Journal of Memory and Language 45, 526–545 (2001)

doi:10.1006/jmla.2000.2780, available online at http://www.academicpress.com on

Episodic Memory for Musical Prosody

Caroline Palmer and Melissa K. Jungers

Ohio State University

and

Peter W. Jusczyk

Johns Hopkins University

Prosodic cues are often informative in speech perception; similar acoustic features distinguish music perform-
ances. Three experiments addressed the role of prosodic cues in memory for music. In Experiment 1, musically
trained and untrained listeners were familiarized with performances of short musical excerpts and later heard the
familiarized performances as well as novel performances of the same music. All listeners identified correctly the
familiarized performances. In Experiment 2, 10-month-old infants were familiarized with the same performances.
In a head-turn preference procedure, infants oriented longer to the familiarized performances than to the novel
performances. In Experiment 3, musically experienced listeners identified familiarized excerpts placed in differ-
ent melodic contexts; identification was more accurate for excerpts whose prosodic cues (intensity and articula-
tion) conflicted with the structure of the melodic context. These findings support episodic memory for music that
incorporates stimulus-specific acoustic features as well as abstract structural features. © 2001 Academic Press
Key Words: episodic memory; music perception; prosody.

When listeners recall a familiar musical tune,
do they remember the acoustic features of a par-
ticular performance, or only an abstract struc-
tural pattern of pitches and durations? This
question is similar to the issue of talker identity
in speech recognition: whether listeners’ mem-
ory for and identification of lexical items incor-
porates specific nonlexical, prosodic features of
an utterance. Prosody can refer both to an ab-
stract level of phonological structure and to its
acoustic realization in speech (Cutler, Dahan, &
Donselaar, 1997; Shattuck-Hufnagel & Turk,
1996). Prosodic cues, including acoustic varia-
tions in fundamental frequency, spectral infor-
This research was supported in part by NIMH grant
MH45764 to the first author, and by NICHD grant HD-
15795 and a Research Scientist AWARD from NIMH (MH-
01490) to the third author. We gratefully acknowledge the
assistance of Grant Baldwin, Zeb Highben, Ann Marie
Jusczyk, and Kristina Krasnov in conducting the experi-
ments, and Rosalee Meyer, Peter Pfordresher, and Diana
Raffman for comments on an earlier draft.
Address correspondence and reprint requests to Caroline
Palmer, Psychology Dept., Ohio State University, 1885 Neil
Ave., Columbus, OH 43210. E-mail: palmer.1@osu.edu.
mation, amplitude, and relative durations of
speech, are said to characterize the uniqueness
of each utterance (Speer, Crowder, & Thomas,
1993). The same acoustic features differentiate
one music performance from another; variations
in frequency, timbre, amplitude, and relative du-
rations, often termed performance “expression,”
form the microstructure of a music performance
and differentiate it from another performance of
the same music (cf. Palmer, 1997). Performance
expression can aid listeners in differentiating
among structural ambiguities (such as phrasal
and metrical boundaries) that arise in music
(Palmer, 1996; Sloboda, 1985), just as prosodic
features of speech can clarify the intended
meaning of a syntactically ambiguous sentence
(Lehiste, Olive, & Streeter, 1976; Price, Osten-
dorf, Shattuck-Hufnagel, & Fong, 1991; Scott,
1982). Thus, prosodic features may play similar
roles in aiding segmentation processes in music
and speech. We investigate here whether musi-
cal “prosody,” or expression that characterizes
particular music performances, plays a similar
function in music perception and recognition as
in speech.

