were several
very large discrepancies (in excess of 80 ms)
between the two measurements of the C02
performance, due to a conscious change in the
author's criteria for treating asynchronies; these
discrepancies were not considered true
measurement errors and were omitted from the
data presented below, though some smaller
inconsistencies of the same kind may still be
included.
The measurement error distributions are shown
in Table 2. The combined BRE+CUR data
represent average accuracy, whereas the C02
data constitute a worst-case scenario; only DAV
was even more difficult to measure. It can be seen
that over 90% of the measurement errors in the
BRE+CUR data did not exceed 10 ms, which is
less than 2% of the average 101. The largest error
was under 35 ms, or about 6% ofthe.average 101,
and the average measurement ,error was 4.3 mS,

DIversItv and Commonalltv In MusIc Performance: An Anail/SIS ofTImll1R MIcrostructure

less than 1%. In the C02 data set, about 90% of

the errors ranged from 0 to 20 ms, and the largest
error was under 50 ms. The average error was
10.4 ms, or about 2%. Measurements of the
virtually noiseless CD recordings presumably
were even more accurate than suggested by the
BRE+CUR data.
Table 2. Distributions of absolute measurement error
in two sets of remeasured data.
BRE+CUR (N=412)
Range (ms)
0-5
6 - 10
11 - 15
16 - 20
21 - 25
26 - 30
31 - 35
36 -40
41 - 45
46 - 50

Percent
77.4
13.8
2.7
1.7
2.2
1.7
0.5

Cum.
77.4
91.3
93.9
95.6
97.8
99.5
100.0

C02 (N=245)
Percent
24.5
40.0
18.8
5.7
3.7
3.3
2.0
1.2
0.4
0.4

Cum.
24.5
64.5
83.3
89.0
92.7
95.9
98.0
99.2
99.6
100.0

IV. RESULTS AND DISCUSSION

The strategy in presenting the results will be to
proceed from more global properties to more
detailed aspects.

A. Overall tempo
Perhaps the most obvious dimension along
which different performances vary is that of
overall tempo. The present set of performances is
no exception. Tentative estimates of global tempo
were obtained by computing the first quartiles
(the 25% point) of the individual eighth-note 101
distributions, multiplying these millisecond values
by 2, and dividing them into 60,000. The choice of
the first quartile was motivated by the
consideration that expressive lengthening of lOIs
is both more frequent and more pronounced than
shortening, and also by the fact that it gave tempi
for two pianists, BRE and DAV, that agreed
closely, respectively, with Brendel's (1981)
statement of his preferred tempo and with Clara
Schumann's (DAV's teacher's) recommended
tempo. (Repp, 1993b, will provide a more extended
discussion.) Table 3 shows that the tempo range
extended from 48 to 79 quarter-notes per minute.
Apart from the fact that the three fastest
performances are all old recordings, there does not
seem to be any systematic relationship between
tempo and the time at which the recording was

l1J

Schumann's "TriiumereI"

235

made, nor with pianists' gender, age at the time of

recording, or country of origin.
Table 3. Overall tempi (qpm) estimated from the first
quartile of the fOf distribution.
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79

ESC
KAT, NEY
ZAK
CAP
KLI
CUR
BUN, NOY
DEM, KUB
ARR, KRU, MOl, SCH
H02
ARG
ASH
H03
HOI, SHE
C03
BRE,ORT
GIA

COl

C02

DAY

B. Repeats
The first step in the data analysis was an
attempt to eliminate redundancy and reduce
random measurement error by averaging across
repetitions of the same (or highly similar) musical
material. Of course, such averaging is meaningful
only if there are no systematic differences in
timing microstructure across repeats. The prime
target for averaging was the first period, bars 1-8,
which was repeated literally, according to
Schumann's instructions in the score. All but two
pianists (DAV, KRU) observed this repeat. 2
To compare the first and second repeats for all
pianists, a grand average timing pattern was first
obtained by computing the geometric mean of
corresponding lOIs across all 28 performances.
(For DAV and KRU, the data of bars 1-8 were

Repp

236

simply duplicated for the missing second repeat.)

The geometric mean was preferred over the
arithmetic mean because it compensated for any
tendency of slower performances to show more
expressive variability, which would have
dominated in an arithmetic average. In this grand
average, the two repeats of bars 1-8 were found to
have extremely similar timing profiles, with only a
slight tendency for the second repeat to be played
slower. The correlation between these two
averages was 0.987, which indicates that any
variations across repeats for individual pianists
were either random or idiosyncratic. The
correlations between the two repeats in individual
performances are shown in Table 4 (column 2);
most of them were quite high.
The close similarity of timing patterns across repeats has been noted in virtually every study in

which such a comparison was made (see, e.g.,

Palmer, 1989; Repp, 1990; Seashore, 1938). Thus,
it seemed justified to average across the two repeats of bars 1-8 in all further analyses, except as
noted.
There were other instances of identical or highly
similar musical material in the piece. Their timing
patterns may be compared in Figure 3, which
plots the grand average eighth-note lOIs. A
logarithmic ordinate scale is used here to
accommodate the longest values; it also makes the
scale comparable to the percentage scale used by
authors such as Gabrielsson (1987). The abscissa
represents metrical distance (or "score time") in
eighth-note steps.3 Longer notes appear as
"plateaus" of multiple eighth-notes; that way, all
lOIs are represented on the same proportional
scale.

Table 4. Correlations of the timing profiles for repeated or similar sections (R = repeat; GM = geometric mean).
Bars
Artist

1-8/1R-8R

ARG
ARR
ASH
BRE
BUN
CAP
COl
CO2

0.781
0.933
0.922
0.953
0.510
0.935
0.859
0.865
0.750
0.937
*
0.938
0.890
0.777
0.918
0.811
0.826
0.825
0.804
*
0.951
0.805
0.920**
0.931
0.677
0.676
0.868
0.839
0.987

cm
CUR
DAY
DEM
ESC
GIA
HOI
H02
H03
KAT
KLI
KRU
KUB
MOl
NEY
NOY
ORT
SCH
SHE
ZAK
GM
*no repeat
**bars 1-4 only (see Footnote 2)

1-4117-20

0.861
0.938
0.951
0.888
0.633
0.922
0.717
0.736
0.741
0.911
0.863
0.924
0.734
0.809
0.958
0.861
0.738
0.901
0.893
0.875
0.913
0.855
0.904
0.906
0.609
0.574
0.831
0.875
0.986

9-12/13-16
0.630
0.850
0.927
0.880
0.632
0.846
0.837
0.851
0.571
0.921
0.863
0.828
0.802
0.914
0.697
0.822
0.770
0.867
0.861
0.940
0.888
0.493
0.872
0.908
0.707
0.748
0.859
0.909
0.950

DlVersitu and Commonalitu

2500

.........

2000

Ul

E
-.J

<t:

In

Music Performance: An Analysis of Timing Microstructure in Schumann

--

a, (bars 04)

--

a2 (bars 21-24)

-e- a, (bars 17-20)

>
a:

--

"Triiumerei"

23;""

b 1 (bars 58)

-e- bz (bars 9-12)

--

b3 (bars 1316)

f-

1000

fW

~ 800

MG5b

MG3a

r-l

MG4b

ITl

a:

MG3b

z 600
f-

~-M~~~M~~~M~~~M~~

6~~~~NNN~~M~~~~~~

BAR/EIGHTH-NOTE

BAR/EIGHTH-NOTE

Figure 3. Grand average lOIs (geometric means across all 28 performances). The two panels show the timing profiles
for phrases of Type a and b, respectively. Primary melodic gestures (d. Figure 2) are indicated by brackets. The bar
numbers on the abscissa refer to bars 0-8 only; for the later phrases, an appropriate constant must be added.

The left-hand panel of Figure 3 compares the

timing profiles for the three Type a phrases. Bars
17-20 (phrase al) are musically identicalwith bars
1-4 (see Figure 1), and it can be seen that their
timing patterns are highly similar (r = 0.986),
though bars 17-20 tend to be played at a
somewhat slower tempo. The individual
correlations for these four bars (Table 4, column 3)
are comparable to the correlations between the
two repeats of bars 1-8. The timing pattern for
bars 21-24 (phrase a2) is initially similar but
diverges soon, due to the fermata (long hold) in
bar 22 and the progressive ritardando (slowing of
tempo) towards the end of the piece.
The right-hand panel of Figure 3 compares the
timing patterns of the three Type b phrases. Once
again a striking similarity can be seen, due to the
similarity or identity of the melodic structure (see

Figure 2). The timing profiles of phrases b2 (bars

9-12) and b3 (bars'13-16) are especially close. Only
the (prescribed) ritardando in bar 16 is much
more pronounced than that in bar 12 (which is not
notated). If the final halves of bars 12 and 16 (the
ritardandi) are excluded, correlations between
these two timing patterns are again extremely
high (see Table 4, last column). Phrase bl (bars 58) also shows a rather similar profile; differences
occur precisely where its musical content deviates
from phrases b2 and b3 (especially in bar 6).
The correlations in Table 4 indicate which
artists were highly consistent in their timing
patterns, and which of them were less consistent.
The consistent group is led by ASH, CUR, and
NOV and includes many others; the less
consistent group includes ARG, BUN, C03, MOl,
ORT, and SCH.4

Repp

238

C. Melodidrhythmic structure and the

average timing pattern

Let us examine now the average timing pattern
and its relation to the musical structure in greater
detail. Although the grand average timing profile
is not necessarily an optimal performance timing
pattern, it captures. features that many individual
performances have in common, whereas it suppresses random and idiosyncratic timing deviations that differ from performance to performance.
In Figure 3 it can be seen that there are global
temporal trends within phrases, particularly those
of Type b. Their timing profiles follow a concave
function on which various local peaks are
superimposed. Bars 21-24 (phrase a2), too, seem
to follow such a curve, though it is grossly
distorted by the fermata and the final grand
ritardando. Bars 1-4 and 17-20 (phrase al) have a
flatter global timing shape. The curvilinear trends
in Type b phrases are reminiscent of the parabolic
curves hypothesized by Todd (1985) as the basic
phrasal timing pattern. Todd's model 5 , however,
operates on the durations of time units equivalent
to whole bars, whereas the present timing profiles
are across eighth-note units and thus provide
much finer resolution. The curves that "cradle" the
present timing patterns (i.e., the "bottom lines" of
the Type b timing profiles) are, in fact, poorly fit
by quadratic functions. However, they do
represent the general principle formalized by
Todd and observed by many others, that there is a
slowing of the tempo at major structural
boundaries, in proportion to the importance of the
boundaries. Thus the most extreme ritardando
(prescribed in the score) occurs at the end of the
piece; a pronounced slowing down (also
prescribed) occurs at the end of the second period
(bar 16); next come the end of the first period (bar
8) and the end of phrase bl (bar 12), which show
about the same amount of ritardando (not explicit
in the score); the end of phrase al (bars 4 and 20)
shows the smallest, but still quite noticeable
slowing down.
The piece begins with a brief melodic gesture
(MG1) that ascends from the dominant to the
tonic, with accent on the latter (hence the
intervening bar line, which indicates accent on the
following beat). MGl recurs in bars 4-5, 8-9,12-13,
16-17, and 20-21, though the upbeat is shortened
in bars 8, 12, and 16 (cf. Figure 1). It thus
contains a single 101 of variable nominal length.
Its realization was strongly context-dependent:
The quarter-note upbeat at the beginning of the
piece (bar 0) was short relative to the following

101, which seemed fairly stable across all phrases.