526 0749-596X/01 $35.00

Copyright © 2001 by Academic Press
All rights of reproduction in any form reserved.
 EPISODIC MEMORY FOR MUSIC
Although the structure of music and speech
differ, the perceptual issues that researchers ad-
dress in both domains are similar: specifically,
to account for the perceptual constancy that lis-
teners experience in the presence of a range of
physical change. For example, as early as in-
fancy, listeners can recognize a melody as the
same when it is transposed to a different set of
frequencies, performed at a different rate or
tempo, or performed in a different instrumental
timbre or “voice” (Chang & Trehub, 1977; Tre-
hub & Thorpe, 1989). Even music whose pitch
or duration content has changed, as in varia-
tions on musical themes, permits perceptual con-
stancy (Dowling & Harwood, 1986; Large,
Palmer, & Pollack, 1995; Serafine, Glassman &
Overbeeke, 1989). Despite the many types of
stimulus variability that music displays across
performers, rates, timbres, and contexts, music
perception is robust and adaptable over a range
of physical changes in the acoustic signal.
Consistent with perceptual constancy, early
studies of music perception took a view similar
to that of speaker normalization: stimulus vari-
ability (including differences across music per-
formances) was treated as noise, to be filtered
out in an abstract memory representation (cf.
Large, Palmer, & Pollack, 1995). Normaliza-
tion refers to the process of converting or trans-
forming physically different tokens into some
common representational format that is the ulti-
mate goal of perceptual analysis, and these
standardized representations are stored in mem-
ory (Pisoni, 1997). Normalization views led to
the search for an idealized symbolic representa-
tion for spoken language in which stimulus
variability is treated as noise; the result of nor-
malization processes was a reduction in stimu-
lus variability and often a loss of prosodic
information. Linguistic properties of speech
that carry the intended message were consid-
ered separate from indexical or extralinguistic
(nonstructural) features such as the particular
talker’s voice, gender, or dialect (Gerstman,
1968; Shankweiler, Strange, & Verbrugge,
1977; Studdert-Kennedy, 1974). Just as this
theoretical perspective guided speech re-
searchers’ choices of stimuli that reduced or
eliminated variability on unwanted dimensions,
527
music perception researchers concentrated on
normalized musical standards that reduced
stimulus variability in perceptual experiments
(cf. Dowling & Harwood, 1986).
More recently, prosodic cues have been con-
sidered a source of information in speech that
aids lexical recognition, rather than as stimulus
variation to be filtered out. In this view, extralin-
guistic properties, such as gender, dialect, and
speaking rate, can help listeners identify linguis-
tic categories (Nygaard, Sommers, & Pisoni,
1994). Talker-specific characteristics of speech
can provide information about the talker’s gen-
der, identity, or age (Fellowes, Remez, & Rubin,
1997; Remez, Fellowes, & Rubin, 1997). Talker
and rate information also influence listeners’
memory for speech; previously heard words are
more accurately recognized when the produc-
tion rate or talker’s voice is retained (Bradlow,
Nygaard, & Pisoni, 1999; Palmeri et al., 1993).
In addition, previously presented sentences are
recognized more accurately when the same
prosodic cues are retained, and prosodic cues
aid listeners’ memory for syntactically ambigu-
ous sentences (Speer, Crowder, & Thomas,
1993). These findings suggest that highly de-
tailed nonlexical features of the acoustic signal
are encoded in memory representations along
with lexical content (Goldinger, 1997; Jusczyk,
1997; Pisoni, 1997).
Patterns of stimulus variability in music per-
formances suggest that performance expression
is systematic and intentional, and likely to be
perceptually informative. Musicians can repli-
cate expressive patterns of timing and dynamics
(correlated with intensity) across their perform-
ances with high precision (Henderson, 1936;
Seashore, 1938), and attempts to play mechani-
cally or without expression significantly
dampen these expressive features (Bengtsson &
Gabrielsson, 1983; Palmer, 1989). Performers’
use of prosodic cues to mark musical units such
as metrical and phrase structure increases with
short-term practice and overall experience
(Palmer, 1996b; Sloboda, 1983). Expression in
music performance can be systematically af-
fected by both structural dimensions (harmony,
melody, rhythm, meter, etc.) and nonstructural
dimensions (affect, tempo, other interpretive de-
528 PALMER, JUNGERS, AND JUSCZYK
cisions), and it is often difficult to separate the
two. Attempts to formalize a rule-based syntax
of the relationship between intended musical
structure and the resulting performance expres-
sion, called the “structure–expression map-
ping,” have met with some success (Clynes,
1986; Sundberg, Askenfelt, & Fryden, 1983;
Todd, 1985), as measured by listeners’ judg-
ments of computationally generated simulations
of performances (Clynes, 1995; Thompson,
1989) and comparisons with human perform-
ances (Repp, 1989). However, the results sug-
gest that both piece- and performance-specific
factors influence performance expression as
much as the structural features modeled by the
simulations. The experiments reported here test
whether prosodic features of music performance
encode contextual information that listeners can
use in memory tasks.
Fewer studies have tested which types of
stimulus variability in music performance are
perceptually informative. Musically trained lis-
teners are able to identify performers’ intended
structure (such as metrical and melodic struc-
ture) on the basis of intensity, articulation (inter-
stimulus intervals between musical tones), and
timing cues (Nakamura, 1987; Palmer, 1996b;
Sloboda, 1985). For example, musically trained
listeners can correctly identify the intended
phrase structure in piano performances more
often when acoustically varying features of in-
teronset intervals and intensities are present
than when they are normalized in the perform-
ances (Palmer, 1996a). Listeners are also accu-
rate at identifying performances of the same
music intended as musically expressive (nor-
mal), mechanical (deadpan), or exaggerated
(Kendall & Carterette, 1990). In addition, listen-
ers can correctly identify a performer’s intended
emotion on the basis of prosodic features; for
example, a slow performance tempo often sig-
nals a sad emotional state (Gabrielsson & Juslin,
1996). The relative importance of different
prosodic cues is unknown; only recently have
systematic analyses of relations among sources
of performance variability been undertaken
(Juslin, 1997).
These perceptual studies rely heavily on
musically trained listeners, perhaps reflecting
an implicit assumption that only through exten-
sive training can one learn the relationship be-
tween intended structure and acoustic variation
in performance. This contrasts with a common
view in speech that prosodic features provide a
low-level cue to aid segmentation for even in-
experienced listeners (Cooper & Sorensen,
1977; Lehiste, 1970; Streeter, 1978). Research
in infant perception also suggests that musical
expertise is not necessary for perception of
prosodic features. Infants show preferences for
infant-directed songs such as lullabies, which
show heightened cues such as higher pitch,
slower tempo, distinctive timbre, and changes
in fundamental frequency (Trainor, Clark,
Huntley, & Adams, 1997; Trehub & Trainor,
1998), over non-infant-directed songs (Trainor,
1996). Four-month-old infants prefer listening
to music with pauses inserted at phrase bound-
aries over the same music with pauses inserted
in the middle of phrases (Krumhansl &
Jusczyk, 1990). Further experiments confirmed
that drops in pitch contour and duration length-
ening are acoustic variables that can mark unit
or phrase boundaries for infants (Jusczyk &
Krumhansl, 1993; Trainor & Adams, 2000), the
same prosodic markers of clausal units that in-
fants respond to in speech (Hirsh-Pasek et al.,
1987). One of the goals of this paper is to ad-
dress whether musical training (in adults) or
experience (in infants) is a prerequisite for
prosodic features to be incorporated in memory
for music.
Although there is little study of infants’ mem-
ory for music, some studies suggest that
prosodic features can be retained in infants’
memory for speech. Two-month-old infants can
use the prosody of a sentence to organize and re-
member words (Mandel, Jusczyk, & Kemler-
Nelson, 1994). Six-month-old infants are better
able to locate prosodically well-formed speech
units than ill-formed units in speech passages,
suggesting their ability to use prosodic cues to
parse speech is operative at an early age (Nazzi,
Kemler Nelson, Jusczyk, & Jusczyk, 2000;
Soderstrom, Jusczyk, & Kemler Nelson, 2000).
Moreover, when they begin to segment words,
infants appear to rely heavily on prosodic cues
to word boundaries (Jusczyk, Houston, & New-
 EPISODIC MEMORY FOR MUSIC
some, 1999). Infants also appear to be storing
information about what they hear; nine-month-
old infants can retain words over a 2-week pe-
riod that were presented in sentences (Jusczyk
& Hohne, 1997). Thus, long-term memory for
word units is operative by this age. One reason
that a perceptual system might preserve fine de-
tails about speech would be to allow early learn-
ers to accurately imitate and produce patterns
heard in their language environment (Studdert-
Kennedy, 1983). We test here the hypothesis
that listeners as early as infancy (10 months old)
have the ability to remember and later recognize
music performances on the basis of particular
prosodic features.
It is possible that prosodic cues aid music
segmentation but are not retained in memory
once segmentation processes are complete.
Raffman (1993) proposes an explanation for
why prosodic cues in music can be discrimi-
nated but not recognized or remembered later.
In her view, expressive nuances are features of
the surface of a performance; they can be influ-
ential in the formation of a structural representa-
tion, but they are not retained as part of it. Based
on findings such as Siegel and Siegel’s (1977),
that musically trained listeners cannot accu-
rately identify out-of-tune pitch intervals (inter-
vals that fall between well-learned pitch cate-
gories in equal temperament scales), Raffman
proposes that only information at a categorical
level (such as notated pitches and durations) can
be encoded in a lasting memory representation.
She argues that expressive nuances in music are
subcategorical: below the level at which listen-
ers possess schemas. Because a primary func-
tion of schemas is to reduce the information
load imposed by the expressive nuances, there
would be no advantage (and likely a memory
capacity disadvantage) to acquiring schemas
whose resolutions were as fine as that of dis-
crimination of prosodic cues. Although listeners
can discriminate among musical prosodic fea-
tures over short time periods, they cannot retain
a lasting memory for such features over longer
time periods in this view.
We address listeners’ retention of prosodic
features in music performance in three ex-
periments. The first experiment investigates
529
whether musically trained and untrained adults
listeners can learn and later identify short music
performances on the basis of their subcategori-
cal acoustic cues. The second experiment ad-
dresses the same question with 10-month-old
infants; using the Headturn Preference Proce-
dure (Kemler Nelson et al., 1995), we consider
whether musical acculturation is required for
listeners to remember particular prosodic fea-
tures of music. The final experiment examines
whether prosodic cues are useful during seg-
mentation, specifically, in the identification of
musical units embedded in larger musical con-