The quarter-note upbeats at the ends of bars 4
and 20, which coincided with MG6a, were
relatively longer. The eight-note upbeats in bars 8
and 12, which coincided with MG6b, were even
longer. Finally, although the grace-note upbeat of
bar 16 is not shown explicitly in Figure 3, the 101
accommodating it was lengthened enormously,
due to the major ritardando at the end of period
B. (More information about the grace-note upbeat
is provided later on.)
The chord following MG1 enriches the harmonic
and rhythmic texture but is not really part of the
melody; however, it may be thought of as a kind of
echo of the tonic and thus could be linked with
MG 1. The chord bisects the inter-gesture interval
between MGl and MG2. Of the resulting two lOIs,
the first (nominally shorter) one was consistently
longer relative to the second (nominally longer)
one, as if pianists tried to equalize the two. This
may reflect final (and/or accent-related)
lengthening of the tonic in MG1, which is
absorbed by the 101 preceding the chord.
MG2 is undoubtedly the most salient melodic
gesture of the piece. It comprises five eighth-notes
that ascend in increasingly larger pitch steps to a
final note which constitutes the apex of the fourbar melodic arch that constitutes each phrase.
Variants ofthe gesture occur in bars 1-2, 5-6, 9-10,
13-14, 17-18, and 21-22. MG2 is unusual because
it culminates on the second beat of a bar, negating
entirely the customary accent on the first beat. As
can be seen in Figure 3, the average temporal
pattern of the five lOIs is quite similar in all
occurrences of the gesture: The first 101 is close to
the proportional duration of the preceding 101, but
the next two lOIs are much shorter. The fourth
101 is longer again, similar to the first, and the
fifth 101 is the longest; it is clearly visible as a
. sharp peak in the timing profile. The duration of
that fifth 101 varies with position in the piece: It is
least extended in bars 10 and 14, longer in bars 2,
6, and 18, where it accommodates the two grace
notes of the left-hand chord (cf. Figure 1), and
longest in bar 22 where it leads to the climactic
fermata. In fact, this eighth-note 101 in bar 22
tended to be lengthened relative to the fermata
chord itself (the plateau following the peak). The
specific accelerando-ritardando shape of MG2 will
be examined more closely later on.
The 101 following MG2 in bars 2, 6, 18, and 22
(plateau in Figure 3) represents another inter-gesture interval, nominally three or four eighth-notes
long. Relative to the' last 101 of MG2, this interval
was much shorter proprortionally, even in bar 22,

Diversitll and Commonalitll in Music Performance: An Ana/lIsls of Timln\; MIcrostructure

where it received the fermata. The fermata had its

maximal effect on the inter-gesture interval, doubling its duration relative to bars 2 and 18, but it
also affected the whole preceding MG2. In bars 10
and 14, the inter-gesture IOI is bridged by the inner-voice imitation gesture MG2i, which has a
timing pattern that is almost the mirror image of
MG2 in that it accelerates during the first three
IOIs and slows down only at the end. The first IOI
is long because it also represents the last 101 of
MG2; the relative length of the second IOI may be
transitional, or it may reflect final lengthening of
MG2 which was absorbed by MG2i. 6
The remaining MGs, which descend in a chain
from the melodic apex reached by MG2, are
patterned differently in phrase Types a and b. The
grouping structure of the Type a chain is indicated
with slurs in the score (Figure 1). MG3a consists
of four (a1) or five (a2) eighth-notes in the soprano
voice, with accent on the penultimate; a shorter
accompanying gesture occurs in the tenor voice
(al only). It can be seen in Figure 3 that
lengthening on the accented note occurred in
performance. MG4a is similar in rhythmic
structure to MG3a and exhibits a similar timing
pattern, except that the final unstressed eighthnote 101 was lengthened as well, especially in
phrase a2, where the final ritardando tilted the
timing profile. MG5a, reinforced by shorter
secondary MGs in the bass voice, is similar to
MG4a, except that in phrase al it leads into a
final half-note that coincides with the onset of
MG6a. Its timing pattern is a progressive
lengthening, similar to that observed in MG4a. In
phrase a2, the final ritardando is in full control.
The final gesture in phrase aI, MG6a, is in the
bass voice and leads back to the low-pitched tonic
which doubles the tonic of the next MG 1. Its first
IOI coincides with the final lengthened IOI of
MG5a, and the last two lOIs coincide with the
quarter-note upbeat of MG1 and are both
lengthened as well. The lOIs in between are
considerably shorter. The total timing profile of
MG6a is not unlike that of MG2, which it
resembles rhythmically. The final gesture in
phrase a2, MG6f, is a truncated version of the
preceding MGs and shows an enormous slowdown
in tempo; note, however, the discontinuity in the
ritardando between MG5a and MG6f, which is a
manifestation of final lengthening in MG5a.
The melodic chain of type b is also divided into
four MGs, but they are organized differently. The
slurs in the score are not entirely consistent here;
what distinguishes the gestures is the fact that
they occur in different voices, imitating and

In

Schumann's "Triiumerei"

239

overlapping each other. Simultaneously, other

voices articulate shorter two-note cadential
gestures which, as can be seen in Figure 3,
essentially govern the timing pattern. In each
MG, both lOIs constituting these accompanying
motifs were lengthened, the first, unaccented one
usually more than the second, accented one. This
is explained by the harmonic function of the tone
cluster initiating the unaccented 101, which is
that of suspense followed by resolution. In part,
however, it may also be due to final lengthening of
the preceding, overlapping MG.
In summary, these observations illustrate
several general principles of performance timing:
(1) Whole phrases tend to show global tempo
curves characterized by an initial accelerando and
a more pronounced final ritardando whose degree
reflects the degree of finality of the phrase in the
hierarchical grouping structure. (2) "Riding" on
the global pattern, individual MGs often show a
similarly curved local pattern, though accent
placement, harmonic factors, and the influence of
overlapping MGs in other voices may modulate
that pattern in various ways. Gesture-final
lengthening is commonly observed.

D. Principal components analysis

An alternative way of capturing commonalities
among different performances is offered by the
statistical technique of principal components
(factor) analysis. This method decomposes the
data matrix (N performances by M lOIs) into a
number of independent components or factors,
each of which resembles a timing profile (i.e., M
standardized "factor scores"). The original data
are approximated by a weighted sum of these factors; the weights, which differ for each performance, are called "factor loadings" and represent
the correlations between the performance timing
profiles and the factor score profiles. The degree of
approximation (the percentage of the "variance accounted for" or VAF) depends on the number of
factors that are considered significant; a common
criterion employed here is that they should have
"eigenvalues" greater than 1. (Eigenvalue = N
times an individual factor's proportion of VAF.)
The first factor always accounts for the largest
amount of variance and represents a kind of central tendency or most common pattern. Additional
factors account for increasingly less variance and
thus represent patterns shared by fewer performances. A standard technique for simplifying the
factor loadings and thereby increasing the interpretability of the factors is called "Varimax rotation." It increases large factor loadings and re-

Repp

240

duces small ones without changing the number of

factors or the total VAF; however, the VAF is redistributed among the rotated factors.
Principal components analysis thus can reveal
whether there is more than one shared timing
pattern represented in the sample of 28
performances. If there is essentially only one way
of performing the piece, then a single factor
should explain most of the variance in the data. If
there are several radically different timing
patterns, then several orthogonal factors should
emerge. Idiosyncratic timing patterns will not
lead to a significant factor and will constitute part
of the variance not accounted for. Since all timing
profiles are standardized and intercorrelated in
the course of the analysis, differences in overall
tempo are automatically disregarded.
Although it would seem desirable to conduct the
analysis on the complete performances, the large
ritardandi observed universally at the ends of
phrases a2, bI, b2, and b3, as well as the large
slowdown at the fermata in phrase a2, dominate

the overall timing pattern and cause high

correlations among the performances, with the
result that the interesting variation among the
shorter lOIs is lost. In fact, a principal
components analysis conducted on the complete
data yielded only a single factor, indicating that
all pianists observed the major ritardandi, which
is not very surprising. Therefore, it was decided to
analyze only the data for bars 0-8 (phrases al and
bI), which did not include any extreme
lengthenings (cf. Figure 3). Because of the
parallelism of the timing profiles of these phrases
to those of similar type (see Figure 3), such a
restricted analysis essentially captures the whole
performances minus the major ritardandi.
This analysis yielded four significant factors.
Together they accounted for 76% of the variance.
For individual performances, the VAF ranged
from 60% to 90%. The timing profiles representing
the first three factors are shown in Figure 4, and
the factor loadings of the 28 performances are
listed in Table 5.

1OOO-rr---r---,--r--:-::!-:--~-.....---r---,

<il lOOO -rr---,c---;---,---,---.,.--r---,-..,--,

.s

OIJ

CJ)

a:

CJ)

a:

otD

..

~ 400 -hji-,TTM-rrrrrn-TlTl+n-rrrr-.+nlTTTrmTTrrrr+TTrITTT+-rnT1'Trh-rTTTn-I
1000rr--,---,---,----,----,---,r---,...-----,

400,

....

~-M~~-M~~_M~~_M~~-M~~_M~~_M~~_M~_

O~~~~NNNNMMMM.~.~~~~~~~~~~~~~~~=~

BARS/EIGHTH-NOTES

Figure 4. Timing patterns of the first three principal components for bars 0-8. The standardized factor scores were
converted into milliseconds by multiplying them with the average within-performance 101 standard deviation and
adding this product to the grand mean 101. Bar lines and MG brackets are added for orientation.

Diversitu and Commonality in Music Performance: An Analuszs of Timzng Mzcrostnlcture

Table 5. Sorted rotated factor loadings of the 28

performances, bars 0-8. (Loadings smaller than 0.4 are
omitted.)
II
SCH
ASH
BRE
DAV
CAP
SHE
KAT
KUB
ARR
ZAK
aRT
CUR
KRU
DEM
HOI
H03
H02
ARG
NEY
ESC
GIA
BUN
KLI
NOV
C03
Cal
CO2
Mal

0.844
0.760
0.738
0.695
0.695
0.692
0.685
0.672
0.664
0.644
0.637
0.615
0.593
0.551

0.482

0.523
0.509

III

IV

0.423

0.438

0.41 I

0.463
0.460
0.429
0.495
0.495
0.888
0.886
0.837
0.770
0.665
0.626
0.595
0.588
0.570
0.549

0.445

0.524
0.491

0.464
0.876
0.875
0.850

0.501

0.582

Factor I (VAF = 30%) represents a timing

pattern shared (partially) by a large number of
performances; half the pianists had their largest
loading on this factor, with SCH, ASH, and BRE
. leading the group. This pattern has the following
features: a relatively long initial upbeat (MGl)
and pronounced lengthening of the last 101 of
MG6, which accompanies the upbeat to the next
MGl (positions 4-8 and 8-8); general lengthening
of long lOIs (plateaus in the profile); a relatively
long accelerando phase in MG2 followed by a
pronounced lengthening of its last 101; and small
but regular gestural accent peaks for MG3-MG6,
except for special emphasis at the end of MG5b
(positions 8-3 to 8-5). This pattern reflects the
melodic/rhythmic structure even more clearly
than the average timing profile displayed in
Figure 3.
Factor II (VAF = 25%) may be called the
"Horowitz factor," since the three performances by
that artist showed the highest loadings, though

In

Schumann', "Triiumerez"

241

ARG, NEY, ESC and a number of other

performances shared features of this pattern. Due
to the statistical orthogonality of the factors, this
timing pattern is radically different from that of
Factor I. It is characterized by a very short initial
upbeat (essentially reduced to an eighth-note,
though it was rarely that extreme in the actual
performances; see below); relatively short long
lOIs (i.e., low plateaus); no accelerando but a
pronounced ritardando during MG2; pronounced
lengthening in MG3a, MG3b, and MG5b, with
extreme prolongation of the IOI in position 8-4,
which accommodates a grace note.
Factor III (VAF = 15%) is the "Cortot factor"; the
three performances by this artist were the only
ones that showed substantial loadings. Its timing
pattern is most unusual: a relatively short upbeat
in MGl; a short final lOI in MG2 (positions 2-2
and 6-2); marked speeding up during MG3 with a
following ritardando extending through MG4 to
the endofMG5; and rushing during the end of
MG5b, followed by a pronounced ritardando
during MG6b.
Factor N (VAF =6%) seems less important and
will not be discussed in detail, as no artist showed
a high loading on it (Table 5).
The three factor timing patterns shown in
Figure 4 do not correspond exactly to any real performance, though they are similar to certain performances, as indicated by the factor loadings
(Table 5). Most artists' performances are best described by a weighted combination of several factors; thus, for example, the CAP timing profile is a
combination of Factors I, II, and III. Such a combination will of course result in an attenuation of
extreme features. About one fourth of the variance
was not explained by the four factors extracted,
and most of that variance was probahly nonrandom, representing the artists' individuality.
It had to be asked at this point whether the
Horowitz and Cortot factors emerged only because
each of these artists was represented by three
different performances. 7 The principal components analysis was repeated with only a single
performance of each of these pianists included
(COl and H02). The three factors that emerged
(VAF = 71%) were highly similar to the first three
factors of the earlier analysis. The second and
third factors still had their highest loadings in
H02 and COl, respectively. The patterns offactor
loadings for the three factors in the two analyses
correlated 0.96, 0.98, and 0.95, respectively. Thus,
the Horowitz and Cortot factors are not artifacts
due to the over-representation of these artists in
the sample. They represent two true alternatives

242

Repp

to the "standard" pattern of performance timing

instantiated by Factor I-alternatives that only
Horowitz and Cortot dared to choose in nearly
pure form, but that several other pianists incorporated partially into their timing strategies.
E. Detailed analyses
In this section, we examine how melodic
gestures and other details of the score were
executed by using, as it were, a magnifying glass
on the data. While the emphasis has so far been
primarily on commonalities, the analyses are now
directed increasingly towards uncovering artistic
diversity. They reveal the different ways in which
a gesture may be shaped by pianists of great
authority. They also reveal some unusual, and
occasionally questionable, interpretations of
notational details in the score. In addition to
demonstrating the range of individual differences,
these analyses illustrate what temporal patterns
are preferred by a majority of the artists, and
what patterns are never used, and hence
presumably unacceptable.
1. MG 1. This gesture consists of merely two
tones forming the pitch interval of an ascending
fourth. It is followed, however, by a chord whose
timing, though not strictly part of the melody, is
also of interest. MG1 appears six times: Three
times with a nominal quarter-note upbeat (bars 0,
4, and 20) and three times with a nominal eighthnote upbeat (bars 8, 12, 16), one of which (bar 16)
is written as a grace note and thus is open to the
assignment of an even shorter value. Two of the
instances of the gesture are relatively unconstrained because the upbeat is unaccompanied
(bars 0 and 16); in the other cases the upbeat coincides with eighth-notes in the bass voice (MG6).
Figure 5 presents the individual data in terms of
two 101 ratios, N(BC) and B/C, where A is the
duration of the intragesture 101 (i.e., the upbeat),
B is the intergesture 101 preceding the chord
(nominally two eighth-notes) and C is the
intergesture interval following the chord
(nominally three eighth-notes). If the score were
played literally, N(BC) should be either 0.2 or 0.4,
depending on whether the upbeat is notated as an
eighth-note or a quarter-note, and B/C should be
0.67. In Figure 5, however, the ratios have been
normalized with respect to the underlying eighthnote pulse (assumed to be isochronous for this
purpose), so that the expected "literal" ratios
(effectively, tempo ratios) are equal to 1 in all
cases.
Let us consider first the ratio plotted on the
abscissa. The top row of panels in Figure 5 shows

the three instances in which the MG1 upbeat is

nominally a quarter-note. The leftmost panel
shows the unconstrained situation at the
beginning of the piece (bar 0). It can be seen that
very few pianists played this initial upbeat-or
the following intergesture interval, as the case
may be-"as written" (ratio of 1). The large
majority played the upbeat quarter-note short
relative to the intergesture interval. Several
pianists (NEY, COl, BUN) come very close to
playing it as if it were an eighth-note (ratio of 0.5),
and one (ARG) does so without question. This may
have been a deliberate anticipation of the
(notational) shortening of the upbeat in later
incarnations of MGl, but it is a true deviation
from the written score and is perceived as such.
Later, however, when the quarter-note upbeat
occurs together with two eighth-notes in the bass
voice (top center and right-hand panels in Figure
5), pianists are more evenly divided between those
who shorten the upbeat and those who lengthen it
relative to the intergesture interval. While the
average ratio is close to 1, there is enormous
variation among individual artists, and some who
severely shortened the initial, unconstrained
upbeat (e.g., ARG, BUN) do not repeat this
tendency later on.
The lower row of panels in Figure 5 shows
instances where the upbeat is nominally an
eighth-note. In two cases (left-hand and center
panels) the upbeat is accompanied by a bass-voice
eighth-note that marks the approaching end of
MG6 and of a whole phrase, so here the
overwhelming tendency is to lengthen the upbeat,
sometimes to an extent corresponding to a
nominal quarter-note (ratio of 2; BUN and ARR in
bar 12). In bar 16, where the upbeat is notated as
a grace eighth-note, a significant minority of
pianists plays the upbeat shorter than an eighthnote, in one case (KAT) shorter than a sixteenthnote (ratio of 0.5). The majority, however, still
lengthen it, and at least one (ASH) plays the grace
note as if it were a quarter-note (ratio of 2).
There is little indication of clustering or
bimodality in these data. Although there are some
"outliers," the values of the ratio seem to be rather
evenly distributed over a wide range. Correlations
of individual timing patterns across the different
instances of MG1 are not high, though some
consistencies can be seen. Thus ARR and BUN
always tend to lengthen the upbeat, whereas
ARG, DEM, and COI-3 tend to shorten it. Some
pianists, perhaps to be regarded as "literalists,"
consistently avoid extremes (e.g., ESC, KRD,
MOI,ZAK).

Bars 4/5

Bars 0/1
1.6

Argo

1.2-t

I
I

N~
C01.

Bun

1~

--

1.6

' ...

.Arr

Oem.

Sene

She

,lre

...

0.8~

0.4

0.6

0.8

Co2

art

1.2

0.8 ,
0.6

-.,..'

Arre

1.2 _ Oeme

Ho3

Co2.
Col

~ Cur

1.2

1.4

1.8

Gia.

Oem

1.2

Arg

Arr

. , . Ash
,

. . ..~u

Gia

1.6

Nov

1.4

I
I

.c ur.

1
I

1.2

1.4

1.6

0.8 ,
0.8
1.8

1.2

1.6

Bu
Gia'"

1.2,-

,,,
,

~a~.
,
,

,.

0.8 .
2.4
0.4

~
;:,:

1Ii

r;
'\:J

-.

1.4

1.2

Hal

....

Bre

Kub

~;:,:

;;,.

,
,

;:,:

;;,.
;:,:

c'

''-""

r;;

Bun'

1.4,- , art

!
~

Arr

'Ho3.

0.6 .
.
0.6 0.8

3
3

fl
I

n
Q

Bre

,
,
1.6 -

ap

0.8

Bun.

'"

;:,:
M.

Arr.

".8

1.2

Ho1

Bars 16/17

1.8
Nov

0.6

Bars 12/13

Bars 8/9
1.6
1.4

Oav.
Co3

0.8

0.8

Cap

Co3

Co1.

Co2

CO

One

Arg.

1.4 -

Arg.

1.2

--

Nov

.Bun

1.4.-

Ash

Nov.

Bars 20/21

1.6

Oem

1.4~

...,.,

:j

;:;

Ash

$:

r;

~::!

r;

;;;

... C01 Co2,

;:,:
(F>

0.8

1.2

1.6

2.4

:::
S
;:,:
;:,:

A/(B+C)

'I":
-j

...

~;

Figure 5. Timing patterns of the six instances of MG1 and the following chord. (See text for explanation of ratios.) Ratios of 1 represent equal underlying
beats (Le., equal tempo). The upper panels plot the three instances where the upbeat.is nominally a quarter-note, while the lower panels plot the three
instances where the upbeat is nominally an eighth-note (a grace note in bar 16). The dashed lines indicate a doubling or halving of the nominal duration in
execution.

:::

;;;

~.
f-.j
oj;"
\.n

244

Repp

With regard to the BIC ratio, plotted on the

ordinate, there is more consistency. In all six
instances, which are notationally equivalent,
there is a strong tendency to lengthen the rest
preceding the chord, sometimes to the extent that
the onset of the chord occurs in the middle of the
intergesture interval (ratio of 1.5). Among those
who tend to play the chord very late are ARG,
ARR, BUN, and NOV. Nevertheless, there are
some pianists who do not show that pattern, most
notably Cortot.
Finally, it should be noted that the two ratios
are uncorrelated across different artists.
Evidently, the timing ofMG1 is quite independent
of the timing of the chord in the intergesture
interval. s
2. MG2. This is the signature melodic gesture of
the piece, and its manner of execution is crucial to
the impression of a performance of "Traumerei." It
offers a unique opportunity to investigate the
temporal shaping of a gesture, for several reasons:
(1) It comprises six notes (i.e., five IOIs)-a
sufficient number of degrees of freedom for any
constraints on temporal shape to emerge very
clearly; (2) it is unidirectional in pitch and
uninterrupted by metric accents-the accent that
would normally occur on the note following the
bar line is clearly suspended in this case; (3) it
recurs six times, with slight variations; and (4)
pianists are likely to give special attention to its
execution.
MG2 occurs in bars 1-2, 5-6, 9-10, 13-14, 17-18,
and 21-22. (We will not consider the imitation
gesture MG2i in detail.) The versions in bars 1-2
and 17-18 are identical; during the last 101, two
grace notes (a written-out arpeggio) occur in the
left-hand accompaniment of the melody. In bars 56 and 21-22, the penultimate pitch interval is
extended from a fourth to a major sixth; in
addition, the melody in bars 21-22 leads to a
fermata and also lacks the grace notes during the
last 101. In bars 9-10 and 13-14, there are no
grace notes, the penultimate melodic interval is
reduced to a minor third in the first instance,
there is a change of key (to B-flat major) in the
second instance, and the occurrence of MG2i in
the middle voices creates a forward movement
that is absent in the other variants. Therefore,
timing differences are to be expected reflecting
these factors.
The average temporal shapes of the six versions
of MG2 are shown in Figure 6. The five lOIs for
each version, representing the geometric means
across all 28 performances, are plotted here on a
linear scale. The five data points for each version

have been fitted with a quadratic function (i.e., a

parabola), which seems to describe their temporal
shape rather wel1. 9 Each gesture accelerates
initially and then slows down at the end. This
final ritardando is least pronounced in bars 9-10
and 13-14; it is more evident in bars 1-2, 5-6, and
17-18; and in bars 21-22, preceding the fermata, it
is dramatic. Except in this last instance, there is a
tendency for the penultimate IOI to rise slightly
above the parabolic trajectory. This was evidently
due to including in the average performances such
as Cortot's which, as we have seen in Figure 4
(Factor III), showed a pronounced tendency to
shorten the last 101.