texts. The contextual relationship between the
prosodic cues of the embedded musical unit and
the larger melodic context is altered, by present-
ing the same unit embedded in different
melodies. Together, these experiments provide
evidence as to whether prosodic features of
music performances can be stored in memory
and later identified by a variety of listeners.
EXPERIMENT 1: MUSICIANS’ AND
NONMUSICIANS’ MEMORY FOR
PROSODIC CUES
We first examine whether adult listeners can
identify specific performances of a short musi-
cal sequence they heard earlier on the basis of
prosodic cues. The experiment contains a famil-
iarization stage and a subsequent test stage, to
parallel as closely as possible the infant para-
digm of Experiment 2. Listeners were familiar-
ized with performances of short melodic se-
quences, and later were presented with both
familiar sequences and novel performances of
the same melodies: sequences that contained the
same melodic and rhythmic structure (same cat-
egorical pitch/duration pattern as notated in a
musical score) but different prosodic cues. The
particular prosodic cues in piano performances
included intensity, interonset interval, and artic-
ulation cues. If listeners can identify the se-
quences heard at familiarization, that would in-
dicate that listeners are using cues other than the
melodic (pitch) or rhythmic (duration) structure
of the performances: specifically, that listeners
are capable of using the subcategorical acoustic
cues to later identify a particular sequence of
musical events.
530 PALMER, JUNGERS, AND JUSCZYK
We also examine whether the ability to iden-
tify subcategorical acoustic cues requires musi-
cal training. Musically trained listeners are com-
pared with listeners who had no musical
training. If both groups can identify perform-
ances on the basis of prosodic cues, this would
suggest a stronger analogy with the speech liter-
ature, in that a wide range of listeners of varying
musical experience show sensitivity to the same
acoustic features of auditory events. If only mu-
sically trained listeners are able to remember
and later identify music on the basis of these
cues, it would suggest that sensitivity to acoustic
cues is acquired with expertise. This finding
would limit an analogy to how the same
acoustic cues are perceived in speech, because
both nonmusician and musician (adult) listeners
display advanced speech abilities.
Method
Participants. Twenty-four musically trained
listeners and 16 listeners without musical train-
ing from the Ohio State University community
participated in this study. Musician listeners
had between 4 and 13 years of private instruc-
tion on a musical instrument (mean ⫽ 8.1
years). Nonmusician listeners had less than one
year of musical training (mean ⫽ 1 month).
None of the listeners reported any hearing prob-
lems. Listeners received either course credit in
an introductory psychology course or a nominal
sum for their participation.
Apparatus. The stimuli were created on
a computer-monitored Yamaha Disklavier
MX100 acoustic upright piano. Optical sensors
detected keypress velocities without affecting
the touch or sound of the acoustic piano. The
timing resolution was 2 ms for note events,
with precision (measured by the standard devi-
ation of interonset intervals during recording)
within 0.8% for interonset intervals in the range
of the performances. The pitch, timing, and
hammer velocity values (correlated with inten-
sity) for each note event were recorded on the
computer. The musical stimuli used in the ex-
periment were first recorded on the piano and
later played back on the same piano and
recorded to compact disc at a 44.1 k sampling
rate for use in the perceptual experiments. All
stimuli were sounded over JBL Studio Monitor
4410 speakers placed around the piano in the
same room and at the same decibel level
(within 3 dB SPL) at which they were origi-
nally recorded on the piano. The sound level
averaged 75 dB SPL, about 25 dB above ambi-
ent noise levels.
Materials. Two performances each of two dif-
ferent five-note melodic sequences, referred to
as “short sequences,” served as stimulus materi-
als. Each short sequence appeared in musical
notation, embedded in two different melodic
contexts: one was a melody in 3/4 time signa-
ture (a ternary meter, with metrical accents on
every third beat) and one was in 4/4 time signa-
ture (a binary meter, with metrical accents on
every second and fourth beats). The melodic
contexts were designed to elicit changes in the
relative importance of note events in the short
sequences, including (but not limited to) the lo-
cation of strong and weak metrical accents. One
of the five-note sequences and its two melodic
contexts are shown in Fig. 1; only the short se-
quences, marked A⬘ and B⬘ in Fig. 1, served as
stimuli in this experiment. The melodic con-
texts, composed by a musician for the experi-
ment, had the following properties: each
melodic context for a given short sequence was
in the same key (C-Major or G-Major) and in
the same frequency range (female vocal range),
and each was 13–15 beats long (each melody
ended on a strong metrical beat). The first tone
of each short sequence began in the middle of a
melody on a strong metrical beat (indicated by
the metrical barline or vertical line preceding
each bracketed excerpt in Fig. 1), and the event
immediately preceding the first tone of the short
sequence was the same in notated pitch and
duration across the two melodic contexts. Ex-
cept for the short sequences embedded within
them, the melodic contexts were composed to
be as different in pitch contour and duration pat-
tern as possible, again, to elicit changes in the
relative importance of note events in the short
sequences.
Piano performances of the entire melodic
contexts were collected from an experienced
pianist, who had 30 years of performing experi-
ence and 18 years of private instruction. The
 EPISODIC MEMORY FOR MUSIC 531