1400

1200

Bars 1/2

Bars 5/6

Bars 9/10

'"

Bars 13/14

1000

Bars 17/18

---- 800

Bars 21/22

---.
(J)

600

400
1-6

1-7

1-8

21

2-2

BAR/EIGHTH-NOTE

Figure 6. Geometric mean lOIs for the six versions of

MG2, with best-fitting quadratic functions. The abscissa
labels refer to bars 1 and 2.

The temporal shapes of the six versions of MG2

were examined and fitted with quadratic functions
in each of the 28 individual performances (a total
of 168 instances). It emerged that most individual
artists' timing patterns (87% of all instances)
could be described well by parabolas, with fits
ranging from good to excellent. What varied
between artists and between different instances of
MG2 were the curvature and elevation of the
parabolic functions, but not so much their
goodness of fit (however, see Footnote 9). The data
of six individual artists are shown in Figure 7.

Diversit1/ alld Commonahtliln

l'vll1SIC

Performance: An Anal1/sls of Timin" Microstmctllre

111

5chumaml's "Triillmere:"

1200 - , - - - - , - - - - - , - - . , . - - - - , - - - . m

1200 ..,......--r---,--,---r-.,-,

1000

1000

800 - 4 \ - \ - - t - - - - t - - . ; - - - I t - - r l ' - i

1-6

1-7

1-8

2-1

2-2

245

800 -t---4---1---+-_++--"'''''''''

1-6 1-7 1-8

2-1
2-2
1600....---,---,-----,--.,-----,-.....,
1400

--.

1000
1200-t---+---f---+---ff---;

(f)

.s

800 +---+---+---+-:>.<.+-,I4-;----i 1000 -j-l,---+---f---+--r-t---;

800-t-~-+---+----l-,4~+---i

400-!--~~f=:::::=-+--+-_I 400-!---+-~-T~-+--f----i
1-6
1-7
1-8
2-1
2-2
1-6 1-7 1-8 2-1
2-2
1400 -.----,--,---,-----r----, 1600..,......--,----r----,,.---.,---.....,
1400

1200
1000 -+---+----;---++-~
____1

1200 - i - - - + - - - + - - - i f - - - + - , f - - - i

1000 -t---t---+-+-I-f--,.<p>o"6-"71
800 -+--'rl-----;---f-+h~_,f_-__1

1-6

1-7

1-8

2-1

2-2

400 +--+--=+=:::::l!!::::::PL-~-l
1-6 1-7
1-8
2-1
2-2

BAR/EIGHTH-NOTE
Figure 7. MG2 timing data for six individual artists, with best-fitting quadratic functions. Where the pattern deviated
markedly from a parabolic shape, the data points are connected with straight lines. See Figure 6 for the legend of
symbols.

The two panels on top illustrate two individual

cases that were fit well by parabolas but differed
in curvature: The highly modulated timing curves
of ARR contrast with the flatter ones of SCH. The
center panels show one representative performance of each of the two great individualists,

Horowitz and Cortot. Horowitz's curves are fit

fairly well by parabolas but generally lack the
initial acceleration shown by most other pianists.
Cortot shows a truly deviant pattern: In all three
performances (which spanned 18 years the last
101. This seems to indicate that he grouped the

246

Repp

penultimate tone with the following long tone,

treating it like an upbeat; in fact, the first four
IOIs are fit well by a parabola. This idiosyncratic
timing shape held only for the first five instances
of MG2, however; in bars 21-22, preceding the
fermata, Cortot produced a beautiful parabolic
timing shape in all three performances. Finally, as
can be seen in the bottom panels of Figure 7, ARG
and BUN, two highly variable pianists, intermittently adopted the alternative timing pattern instantiated by Cortot; one other pianist who did so
was CUR. The only other type of deviant timing
pattern was shown by ORT who, in bars 9-10 and
13-14 only, lengthened the third IOI above the
parabolic trajectory, which led to a W-like shape.
The quadratic functions (y = C + Lx + Qx 2 ) that
fit the large majority of individual phrasal shapes
are characterized by three parameters: a positive
constant C that reflects vertical displacement,
related to overall tempo; a generally negative
coefficient L that reflects horizontal-vertical
displacement of the curve in x-y coordinate
space;10 and a positive coefficient Q that reflects
the degree of curvature of the concave (U-shaped)
parabola. A question of great theoretical interest
was whether there are any constraints among the
three parameters; in principle, of course, they are
quite independent of each other. The coefficients L
and Q were plotted against each other for all 146
individual gestures that followed a parabolic
shape. There was a remarkably tight linear
relationship between these two parameters, with
correlations ranging from 0.93 to 0.98 across the
six versions of MG2. The reason for this result is
that the location of the minimum of the function is
relatively fixed. Therefore, as the curvature of the
parabola increases, the negative slope of the
tangent at xO also increases.
If the x coordinate of the minimum is relatively
fixed, then the constant C must be correlated with
the y intercept of the function and should increase
as a function of curvature Q. There were indeed
positive linear correlations between C and Q,
ranging from 0.79 to 0.93. 11 Thus, both Land C
are fairly predictable from Q, which leads to the
conclusion that the expressive timing patterns of
the large majority of the performances of MG2 can
be characterized by a single family of parabolas,
varying in Q only. This family of functions is
shown in Figure 8; it was generated by increasing
Q from 20 to 140 in steps of 20, and by setting L
and C according to their average linear
regressions on Q. Figure 8 embodies a constraint
on the expressive timing shape of MG2 that was
obeyed by the majority of pianists. With few

exceptions, any individual execution of this phrase

can be characterized by a single curvature
parameter, with some additional freedom in global
tempo (vertical displacement), which is not
represented in the figure.

1400

1200

..--...
(j)

1000

---

Q = 20

Q= 40

Q = 60

A.

Q= 80

Q = 100

Q = 120

\I

800

600

400
1-6

1-7

1-8

2-1

2-2

BAR/EIGHTH-NOTE

Figure 8. Family of quadratic functions that describes the

timing constraint on MG2 observed by the large majority
of pianists. The functions vary in curvature (Q), roughly
over the observed range. The other coefficients of the
quadratic functions were obtained by the average
regression equations. Vertical displacement should be
considered an additional free parameter.

To give some impression of individual

differences in tempo modulation during MG2,
Figure 9 plots the average Q coefficient for bars 12,5-6, and 17-18 (which were generally executed
similarly; cf. Figure 6) against that for bars 21-22,
with individual artists identified. (Cortot is
excluded from this plot.) Most pianists employed
moderate modulation in the earlier instances of
MG2, but gave a very expressive shape to the last
instance in bars 21-22, which led to the fermata.
Some, most notably DEM, used strong modulation
in all instances. Horowitz's three performances
are at the other extreme. They appear relatively
unmodulated despite a substantial ritardando
because he did not show any initial accelerando;
therefore, his timing patterns were fit by
parabolas with low curvature.

DiversitVi1nd Commonalitv in Music Performance: An Analusis of Timll1\i Microstructure in Schumann's Triilin/erei"

160

Dav

De_

Na~

140

Esc

(\J

~ 100

She

Cur 0

,,-

80
Gia

0:

0)

60
40
20

o,
o

Ie

Arg

An

Ka~

Sre

Neye
SCh

(\J

(f)

Cap

Ash

Ki~ Zak

ru

120

Kli

Un!.

HaZ.
Ha3

Hal

I
20

40

60

80

100

120

140

160

BARS 1/2, 5/6, 17/18

Figure 9, Individual variations in tempo modulation

during MG2. The average curvature (Q) for three
instances of the gesture is plotted against the curvature
for the last instance (bars 21-22). Only cases following
the parabolic timing shape are included.

This "parabolic constraint" on the timing shape

of an expressive gesture is of great theoretical
interest, as it supports Todd's (1985) suggestion of
the parabola as a basic timing function. It also
appears to be consistent with his more recent
modelling (Todd, 1992b), even though he proposed
that a linear function can account for most of the
timing variation; clearly, a quadratic function did
even better. The parabolic constraint may thus be
understood as an allusion to physical motion (cf.
Todd, 1992b), which most performers aim at, and
which listeners with some musical education find
aesthetically pleasing (Repp, 1992b).
Since MG1, the following inter-gesture interval,
and MG2 may form a coherent half-phrase that is
planned and executed as a single expressive unit,
their timing relationships acrosS the 28
performances were examined (in bars 0-2 only).
None of the three coefficients characterizing the
parabolic shape of MG2 correlated significantly
with either of the two 101 ratios considered in
connection with MG1 (Figure 5). However, the
total interval preceding the onset of MG2
correlated moderately (0.59-0.72) with all three
MG2 coefficients. Thus, the slower the initial

24;"

tempo, the slower and the more inflected was MG2

(see also Footnote 8).
3. Grace notes accompanying MG2. Only three of
the six instances of MG2 have grace notes (really
a written-out arpeggio) in the left hand during the
last 101; they occur in bars 2, 6, and 18. Since all
six instances of MG2 followed a parabolic
trajectory, on the average (cf. Figure 6), the grace
notes as such were not responsible for the
lengthening of the final 101 and the systematic
constraints that it followed. Rather, they seemed
to be fitted into whatever temporal shape the
pianists chose for the primary melodic gesture. We
now turn our attention to the execution of these
grace notes, which exhibited surprising
variability.
In the score, the grace notes are represented as
nominal sixteenth-notes. However, since the
preceding notes in the left hand are nominally
(tied-over) quarter-notes, there is literally no time
for the grace notes in the rhythmic scheme; thus
they must be "taken away" from the preceding or
following note values. The standard way of
execution suggested by the spatial placement of
the grace notes in the score is to play them within
the last 101 of MG2-that is, after the onset of the
penultimate melody tone in the soprano voice, but
before the onset of the final long tone and the
accompanying two-tone chord. However, only 12 of
the 24 pianists consistently played the passage
according to this conventional interpretation of
the notation. (The two repeats of bars 1-8 were
examined separately in this analysis.) It seems
that this brief passage offered pianists ample
opportunity to indulge in "deviations from the
score."
Most of the variants are illustrated
schematically in Figure 10, which uses horizontal
displacement of the critical notes in bar 2 to
indicate the actual succession of tones. The
standard version is shown in (a). As shown in (b),
some pianists delayed the last note in the left
hand until after the octave chord in the right
hand: ARR (slightly, bar 6 only), DAV (bar 2 only),
MOl (bar 6, once in bar 2), NOV (only the first
time in bar 2, but very dramatically), and ZAK
(consistently). Others (c) played all notes of the
left hand before those of the right hand: ARG
(except the second time in bar 6), HOI and H02
(bar 2). In a variant of that pattern, the righthand chord was arpeggiated (DEM, consistently)
or mysteriously interleaved with the broken chord
in the left hand (Horowitz, in bar 6 only, but in all
three performances).

Jl

(a)

'3M.

r1J

Jl

(c)

r1J
I.J

(d)

Jl

,r1J

Jl

(e)

Jl

(f)
I

Figure 10. Schematic illustration of six observed manners

of execution of the grace notes accompanying MG2. The
notation for bars 2 and 18 is shown, but it stands for bar
6 as well. Spatial displacements of the notes symbolize
temporal displacements.

rn all instances mentioned so far, the onset of

the last melody tone of MG2 did follow the two
grace notes, so the grace notes were within; the
last r01 of the gesture. This made it possible to
ask about the relative timing of the grace notes
within the ror (see below). rn some further cases,
however, one or both of the grace notes coincided
with the final melody tone. Thus, SCH consistently played the right-hand octave tones simultaneously with the first grace note (d). BUN (bars
2 and 18) and KUB (bars 2 and 6, second repeat
only) played the right-hand tones simultaneously
with the second grace note (e). Finally, Cortot (bar
6, in all three performances; also bar 18 in COl)
played all tones simultaneously as a single chord
CD, as prescribed only in bar 22 of the score. (BUN
also did this once in bar 6.)12
To investigate the relative timing of the grace
notes within the last ror of MG2, provided they
did fall within that lOr (cases a-c in Figure 10),
the r01 was divided into three parts (A, B, and C),
defined by the respective tone onsets. Two interval
ratios were calculated: A/(BC), which indicates the
relative delay of the onset of the first grace note,
and B/C, which represents the duration of the first
grace-note ror relative to the second. If the first
melody tone and the two grace notes were played
as a triplet, for example, the two ratios would be
0.5 and 1, respectively; if the grace notes were
played as thirty-second notes following a
sixteenth-note rest, the ratios would both be 1.
Figure 11 plots A/(B+C) against B/C for all instances where these ratios could be determined.
(Bars 2 and 6 are each represented by a single ratio, averaged across the two repeats.) A wide
range of ratios was found, but most data points
form a cluster in the lower left quadrant, suggesting that the majority of pianists agreed on a particular timing pattern. The central tendency of
that cluster is around A/(B+C) = 0.4 and B/C = 0.5.
This means that the grace notes were not played
evenly: The first grace note started about 40% into
the ror, and the second grace note was about
twice as long as the first, which means it started
about 60% into the ror, on the average.
Considering the compressed scale of the ordinate
in the figure, however, it is clear that there was
wide variation in the relative durations of the two
grace notes, even within the cluster of relative
conformity. (Some of this variation undoubtedly
reflects measurement error magnified by ratio calculation.) What was almost always true, however,
is that the first grace note started during the first
half of the lOr (A/(B+ C) <1), and the second grace
note ror was longer than the first (H/C < 1).

Diversitv and Commonalihl

In

Music Performance: An Anal\fsls of Timing Microstructure

2.5

11l

Schumann's "Triiumerei"

249

I
0

2 -f---

Bar 2
Bar 6

V\

Zak

0
Zak

t::. Bar 18

She.6

--

1.5
Co2

Bun

CO

~o

oJa-o_t
t::.

~I

0.5

L...J

0.2

R~

let::.

~_O

t::..

She

t::.

Co10

o Kli

Cao

-t::.

:e~
0.4

~u t

0.6

0.8

1.2

1.4

1.6

A/(B+C)
Figure 11. Relative timing of the grace notes within the last 101 of MG2, for cases (aHc) in Figure 10. The abscissa
shows the ratio of the time before the onset of the first grace note (A) and the remainder of the 101 (Be); the ordinate
shows the ratio of the time between the two grace note onsets (B) and the time from the onset of the second grace note
to the end of the 101 (C). Ratios for bars 2 and 6 are averaged over the two repeats.

There were some notable exceptions to this

pattern, though no individual pianist was
consistently deviant. Assuming that these
instances are not simply slips of motor control, it
would be interesting to know what led individual
artists to vary their timing patterns in these ways.
4. MG3a-MG6a. We turn now to a closer
examination of the MG chains making up the
second half of each phrase. The Type a chain
occurs in identical form in bars 2-4 and in bars 1820 (see Figure 2). It also occurs in varied and
abbreviated form in bars 22-24. We will not deal
specifically with MG3a and MG3b in bars 22-23,

which on the whole were played as in bars 2-3 and

18-19. MG5a and MG6f in bars 23-24, which
carried the final large ritardando, will be
considered separately later on.
As noted earlier, the average timing profiles for
bars 2-4 and 18-20 were extremely similar, though
bars 18-20 were played at a somewhat slower
tempo overall (see Figure 3). In the principal
components analysis conducted on bars 0-8, the
first factor (see Figure 4, top panel) exhibited a
regular sequence of peaks for MG3a-MG6a, which
reflect lengthening of accented tones and MG-final
lOIs. Detailed examination of the individual

Repp

250

timing patterns, however, revealed enormous

variety, though not without constraints. No two
pianists' interpretations of this MG chain were
quite the same, and some artists also changed
their patterns significantly between bars 2-4 and
bars 18-20. (The two repeats of bars 2-4 will not be
considered separately here, though some pianists
played even these differently.)
To explore this variability in a reasonably
economical way, principal components analysis
was used once again, but this time only on the 2 x
18 36 IOIs comprising MG3a-MG6a in bars 2-4
and 18-20, concatenated. From this analysis, six

1000

Q]

900

Bars 2-4

900

-e-

Bars 18-20

700

700

600

600

500

500

....
c\J

-.-

800

....--..
(f)

1000

800

400

.........

MG6a

significant factors emerged-more than from the

earlier analysis of bars 0-8, where the larger
timing deviations in MG1 and MG2 dominated the
correlation structure. Thus, there were at least six
distinct (i.e., uncorrelated) timing patterns
underlying the variation in the data; these
patterns are shown in Figure 12. None of them is
fully representative of any individual
performance, however, and together they account
for only 81% of the variance. The factor scores for
bars 2-4 and 18-20 were highly similar and are
superimposed in the figure. The factor loadings of
the individual performances are shown in Table 6.

(")

If)

....

(")

If)

....

400

(")

If)

....

....

(")

If)

(")

If)

....

<'?

"(

r-;-

1000

1000

OJD

900

900

a:

800

()

700

700

600

600

I- 500

500

....

c\J

800

(f)

a:
0

()

u..

400

....
c\J

(")

If)

....

(")

If)

....

400

~.

1000

(")

If)

....

1000

900
800

800
700

600

600

500

500

....
c\J

@J

900

700

400

....

c\J

(")

If)

....

<'?
~

If)

....

400

....

c\J

C')

If)

....

BAR/EIGHTH-NOTE
Figure 12. Rescaled factor scores (underlying timing patterns) for the MG chain of Type a in bars 2-4 and 18-20. The
abscissa labels refer to bars 2-4.

Diversitll and Commonalitll m ivfusic Performance: An Anaillsis o(Timing Microstructure m Schumann's "Triiumerei"

251

Table 6. Sorted rotated factor loadings from the principal components analysis ofMG3a-MG6a. (Loadings below 0.4
are omitted.)
II

ARR

NEY
BRE
ASH
KAT
ESC
H03
H02
HOI
ARG
COl
CO2

0.773
0.758
0.744
0.708
0.671
0.606

IV

0.446

VI

0.402
0.873
0.832
0.784
-0.593
0.823
0.808
0.786
0.676

0.564
0.456

0.457
0.427
0.431
0.414

0.437
0.416
0.908
0.893
0.830
0.629

cm

ORT
KUB
NOV
SCH
CUR
MOl
SHE
DAV
GIA
KLI
DEM
ZAK
BUN
KRU
CAP

III

0.410
0.576
0.436
0.515

Factor I (VAF = 18%) is characterized by acceleration during MG3a; a fast traversal of MG4a; a
dramatic ritardando within MG5a with maximal
lengthening of the final, unaccented note (position
4-2), further augmented in bars 18-20; and a
pronounced lengthening of the last 101 in MG6a
(position 4-8). The pianists whose interpretations
come closest to this pattern are ARR, NEY, BRE,
and ASH.
Factor II (VAF = 17%) is the "Horowitz factor."
It is characterized by significant lengthening of
the accented tone in MG3a as well as of the preceding unaccented tone; slight deceleration during
MG4a; final lengthening in MG5a; and a ritardando during MG6a, followed by a sudden shortening of the last 101. Some of these tendencies
were exaggerated in bars 18-20 relative to bars 24. Horowitz's actual performances (especially H02
and H03) resemble this pattern, except that the
exaggeration in bars 18-20 is much more dramatic. H03 differs in that bars 2-4 and 18-20 are
executed very similarly, both showing dramatic
lengthening in positions 4-2 and 4-3.
Factor III (VAF = 14%) is the "Cortot factor." It
is characterized by a fast start; progressive slowing down during MG4a and MG5a, with final

0.535

0.541
0.864
0.693
0.574
0.564
0.602
0.565
0.540

0.410

0.491
0.500
0.455

0.436

-0.435

lengthening in MG5a; and some shortening of the

final 101 in MG6a. Some additional peaks appear
in bars 18-20. Cortot's real performances are similar, especially COl and C02, except that the tendencies in bars 18-20 are much more exaggerated.
In C03, there is much lengthening in position 202 instead of 20-3 (read 4-2 and 4-3 on the abscissa
in Figure 12). Note the sizeable negative loading of
ORT.
Factor N (VAF = 14%) shows pronounced final
lengthening in both MG3a and MG4a; a "flat"
MG5a with slight final lengthening; and a smooth
ritardando through MG6a, with a very long last
101. Bars 2-4 and 18-20 are executed very similarly. This beautifully regular pattern comes closest to the actual interpretations of KUB, NOV,
and SCH.
Factor V (VAF = 11%) shows a rather flat
pattern, with initial acceleration and pronounced
final deceleration, but with the final 101 shorter
that the penultimate one. This pattern is the least
differentiated, as if the whole MG chain were
thought of as a single gesture. This pattern comes
closest to MOl's performance, though he does
show some small local peaks in positions 3-5 and
4-1.

Repp

Factor VI (VAF = 7%), finally, begins with a

pattern for MG3a and MG4a that is very nearly
the opposite of Factor II; MG5a shows striking
final lengthening; and MG6a shows only a small
lengthening of the final 101. There is an overall
trend to accelerate during the MG chain. This
pattern is not representative of any individual
performance and occurs only in mixed patterns,
such as those ofKLI, DEM, and ZAK, all of which
exhibit pronounced lengthening in position 4-3.
In summary, the information contained in
Figure 12 and Table 6 offers only a moderate
amount of data reduction. The timing patterns of
individual artists were remarkably diverse.
Nevertheless, there are many possible patterns
that never occurred, such as lengthening in
position 3-4 (the second 101 of MG4a), which
clearly was treated as transitional by all pianists.
Nor were the patterns random in any way; the
majority of the artists produced very similar
timing patterns in bars 2-4 and 18-20, and also in
the two repeats of bars 2-4 (which were averaged
here). The performances with the highest
consistency between bars 2-4 and 18-20 were
ARR, CAP, HOI, and KUB, whereas the most
variable ones were BUN, Cal, C02, H02, H03,
and aRT. That Cortot's and Horowitz's withinperformance variations were carefully planned is
suggested by the fact that they were replicated in
different performances (only HOI retained the
pattern of bars 2-4 for bars 18-20). Whether
BUN's and aRT's inconsistencies were similarly
planned is not known, since only a single
performance by each pianist was available for
examination.
5. MG3b-MG6b. The Type b version of the MG
chain occurs three times in the piece: first in bars
6-8, and then in bars 10-12 and 14-16. Because of
the substantial ritardandi that occur during
MG6b, it was decided to treat the last six lOIs of
each chain separately. The remaining 3 x 12 = 36
lOIs were subjected to principal components analysis, which yielded six significant factors, just as
for the Type a passage. They accounted for 80% of
the variance. Their timing profiles are shown in
Figure 13, and the matrix of factor loadings is
shown in Table 7. As for the Type a chain, the
factor patterns are only rarely representative of
individual artists' patterns, which more often are
a combination of several factor patterns. It may be
recalled that the grand average timing profile
showed a fairly regular lengthening of pre-accented and accented lOIs in each MG, which
really represents final lengthening of the
accompanying secondary MG, which ends with an