FIG. 1. Example of a short musical sequence, indicated in brackets, and its two melodic contexts, A and B,
used in the experiments. Intensity (hammer velocity units) and articulation values (in ms) for the short sequence
appear below the musical notation.

pianist was instructed to perform the melodies
in an exaggerated fashion, “as if playing to a
child,” to mimic instructions to speakers in in-
fant studies of speech perception (Nazzi et al.,
2000). The performances were recorded to a
metronome set to 152 beats per minute (quarter-
note ⫽ beat), or 394 ms/quarter-note, to ensure
similar tempi across performances. When asked
afterward, the pianist reported not being aware
that an identical short sequence was embedded
532 PALMER, JUNGERS, AND JUSCZYK
in each melody pair. The MIDI event informa-
tion (pitch, interonset interval, and hammer
velocity values (correlated with intensity) as
recorded by the optical sensors in the piano) de-
noting the short sequences was excised from the
melodic contexts, beginning at the onset time of
the first tone, up to the offset time of the last
tone.
Analyses were conducted on the interonset
interval, intensity, and articulation values in the
performances. Analyses of each event’s interon-
set (IOI) values indicated that the mean quarter-
note interval was 394 ms (equivalent to the indi-
cated metronomic value of 394 ms), and there
were no significant differences in mean quarter-
note IOIs by stimulus melody (4), meter (3/4 or
4/4), or beats within each measure (1–3, the
maximum number of beats that can be com-
pared across all melodies). Thus, interonset val-
ues did not differ significantly from the dura-
tional categories notated in the score, and there
were no differences in tempo (largely because
of the presence of the metronome). Analyses of
the hammer velocity (intensity) values in arbi-
trary MIDI units which range from 0 (silent) to
127 (loudest value) indicated that one of the
four performances was louder than the others,
F (3,36) ⫽ 3.29, p ⬍ .05. To avoid stimulus dif-
ferences in absolute intensity level, the MIDI
hammer velocity values of this melody were
lowered by a constant value so that the mean of
the revised values was equal to the mean value
of the remaining three short sequences, but the
melody retained its note-by-note intensity dif-
ferences. Analyses of the adjusted intensities
indicated that intensities were higher on metri-
cally strong beats; as shown in Fig. 1, perform-
ances of melodies in 3/4 meter had higher inten-
sities only on beat 1 (strong metrical accent),
whereas performances of melodies in 4/4 meter
had higher intensities on beats 1 and 3 (strong
metrical accents) (Dunn-Bonferroni, p ⬍ .05).
Event intensities across the short sequences
were correlated with music-theoretic predic-
tions of metrical accent strength (Lerdahl &
Jackendoff, 1983), based on the number of ac-
cents in a metrical grid predicted by the time
signature that aligned with each note event (see
Palmer & Krumhansl, 1987, for further details).
These correlations were significant, r ⫽ .72,
p ⬍ .01, confirming that intensity cues in these
performances increased for events at strong
metrical accents.
Articulation values were measured by the
offset time of the current event (key release
time) minus the onset time of the next event
(keypress) in ms, as in previous studies of piano
performance (Palmer, 1989, 1996b; Sloboda,
1983); a positive value denotes a smooth or
“legato” style, whereas a negative value denotes
a gap or a “staccato” style. Articulation values
differed across performances, F (3,36) ⫽ 33.3,
p ⬍ .01, and beats, F (2,36) ⫽ 10.8, p ⬍ .01, re-
flecting primarily a melody by beat interaction,
F (6,36) ⫽ 6.8, p ⬍ .01; as shown in Fig. 1,
legato/staccato patterns were unique within
each performance. There was no correlation of
articulation values with music-theoretic predic-
tions of metrical accent strength, and articula-
tion values did not correlate with intensity val-
ues in the short sequences or in the entire
melodies (p ⬎ .10). Thus, the short sequences
associated with each melody were identical in
terms of the pitch and duration categories in the
musical score, but each performance had differ-
ent patterns of intensity and articulation cues,
which reflected in part the metrical structure
and the specific melodic context, respectively.
Design and Procedure. In the familiarization
stage of the experiment, listeners heard two of
the four short sequences, the ternary-meter (3/4)
performance of one sequence and the binary-
meter (4/4) performance of the other sequence,
to ensure that meter was not confounded with
familiarization. Which short sequences were
present at familiarization and the order in which
sequences were presented were counterbalanced
across subjects. In the test stage of the experi-
ment, listeners heard all four short sequences.
Each block of trials contained the four short se-
quences presented in random order, and there