accented tone. The factor profiles essentially

represent a varying focus on individual MGs in
the chain.
Factor I (VAF = 22%) shows substantial
lengthening in MG3b, much less in MG4b, and
somewhat more in MG5b. The timing profiles for
the three instances of the MG chain are very
similar. The pianist most closely associated with
this pattern is ARR, and a number of other artists,
including Horowitz, show substantial loadings in
this factor. (In contrast to the Type a chain, there
was no "Horowitz factor" here.)
Factor II (VAF = 21%) is characterized by total
absence of lengthening in MG3b and variability
during MG4b and MG5b, usually with more preaccentuation lengthening than accentuation
lengthening. Two performances showing this pattern in relatively pure form are Cal and KRU,
and several other performances (including C02)
show substantial loadings. Thus this was again a
kind of "Cortot factor," though it was not unique to
Cortot. However, C03 deviates from this pattern
and is responsible for a unique factor, after all
(Factor VI).
Factor III (VAF = 14%) is characterized by a flat
pattern (bars 6-7) or pronounced accelerando (bars
10-11 and 14-15) during MG3b, variable lengthening in MG4b, and pronounced lengthening in
MG5b. NEY, KAT, and DAV are the best representatives.
The remaining three factors account for much
less variance (8% in the case of Factors IV and V,
6% in that of Factor VI), and each of them is
largely associated with one particular artist.
These three artists, which earlier analyses have
already revealed to be somewhat eccentric, are
aRT, BUN, and C03. Thus, aRT put increasing
emphasis on MG4b in successive renditions, BUN
showed a very atypical peak in position 7-7 and
tremendous lengthening in MG4b in bar 7 only,
and C03 showed no lengthening in MG3b and
MG4b, and a very variable execution of MG5b.
On the whole, within-performance variability
across the three instances of the Type b chain
seemed larger than for the Type a chain, whereas
between-performance variability seemed somewhat lower. The former observation is not
surprising, for the two MG chains of Type a are
identical, whereas the three chains of Type b
differ harmonically and melodically. No pianist
showed nearly identical patterns for all three
instances of Type b. Some pianists were markedly
more variable than others, however, not to say
erratic. BUN leads the list, which also includes
ARG, C03, DAV, and MOL The relatively most

Diversitll and Commonality in Music Performance: An Anaillsis of Timing Microstructure in Schumanrz's "Triiumerei"

at the end of the piece. A substantial slowdown

occurs also in bar 16, at the end of period B. Less
pronounced ritardandi occur in bar 8, at the end
of period A, and in bar 12, at the end of phrase b2.
Thus there are four locations where major
ritardandi are observed. It may be asked whether
the temporal shapes of these ritardandi followed
any common pattern.

consistent pianists were ASH, DEM, H02, KLI,

NEY, and ZAK. Some pianists (BRE, KRD, SCB)
were notable for their relatively flat and
uninflected rendering of the whole Type b chain. 13
6. The ritardandi. As is evident from the grand
average timing pattern plotted in Figure 3, major
ritardandi occurred at the ends of periods and
phrases. The largest ritardando is in bars 23-24,

-+- Bars 68

750

[I]

650

( J)

253

550

550

450

450
350

350
7-3

67

7-5

7-7

8-1

750

8-1

6-7

7-1

7-3

75

7-7

6-7

71

7-3

7-5

77 . 8-1

7-5

7-7

750

---(J)

DE]

W 650

650

a:
0

550

<-) 550
(J)

a:

450

450

f-

350

LL

350

67

7-1

7-3

7-5

77

8-1

750

750

650

650

550

550

450

450

C2TI

350

350
6-7

7-1

7-3

7-5

77

8-1

6-7

7-1

7-3

8-1

BAR/EIGHTH-NOTE

Figure 13. Rescaled factor scores <underlying timing patterns) for MG chain of Type b in bars 6-8, and 10-12, and 14-16.
The abscissa labels refer to bars 6-8.

Repp

254

Table 7. Sorted rotated factor loadings from the principal components analysis of MG3b-MG5b. (Loadings below 0.4
are omitted.)
II
ARR
HOI
ARG
H03
KLI
MOl
ASH
DEM
ESC
GIA
CAP
COl
KRU
BRE
CO2
NOV
SHE
CUR
SCH
KUB
NEY
KAT
DAV
ZAK
ORT
BUN
C03
H02

0.923
0.788
0.745
0.740
0.715
0.701
0.648
0.589
0.553
0.531
0.515

0.418

III

IV

VI

0.521
0.582
0.490
0.455
0.438
0.860
0.829
0.733
0.699
0.691
0.675
0.664
0.629
0.530

0.407

0.450

0.407
0.488
0.404
0.416
0.814
0.746
0.721
0.642

0.416
0.893
0.756
0.863

0.413

This question was addressed previously by

Sundberg and Verrillo (1980) and Kronman and
Sundberg (1987) with regard to the ritardandi
observed at the ends of performances of "motor
music," mostly by Bach. Sundberg and Verrillo
plotted an average "retard curve" in terms of
inverse lOIs and observed that it could be
described in terms of two linear functions: a
gradual slowdown followed by a more rapid one.
The latter phase often included only three data
points, however, and tended to coincide with the
last melodic gesture in the music. Kronman and
Sundberg abandoned the bilinear model and
interpreted the average retard curve as a single
continuous function. That function was described
well by expressing local tempo (lIIOl) as the
square root of the normalized distance from a
. hypothetical zero point located one beat beyond
the onset of the final tone. Kronberg and
Sundberg speculated that this function may be an
allusion to physical deceleration in natural
activities such as walking.
That the average ritardando followed a
quadratic function is intriguing given the present

0.452

results concerning the temporal shape of MG2.

The decelerando in that ascending melodic gesture
may be considered an instance of a' local
ritardando, and the principles involved may be
quite the same. A square-root (convex) function
fitted to reciprocal lOIs implies a quadratic
(concave) function fitted to the original lOIs.
However, it is difficult to reach strong conclusions
from Sundberg's examination of timing patterns
averaged across performances of different music.
The melodic grouping structure must be taken
into account, which was done only to a very
limited extent by Sundberg and his colleagues.
This is much easier, of course, when dealing
with multiple performances of a single piece of
music, as in the present case. Inspection of the
grand average timing profile in Figure 3 suggests
that ritardandi progress smoothly within melodic
gestures but are interrupted between gestures.
Thus, the final ritardandi in bars 12 and 16 (read
bar 8 on the abscissa) can be seen to begin with
the second 101, but a "reset" occurs after the
fourth 101, which corresponds to a gestural
boundary. The final lengthening of MG5b causes

Diversitll and Commonalitllll1 lvfusic Performance: All Allah/sis of Timzlll: Microstructure

the initial ritardando to "overshoot" its trajectory,

which is resumed with the onset of MG6b. These
ritardandi thus are divided into two sequences of
3 and 4 IOIs, respectively. Similarly, in the large
final ritardando in bars 23-24, which begins as
early as position 23-3 or 23-4, two resets occur,
corresponding to the ends of MG4a and MG5a,
which in this instance are unmistakably indicated
in the score by "commas." This grand ritardando
thus is divided into three groups of IOIs: 3(4), 4,
and 2. Because meaningful fitting of quadratic
curves requires at least four data points, the
following analyses examine one sequence of four
coherent lOIs from each of the three ritardandi,
forming either part of MG6b or of MG5a. The
analyses were analogous to those conducted on
MG2.
In each case, it was found that a quadratic curve
fit the average 101 pattern quite well, perfectly so
in the case of MG5a. These curves are shown in
Figure 14. Quadratic functions were subsequently
fitted to all 28 individual performance timing
patterns of each ritardando section. Most fits
ranged from excellent to acceptable (see Footnote
9). In the case of MG6b (bar 12), there were five
clearly deviant cases (ARG, H03, KUB, and ORT,
all of whom shortened the last IOI, and KLI, who
showed a rather flat pattern); for MG6b (bar 16)
there was none; for MG6a there was one (ARG,
who shortened the last 10l).
The coefficients of the quadratic equations
describing the acceptable (and better) fits were
found to be highly correlated in all instances, just
as for MG2. Thus it was again possible to
construct families of parabolas that capture a
major portion of the individual variation in the
shapes of the ritardandi. These curves are shown
in Figure 15. Taking into account the different
time scales on the abscissa, it seems that each
ritardando has its distinctive range of variation,
but all share the property of being largely
parabolic in shape. These results for strictly
intragestural ritardandi support the observations
on more heterogeneous materials by Sundberg
and Verrillo (1980) and Kronman and Sundberg
(1987).
A final analysis was conducted on the last two
lOIs of the piece. While it did not make sense to fit
any function to them, their correlation could be
examined, which was 0.87. Thus, the longer the
penultimate 101, the longer the final 10I.l4 Not
only was the relationship quite linear, but the
regression line passed almost through the origin,
suggesting that the two lOIs were in a constant
proportion (1:1.8).

III

Scizwnann 's "Traumae:"

1000
900
---.
(f)

--0E

700
600

800

:!:1:'

500
12-5

12-7

12-6

2200

---.

12-8

I
V

1800

en

51400

/'

0
/

1000

-.-

~/

600
16-5

16-7

16-6

16-8

1400

1200
---.

(f)

51000

o
800

1/
~

600

23-7

23-8

24-1

24-2

BAR/EIGHTH-NOTE

Figure 14. Ritardando functions: Quadratic functions

fitted to the grand average (geometric mean) lOIs in
MG6b (bars 12 and 16) and in MG5a (bars 23-24).

256

Repp

1200

o ~ -40
o ~ 20

1000

0=0

(/J

.L

0=20

0=40

800

0= 60

"l

0=80

0=100

0=0

0=40

600
400
125

126

12-7

12-8

2400
2100
~1800

0=80

-=-1500

.L

0=120

Q 1200

0= 160

900

0=200

"l

0=240

(/J

i=

600
16-5

16-6

16-7

16-8

1600
1400
:-[1200

1000
800
600
237

23-8

24-1

0=20

0=0

0=20

.L

0=40

0=60

0=80

"l

0= 100

0=120

24-2

BAR/EIGHTH-NOTE

Figure 15. Families of parabolic timing functions for the

ritardandi in MG6b (bars 12 and 16) and in MG5a (bars
23-24).

v. GENERAL DISCUSSION
The present comprehensive analysis of
expressive timing patterns in 28 performances of
Schumann's "Traumerei" provides an objective
view of the commonalities and differences among
great artists' interpretations of one of the
masterpieces of the piano literature. At first
glance, the differences are perhaps more striking
than the commonalities. There is ample material
here to support the view that every artist's
performance is, in some sense, unique and unlike

any other artist's, even if just one physical

dimension (timing) is considered. Even the same
artist's performances on different occasions, while
demonstrably similar, are sufficiently different to
be considered distinct and individual events. Yet,
there are also significant commonalities that
apparently reflect constraints on performance, at
least on those performances that have been
deemed suitable for commercial distribution. The
individual variations among performances largely
take place within these constraints, although
there are always exceptions, representing
conscious or unconscious transgressions of the
boundaries established by musical convention.
How should those boundaries be characterized?
And are they purely conventional (i.e., arbitrary),
or do they represent more general laws of motor
behavior and perception that music performance
must conform to in order to be naturally
expressive?
It is probably futile to attempt to characterize
the boundaries of acceptable performance practice.
They are manifold and are likely to contract and
expand as a function of many factors.- It is
theoretically more parsimonious to conceive of a
performance ideal (norm, prototype) that lies at
the center of the hypothetical space enclosed by
the boundaries. This ideal may be thought of as a
relatively abstract specification that contains free
parameters, so that a multiplicity (if not an
infinity) of concrete performances can be
generated that all more or less satisfy the norm.
Because of the enormous complexity of serious
music, any concrete performance ideal is very
difficult to attain; and, if one were attained, the
artist would probably change it the next time,
because art thrives on variety. However, the
underlying abstract ideal may remain constant, as
it embodies generally accepted rules of
performance practice. Concrete realizations are
conceived, perceived, and judged with reference to
the underlying norm, but deviation from the norm
(within certain limits of acceptability) can-to
some extent, must-be an artistic goal. That is,
diversity is as necessary as is commonality: Both
uniformity and lack of an aesthetic standard are
detrimental to art.
One obvious free parameter is tempo. Even
though there may be an "ideal tempo" for a piece
of music, this is again an abstraction. There is in
fact a range of acceptable tempi, and individuals
may differ considerably in what they consider
"the" ideal tempo. This is illustrated by the performance sample examined here, which represents
a very wide range of tempi, nearly all of which