were 9 blocks of test trials, yielding a total of 36
trials. The independent variable of whether each
sequence in the test trials was presented at fa-
miliarization was a within-subjects variable.
At the beginning of the experiment, listeners
were instructed that they would hear different
performances of the same melody during the ex-
periment and they would be tested to see if they
could recognize different performances. During
 EPISODIC MEMORY FOR MUSIC
the familiarization stage of the experiment, one
short sequence was sounded 12 times, separated
by 750 ms of silence, followed by the second
short sequence sounded 12 times. Listeners
were told to listen carefully to the familiariza-
tion trials because they would be asked to recog-
nize those particular performances later. They
were also told they would hear a number of per-
formances afterward, some of which were from
the familiarization stage.
During the test stage of the experiment, a
short sequence was sounded twice in a row (with
1 s of silence between the two hearings) on each
trial. Listeners were instructed to respond “yes”
or “no” as to whether the test trial was identical
to one of the sequences they had heard during
the familiarization stage. Listeners were told
that some of the test sequences would have the
same notes as the familiarization sequences, but
they would not be identical because they were
from a different performance. Listeners were
also asked to mark their degree of confidence in
their answer on a three-point scale (1 ⫽ not con-
fident, 3 ⫽ confident) for the test trials. There
were 5 s of silence between test trials.
Nonmusician listeners were given additional
instructions before the experiment, to clarify
the nature of the task. These listeners were told
they would hear two different performances of
the same musical piece, which the experi-
menter then sounded for them. Then they were
told they would hear two identical perform-
ances of the same musical piece, which the ex-
perimenter then played. The reason for the ad-
ditional instructions was to help listeners
distinguish the goal of recognizing the same
performance (same prosodic cues), the aim of
this experiment, from the separate goal of rec-
ognizing the same melody (same pitches and
durations but different prosodic cues). The mu-
sical examples used in the additional instruc-
tions were from a similarly constructed musical
sequence that was not included in any of the
experiments.
Results and Discussion
The sequences that were heard at familiariza-
tion, the random order of test trials, and trial
block were not significant factors in listeners’
identification responses, so percentages of iden-
533
tification (“yes”) responses across trial blocks
were computed for each listener. A two-way
analysis of variance (ANOVA) on listeners’ per-
centages of identification responses by familiar-
ization condition (was test sequence familiar-
ized/not familiarized) and musical training
(musician/nonmusician) yielded a significant ef-
fect of familiarization, F (1,38) ⫽ 78.0, p ⬍ .01.
There were no effects of training or interactions
with familiarization. As shown in Fig. 2, both
groups of listeners were able to identify the par-
ticular performances with which they had been
familiarized, when presented among other per-
formances of the same pitch and rhythmic (du-
rational pattern) content; mean responses for
each condition shown in Fig. 2 differed signifi-
cantly from 50%, the chance estimate (p ⬍ .01).
The d-prime values were computed to verify
that musicians did not have greater sensitivity
than nonmusicians; musicians’ (d⬘ ⫽ 1.73) and
nonmusicians’ mean d-prime values (d⬘ ⫽ 1.58)
were both greater than zero (p ⬍ .01) and did
not differ significantly from each other (non-
parametric Mann-Whitney U ⫽ 218, p ⬎ .10).
The confidence scale ratings did not differ by fa-
miliarization or musical training.
These results indicate that listeners are able to
encode prosodic cues associated with a particu-
lar music performance and use them later to
identify familiar performances from a set of per-
formances of the same music. Acoustic features
of intensity and articulation were likely to have
influenced listeners’ identification judgments;
interonset intervals did not differ from the cate-
gorical notated durations in these performances,
due to the presence of a metronome. In addition,
musical training was not necessary to perform
this task; listeners with no musical training per-
formed as well as listeners with musical train-
ing. Thus, prosodic features in music perform-
ance were useful for a wide range of listeners
and could be encoded and later used to identify
particular musical sequences.
EXPERIMENT 2: ROLE OF MUSICAL
ACCULTURATION
Musical training does not appear to be neces-
sary for identification of musical sequences
based on their prosodic cues. However, identifi-
cation may require musical acculturation:
534 PALMER, JUNGERS, AND JUSCZYK

FIG. 2. Experiment 1: mean percentage of yes responses for musically trained listeners (black bars) and un-
trained listeners (white bars) by familiarization condition (was test sequence familiarized/not familiarized), with
standard error bars.

knowledge of performance styles and genres
that listeners acquire passively (without explicit
training) over years of exposure to musical
forms. If, however, the capacity to store and re-
member acoustic features specific to music per-
formances is available as early as the first year
of life, then even infants, who do not have years
of passive exposure to music, may be capable of
identifying performances they have heard be-
fore. Indeed, evidence from speech perception
studies suggests that infants store indexical
properties (such as talker voice information)
along with the phonetic properties of utter-
ances. Houston and Jusczyk (2000) found that
7.5-month-olds are affected by talker voice
characteristics in recognizing words in fluent
speech, suggesting that such information is in-
cluded in their representations of the sound pat-
terns of words. Moreover, Jusczyk, Hohne,
Jusczyk, & Redanz (1993) found evidence that
infants remember the voice characteristics of an
unfamiliar talker reading stories for as long as 2
weeks. Thus, infants appeared to encode not
only information about what was said but also
information about how it was said. Might they
also show the same abilities for music perform-
ances?
We use the Headturn Preference Procedure
with 10-month-old infants to test their ability to
identify performances heard earlier on the basis
of prosodic features. The procedure takes ad-
vantage of the finding that, when varied stimuli
are used, preferences among infants over 4.5
months of age tend to be correlated with famil-
iarity. Specifically, the amount of time spent lis-
tening to stimulus items, a measure of prefer-
ence, often correlates with amount of familiarity
with those items (Jusczyk, 1998; Kemler Nelson
et al., 1995). In this experiment, infant listeners
were familiarized with the same short musical
sequences as used in Experiment 1, and the in-
fants were presented with all performances (fa-
miliar and novel) at test. The amount of time
spent looking (as measured by orienting toward
the loudspeaker over which the test sequence
was sounded) provides a measure of familiarity
with the stimulus. If prosodic cues in music per-
formance are remembered and later used for
identification at as early as 10 months of age,
then infant listeners should exhibit larger orien-
 EPISODIC MEMORY FOR MUSIC
tation times for familiar sequences than for
novel sequences.
Method
Participants. Sixteen 10-month-old infants
(mean age ⫽ 44 weeks, 4 days; range, 43 weeks,
3 days to 46 weeks, 2 days) were recruited from
the Baltimore, MD, community to participate in
the study. Two additional infants were tested but
not included because of restlessness (1) and
parental interference (1).
Materials. The same musical stimuli were
used in this study as in Experiment 1.
Apparatus. A Macintosh Centris 650 com-
puter controlled the presentation of the stimuli
and recorded the observer’s coding of the in-
fants’ responses. The stimuli were stored in dig-
itized form (at a 20 kHz sampling rate) on the
computer. A 16-bit D/A converter was used to
recreate the audio signals, which were sounded
at an amplitude of 72 ⫾ 2 dB SP, approximately
20 dB above the ambient noise level.
The stimuli were played out through an-
tialiasing filters and a Kenwood audio amplifier
(KA 5700) over Cambridge Soundworks loud-
speakers mounted behind the side walls of the
testing booth. A red light was mounted in view
on each of the side walls and a green light was
mounted on the center panel of the testing
booth, all approximately at the seated infant’s
eye level. Directly below the center light a 5-cm
hole accommodated the lens of a video camera
used to record each test session. A white curtain
suspended around the top of the booth shielded
the infant’s view of the rest of the room. A com-
puter terminal and response box were located
behind the center panel, outside the infant’s
view. The response box, connected to the com-
puter, was equipped with buttons that were used
to start and stop the flashing center and side
lights and record the direction and duration of
headturns. Information about the direction and
duration of headturns and the total trial duration
was stored in a data file on the computer.
Design and procedure. The experimental de-
sign was as similar as possible to that in Experi-
ment 1. Two of the four short sequences were
presented to infants during the familiarization
stage, one from each meter; which two of the
535
four test sequences were presented at familiar-
ization was counterbanced across infants. Dur-
ing the test stage, infants heard all four test se-
quences; the independent variable of whether
each test sequence was presented at familiariza-
tion is a within-subjects variable. The dependent
variable is the amount of time spent listening
(with head oriented toward the sound source) to