Diversity and Commonalitv in Music Performance: An Analysis of Timing Microstructure in Schumann's "Triiumerei"

seem acceptable to a musical listener (the author

must rely on subjective judgment here), even
though some sound clearly slow, while others
sound fast, To the author, only ESC really sounds
"too slow" (though perhaps even that judgment
might change if his performance were heard in the
context of the whole "Kinderszenen" suite), while
that of DAV sounds "too fast," though apparently
(Clara) Schumann wanted it that way (see Repp,
1992d). An unusually fast "Traumerei" indeed
seems to come across better than an unusually
slow one.
The present investigation substantiates two abstract timing constraints embodied in the hypothetical performance ideal. One of them concerns
the temporal marking of the melodic/rhythmic
structure, the other one has to do with the temporal shaping of individual melodic gestures, particularly of the ritardandi within them. The first
constraint is by now well known and is implemented in Todd's (1985) model of expressive timing at the phrase leveL The principle is that
boundaries in the hierarchical grouping structure
are generally marked by ritardandi whose extent
is roughly proportional to the "depth" of the
boundary. Thus the most extensive ritardando occurs at the end of the piece, where boundaries at
all levels coincide; substantial ritardandi occur at
the ends of major sections, such as 8-bar periods;
smaller ritardandi occur at the ends of individual
phrases and gestures. A performance system such
as that devised by Todd may come close to the
principles embodied in the performance ideal,
though it is not known at present whether a performance strictly following such rules would in
fact be perceived as "ideal" by listeners. There is
some perceptual evidence, however, that listeners-even those without much musical education-expect phrase-final and gesture-final
lengthening to occur (Repp, 1992a).
What is clear from the performances examined
here is that variability increases at lower levels of
the structural hierarchy. Virtually every performer observes the major ritardandi at the ends
of major sections, though in different degrees.
When the complete performance timing profiles
were subjected to principal components analysis,
only a single factor emerged, which reflected the
qualitative conformity in that regard. When the
analysis was restricted to the first 8 bars only,
four factors emerged. When the analysis focused
on half a phrase, omitting major ritardandi, six
factors emerged. It seems paradoxical that the
number of independent factors increases as the
number of data points decreases. What this

257

reflects is the increasing pattern variability at

lower levels of the structural hierarchy.
This variability is not likely to reflect a decrease
of control over precision in timing, though some of
it may. Pianists at the level of accomplishment
studied here can control their timing patterns
down to a very fine grain. What the variability
presumably reflects is a relaxation of the performance constraints imposed by the grouping
structure at lower levels of the hierarchy. Not only
are the boundaries between individual melodic
gestures perhaps less definite than those between
larger units, but they are weaker determinants of
the timing pattern and compete with other local
factors including harmonic progression, melodic
pitch contour, and texture. It also seemS that the
major source of timing variation is not artists'
choice of unexpected locations for expressive
lengthening (though some of this occurred, too)
but varying degrees of emphasis on expected
locations. Thus, pianists often omitted
lengthening where the detailed grouping structure
might have predicted it, whereas they
overemphasized other grouping boundaries, as if
to compensate. Sometimes this resulted in the
creation of hypergestures, which combined two or
three elementary melodic motifs. In other words,
the lowest level of the grouping hierarchy is
structurally flexible; it permits the marking of
group boundaries but does not prescribe it. Artists
choose from among the possibilities in a manner
comparable to varying focus in a spoken sentence.
The other constraint observed in the present
data is more novel and more tentative. It is that,
within melodic gestures requiring a ritardando for
whatever reason, this local tempo change is best
executed such that successive lOIs follow a
parabolic function. This constraint was not only
observed in the majority of performances at four
different locations in the music (nine, if the
different occurrences of MG2 are counted
separately), but it also is in agreement with the
observations of Kronman and Sundberg (1987) on
the shape of final ritardandi, as originally
described by Sundberg and Verrillo (1980).
Sundberg and Verrillo also provided perceptual
data suggesting that listeners prefer ritardandi
corresponding to the original (quadratic) timing
curves over other possible timing profiles, and a
recent study by this author (Repp, 1992b) on the
perceptual evaluation of different timing patterns
for MG2 led to similar conclusions.
It is not clear at present whether the parabolic
timing constraint can also be applied to melodic
gestures that do not include any pronounced

258

Repp

ritardando; certainly other factors, such as accent

and Cortot are two of the most unusual pianists of
this century, and their greatness may lie precisely
location, would have to be taken into account. The
in their individuality, which challenges the
larger the number of successive tones in a melodic
listener. However, in the context oflistening to 28
gesture, the stronger the constraint is likely to be;
different performances in sequence, which heighta minimum of five tones (4 lOIs) is required. It
ens one's sensitivity to differences and perhaps inalso remains to be seen whether a similar timing
creases one's reliance on an internalized perforconstraint operates at higher levels of the
mance ideal, the unusual performances do not fare
grouping hierarchy. Todd (1985) postulated a
parabolic timing function as the prototype for . so well. The author's favorites are the performances by Curzon, Brendel, and Ashkenazy,
within-group timing at the phrase level, though
which usually were in the middle field in the variapparently more for the sake of convenience than
ous analyses reported above, and hence were
for any stringent theoretical or empirical reason.
mentioned only rarely. The author's aesthetic
Kronman and Sundberg (1987) hint at a
ideal
seemed to correspond to the central tendency
grounding of the parabolic timing constraint in
the
sample examined here. Whether this ideal
of
elementary principles of (loco)motion, but without
represents a stable representation of traditional
elaborating on this hypothesis. Todd (1992b) has
performance norms or whether it is a form of psyclaimed that a piecewise-linear function is
chophysical adaptation to the range of the perforsufficient to describe local timing changes during
mance sample is an intriguing question that wara performance, but his own (rather limited) data
rants further investigation. Also, the absence of
suggest that a quadratic function provides a better
the context of the preceding and following pieces
fit. At this point, the general hypothesis may be
of "Kinderszenen" must be acknowledged; interstated that a parabolic timing profile is in some
sense "natural" for both performer and listener,
pretations that sound unusual in isolation may
and probably not for purely conventional reasons.
sound more convincing in context. Finally, it must
be noted that the author's impressions derived not
A better understanding of the origins of this
only from the timing patterns of these perforconstraint may elucidate the meaning of the
mances, but also from their intensity patterns
motion metaphor that is often applied to music
(both "horizontal" and "vertical"), their articulaand its performance (cf. Todd, 1992a, 1992b;
tion and pedaling, and their overall sound quality.
Truslit, 1938).
The present performance analyses revealed no
There are many parameters that contribute to the
subjective impression of a performance, but the
clustering of individual artists according to sex,
timing pattern is probably the most important
age, or national origin. Their individuality
apparently transcended these other factors, whose
one. Nevertheless, future performance analyses
relevance to music performance may be
will have to consider these other parameters
questioned in any case. There is no doubt,
(which are much more difficult to measure in
however, that in this sample of highly
recorded performances) as well as the subjective
accomplished artists, perhaps precisely because of
impressions of more than one experienced listener.
their level of artistry, some performances seemed
Ultimately, we would like to know why individual
unusual and even deviant. Subjective impressions
performers play the way they do, and what meswill have to suffice for the time being: Long before
sage they are sending to listeners (cf. Kendall &
Carterette, 1990).
objective measurements confirmed the actual
deviations in their timing patterns, the
The present research illustrates one of two
performances by Argerich, Bunin, and Cortot, and
complementary approaches to the objective
to a lesser degree those by Horowitz and Ortiz,
investigation of musical performance. The other
struck the author as eccentric and distorted.
approach is represented by the work of Sundberg
Interestingly, some of these pianists (especially
and his colleagues on a system of performance
Bunin) also turned out to be inconsistent within
rules (e.g., Friberg, 1991; Sundberg, 1988;
their own performance, as if they had no fixed
Sundberg, Friberg, & Fryden, 1989) and by Todd's
concept and were exploring possibilities. Other
(1985, 1992a, 1992b) modelling of musical motion.
artists, however, were remarkably consistent, for
These models are ingenious and important, but
example Arrau. The author's subjective
they need to be tested on large performance data
impressions were not only highly reliable on
bases, which do not exist at present. In this study,
relistening, but they also correspond to what
a very modest beginning was made towards
professional critics have to say about some of
accumulating and organizing data of sufficient
these pianists' performances. Certainly, Horowitz
depth and variety (though still restricted to a

Diversity and CommonaLitv in Music Performance: An AnaLvsis of Timing Microstructure in Schumann's "Triiumerei"

single composition and a single parameter,

timing) to present a challenge and testing ground
for emerging models of music performance. 15
Soon, of course, much more extensive data bases
should become available with the help of
technological marvels such as the MIDI-controlled
grand piano. There are exciting times ahead for
research on music performance, one of the most
advanced and culturally significant human skills.

REFERENCES
Bengtsson, 1., & Gabrielsson, A. (1980). Methods for analyzing
performance of musical rhythm. Scandinavian Journal of
PsychoLogy,21,257-268.
Brendel, A. (1981). Der Interpret muss erwachsen sein. Zu
Schumann's "Kinderszenen." Musica, 6, 429-433.
Chissell, J. (1987). Liner notes for Claudio Arrau's recording
(Philips 420-871-2).
Clarke, E. F. (1982). Timing in the performance of Erik Satie's
'Vexations.' Acta Psychologica, 50, 1-19.
Clarke, E. F. (1983). Structure and expression in rhythmic
performance. In P. Howell, I. Cross, & R. West (Eds.), Musical
structure and cognition (pp. 209-236). London: Academic Press.
Clarke, E. F. (1985). Some aspects of rhythm and expression in
performances of Erik Satie's "Gnossienne No.5." Music
Perception, 2,299328.
Clarke, E. F. (1988). Generative principles in music performance.
In J. A. Sloboda (Ed.), Generative processes in music (pp. 1-26).
Oxford: Clarendon Press.
Cline, E. (1985). Piano competitions: An analysis of their structure,
value, and educational implications. Doctoral dissertation, Indiana
University.
Clynes, M. (1983). Expressive microstructure in music, linked to
liVing qualities. In J. Sundberg (Ed.), Studies of music performance
(pp. 76-181). Stockholm, Sweden: Publication issued by the
Royal Swedish Academy of Music No. 39.
Clynes, M. (1987). What can a musician learn about music
performance from newly discovered microstructure principles
(PM and PAS)? In A. Gabrielsson (Ed.), Action and perception in
rhythm and music (pp. 201-233). Stockholm, Sweden: Publication
issued by the Royal Swedish Academy of Music No. 55.
Friberg, A. (1991). Generative rules for music performance: A
formal description of a rule system. Computer Music Journal, 15,
56-71.
Gabrielsson, A. (1974). Performance of rhythm patterns.
Scandinavian Journal of Psychology, 15,63-72.
Gabrielsson, A (1985). Interplay between analysis and synthesis
in studies of music performance and music experience. Music
Perception, 3, 59-86.
Gabrielsson, A. (1987). Once again: the theme from Mozart's
Piano Sonata in A major (K. 331). In A. Gabrielsson (Ed.), Action
and perception in rhythm and music (pp. 81-103). Stockholm,
Sweden: Publication issued by the Royal Swedish Academy of
Music No. 55.
Gabrielsson, A. (1988). Timing in music performance and its
relation to music experience. In J.A. Sloboda (Ed.), Generative
processes in music (pp. 27-51). Oxford: Clarendon Press.
Gabrielsson, A., Bengtsson, I., & Gabrielsson, B. (1983).
Performance of musical rhythm in 3/4 and 6/8 meter.
Scandinavian Journal of Psychology, 24,193-213.
Hartmann, A (1932). Untersuchungen tiber metrisches Verhalten
in musikalischen Interpretationsvarianten. Archiv fur die
gesamte Psychologie, 84, 103-192.

259

Horowitz, J. (1990). The ivory trade: Music and the business of music
at the Van Cliburn International Piano Competition. New York:
Summit Books.
Kendall, R. A, & Carterette, E. C. (1990). The communication of
musical expression. Music Perception, 8, 129-163.
Kronrnan, U., & Sundberg, J. (1987). Is the musical ritard an
allusion to physical motion? In A. Gabrielsson (Ed.), Action and
Perception in Rhythm and Music (pp. 57-68). Stockholm:
Publications issued by the Royal Swedish Academy of Music
No. 55.
Lerdahl, F., & Jackendoff, R. (1983). A generative theory of tonal
music. Cambridge, MA: MIT Press.
Lipman, S. (1984). Glenn Gould: His dissent/ An obituary
(originally published in 1982). In S. Lipman, The house of music:
Art in an era of institutions (pp. 79-95). Boston: Godine.
Lipman, S. (1990). Arguing for music/Arguing for culture. Boston:
Godine.
Lussy, M. M. (1882). Musical expression: Accents, nuances, and tempo
in vocal and instrumental music. London: Novello.
Palmer, C. (1989). Mapping musical thought to musical
performance. Journal of Experimental Psychology: Human
Perception and Perfornlance, 15, 331-346.
Povel, D.-J. (1977). Temporal structure of performed music: some
preliminary observations. Acta Psychologica, 41, 309-320.
Repp, B. H. (1990). Patterns of expressive timing in performances
of a Beethoven minuet by nineteen famous pianists. Journal of
the Acoustical Society of America, 88, 622-641.
Repp, B. H. (1992a). Probing the cognitive representation of
musical time: Structural constraints on the perception of timing
perturbations. Cognition, 44, 241-281.
Repp, B. H. (1992b). A constrainton the expressive timing of a
melodic gesture: Evidence from performance and aesthetic
judgment. Music Perception, 10,221-242.
Repp, B. H. (1993a). Objective performance analysis as a tool for
the musical detective. Journal of the Acoustical Society of America,
93 (forthcoming).
Repp, B. H. (1993b). On determining the global tempo of a
temporally modulated music performance.
Reti, R. (1951). The thematic process in music. New York: Macmillan.
Riemann, H. (1884). Der Ausdruck in der Musik. In P. Graf
Waldersee (Ed.), Sammlung musikalischer Vortriige (pp. 43-64).
Leipzig: Breitkopf und Hartel.
Seashore, C. E. (1936) (Ed.) Objective analysis of music performance.
Iowa City, IA: The University Press (University of Iowa Studies
in the Psychology of Music, Vol. IV).
Seashore, C. E. (1938). Psychology of music. New York: McGrawHill. (New York: Dover Publications, 1967.)
Seashore, C. E. (1947). In search of beauty in music: A scientific
approach to musical esthetics. New York: Ronald Press.
Shaffer, 1. H. (1980). Analysing piano performance: a study of
concert pianists. In G. E. Stelmach & J. Requin (Eds.), Tutorials
in motor behavior (pp. 443-455). Amsterdam: North-Holland.
Shaffer, 1. H. (1981). Performances of Chopin, Bach, and Bartok:
Studies in motor programming. Cognitive Psychology, 13, 326376.
Shaffer, 1. H. (1984). Timing in solo and duet piano performances.
Quarterly Journal of Experimental Psychology, 36A, 577-595.
Shaffer, 1. H. (1989). Cognition and affect in musical performance.
Contemporary Music Review, 4, 381-389.
Shaffer, 1. H., Clarke, E. F., & Todd, N. P. (1985). Metre and
rhythm in piano playing. Cognition, 20, 61-77.
Shaffer, 1. H., & Todd, N. P. (1987). The interpretative component
in musical performance. In A. Gabrielsson (Ed.)! Action and
perception ill rhythm alld music (pp. 139-152). Stockholm,
Sweden: Publication issued by the Royal Swedish Academy of
Music No. 55.

Rqp

260

Stein, E. (1962). Form and performance. New York: Knopf. (New

York: Limelight Editions, 1989.)
Sundberg, J. (1988). Computer synthesis of music performance. In
J. A. Sloboda (Ed.), Generative processes in music (pp. 52-69).
Oxford, UK: Clarendon Press.
Sundberg, J., Friberg, A., & Fryden, L. (1989). Rules for automated
performance of ensemble music. Contemporary Music Review, 3,
89-109.
Sundberg, J., & Verrillo, V. (1980). On the anatomy of the retard:
A study of timing in music. Journal of the Acoustical Society of
America, 68, 772-779.
Todd, N. P. (1985). A model of expressive timing in tonal music.
Music Perception, 3, 33-58.
Todd, N. P. (1989). A computational model of rubato.
Contemporary Music Review, 3, 69-88.
Todd, N. P. MeA. (1992a). The dynamics of dynamics: A model of
musical expression. Journal of the Acoustical Society of America,
91,3540-3550.
Todd, N. P. McA. (1992b). The kinematics of musical expression.
Journal of the Acoustical Society of America (in press).
Traub, A. (1981). Die "Kinderszenen" als zyklisches Werk. Musica,
6,424-426.
Truslit, A. (1938). Gestalt und Bewegung in der Musik. BerlinLichterfelde: Chr. Friedrich Vieweg.

FOOTNOTES
'Journal of the Acoustical Society of America, 92, 2546-2568 (1993).
I An effort is made in this manuscript to distinguish between notes,
which are graphic symbols, and tones, which are sound events.
However, when it comes to grace notes, the distinction cannot
easily be made, since "grace tone" is not an acceptable term and
"grace-note tone" is awkward. It should be understood, then,
that "grace note" refers either to the notated symbol or the
resulting sound, depending on the context. The same ambiguity
holds for "chord," although "tone cluster" is an acceptable term
for the sound event that may be used on occasion.
21n a third case (NEY), it was discovered that the two repeats were
virtually identical from bar 5 on, suggesting duplication by the
recording engineers (see Repp, 1993a).
3In the designation for a "position," such as 5-8, the two
numbers stand for the bar and the eighth-note 101 within it,
respectively. In an expression such as "bars 5-8," however, the
two numbers stand for the first and last bars in the range
referred to.
4Most of the performances in the latter group strike the author as
mannered. It seems that these pianists deliberately tried to play
differently from the norm, but were not willing or able to do so
consistently. Perhaps they intended to convey an
improvisatory quality.
5Todd (1992a, 1992b) has recently revised and extended his
model of expressive timing. An application of this model to the
present data would be most interesting but exceeds the scope
of this paper.
6That is to say, there may be an implicit, expressively modulated
eighth-note pulse going through the longer lOIs. As suggested
by the extent of the MG boxes in Figure 2, the first of these
pulses "belongs to" the preceding MG, but its actual duration
cannot be determined unless it is marked by some tonal event
in another voice. MG2i, which breaks up the inter-gesture
interval in the soprano voice, may track the implicit pulse
induced by the primary MG2 and thus may reveal its final
lengthening.
7Although these performances were by no means identical, they
were more similar than almost any pair of performances by

different artists. Cortot's three performances intercorrelated

between 0.80 and 0.82, and Horowitz's between 0.81 and 0.92.
Only three other values in the 28 x 28 intercorrelation matrix
exceeded 0.80: CAP /ZAK (0.86), KAT/SHE (0.82), and
CUR/KRU (0.81). On the other hand, some of the lowest
correlations were observed between Cortot's and Horowitz's
performances (0.15 to 0.44), and also between Cortot and
several other artists (ARG, BUN, ESC, KLI, NEY, NOV). Only
two other correlations fell below 0.30: BUN/MOl (0.26) and
H01/SCH (0.26). Most correlations were between 0.4 and 0.7.
Thus Cortot and Horowitz were rather extreme cases in the
present sample, which agrees with their general reputation as
highly individual artists.
8The possible covariation of adjacent as well as distant lOIs was
investigated in a principal components analysis on the large
(190 x 190) matrix of intercorrelations among all lOIs,
computed across the 28 performances. Its purpose was to
"retrieve" the melodic grouping structure through the factors
extracted, on the expectation that lOIs would covary more
strongly within MGs than between. This expectation was not
fulfilled; the analysis yielded a very large number of significant
factors, none of which captured much of the variance. They did
reflect correlations among some adjacent lOIs and especially
between lOIs in structurally analogous positions across phrases
of the same type. However, for example, there was no
significant relationship between the timing of MG1 and MG2,
nor even between the beginning and the end of MG2.
9Unfortunately, the software used for curve fitting (DeltaGraph)
did not prOVide a measure of goodness of fit. Therefore,
statements about this aspect of the data will remain
impressionistic in this paper.
lothe L coefficient can be understood as follows: If the constant C
(without loss of generality) is assumed to be 0, the parabola
must pass through the origin, the (0,0) point. L represents the
slope of a tangent through the origin. It is zero when the
minimum of the parabola is located at the origin. It becomes
negative when the minimum of the parabola moves to a
positive value along the abscissa, but since the left branch of
the parabola must still pass through the origin, this implies a
negative value along the ordinate for the minimum.
lIthe correlations varied systematically across the six instances of
MG2: The highest correlations (0.91-0.93) were obtained for
bars 1-2 and 17-18; lower correlations (0.85-0.86) hold for bars
9-10 and 13-14; and the lowest correlations (both 0.79) were
found for bars 5-6 and 21-22. The less tight relationship here
must be due to variations in global tempo among the pianists,
which may be only weakly related to the degree of tempo
modulation in MG2.
12For that matter, the fermata chord in bar 22, which is clearly
notated as reqUiring simultaneity of all tones, was not played
by all pianists in that way: DAV and NOV played a grand
arpeggio, NEY played a partial arpeggiO (two grace notes in the
left hand, as in bars 2, 6, and 18), and ASH played only the
lowest tone in advance. These variants may have been
occasioned by small hands (note that three of the pianists are
female), for the chord requires a large span.
l30ne detail skipped in this analysis is the timing of the melody
grace note during the last 101 of MG5b (position 8-4). Its onset
generally occurred near the middle (40-60%) of the 101. A few
pianists played it a little earlier (CAP, C02) or later (ARG, ESC,
SCH, H02, ARR). Two artists (C03, DAV) omitted it
altogether.
14In a final display of eccentricity, BUN produced a terminal 101
of nearly 5 seconds duration, with ORT not far behind.
15The data matrix is available from the author.