each test sequence, summed across test trials.
A modified version of the Headturn Prefer-
ence Procedure (Kemler Nelson et al., 1995)
was used to measure orientation times for test
sequences. The infant sat in the middle of the
testing room on the lap of the caretaker; an ob-
server hidden behind the center panel watched
the infant through a peephole and recorded the
direction and duration of the infant’s headturns
using a response box. The observer and the care-
giver wore earplugs and listened to masking
music over tight-fitting Peltor Aviation-7050
headphones. Each trial was as follows: the cen-
ter green light flashed, and when the infant ori-
ented toward it, an observer pushed a button to
extinguish the center light and to start flashing
one of the red side lights above a loudspeaker
(the program controls randomization of the side
lights). When the infant made a headturn of at
least 30 ° in the direction of the speaker, the ob-
server pushed another button to initiate the
sounding of the sequence, which continued until
it ended or until the infant looked away for two
consecutive seconds. The observer pressed a
button whenever the infant looked away or re-
oriented toward the speaker. The computer pro-
gram kept track, based on the button presses, of
actual looking times to the side lights; total
looking time for a given trial excluded any time
the infant looked away. If the infant looked
away for more than 2 s, the side light was extin-
guished, the trial was interrupted and ended, and
the next trial began. The flashing red light re-
mained on for the duration of the trial.
During the familiarization stage, each infant
heard two of the four sequences. Each trial con-
sisted of one of the sequences repeated eight
times, yielding a total trial length of approxi-
mately 21 s. Each infant had to achieve 30 s of
accumulated listening time to each of the famil-
iarization sequences in order to progress to the
536 PALMER, JUNGERS, AND JUSCZYK
test stage, and trials containing the two familiar-
ization stimuli alternated until the infant had
achieved this threshold. If the infant met the 30-
s criterion for one sequence before the other, the
two trial types (containing the different se-
quences) continued to alternate until the crite-
rion was met for both. The presentation of each
stimulus was assigned randomly to a particular
loudspeaker on each trial.
During the test stage, each infant heard all
four sequences on different trials. There were 3
blocks of 4 trials each, yielding a total of 12 tri-
als. Each of the four sequences was sounded
within each block, and trials within a block were
randomly ordered. Each trial was created the
same as in the familiarization stage, and trial
onset and offset times were determined exactly
as in the familiarization phase. The dependent
variable was each infant’s total orientation time
for each sequence summed across the test trials,
which represented the time spent looking at the
light during the sounding of each sequence, with
time looking away excluded.
Results and Discussion
A one-way ANOVA on infants’ mean listen-
ing times by familiarization condition (was test
sequence familiarized/not familiarized) yielded
a significant effect of familiarization, F (1,15) ⫽
7.77, p ⬍ .05. As shown in Fig. 3, infants lis-
tened longer to the particular performances with
which they had been familiarized, when pre-
sented among other performances with identical
FIG. 3. Experiment 2: infants’ mean orientation times by familiarization condition (was test sequence famil-
iarized/not familiarized), with standard error bars.
pitch/duration content. Thirteen of the 16 in-
fants had longer mean listening times for the test
sequences heard at familiarization than for the
novel test sequences. Neither the sequences in-
cluded at familiarization nor the random order
in which subjects heard trials was a significant
factor.
These results indicate that further musical ac-
culturation beyond 10 months of age is not re-
quired for listeners to use prosodic features in
identifying particular music performances. This
result is consistent with exemplar views of lan-
guage acquisition that consider prosodic fea-
tures as important not only to the identification
of spoken words but also to the formation of
memories that distinguish one unit or word from
another (Houston & Jusczyk, 2000; Jusczyk,
1993, 1997). As in speech, musical prosody car-
ries information that may help to identify a par-
ticular performance.
What purpose could episodic memory for
music performance serve? It seems likely that
prosodic features aid segmentation of musical
sequences into meaningful units. To test whether
prosodic features serve to mark units in a contin-
uous stream of music, the short melodies were
embedded in the next experiment in computer-
generated musical contexts that either matched
or mismatched the features of the short se-
quences. When the short sequences are embed-
ded in a larger melodic context, their prosodic
features may be more salient when their struc-
tural correlates do not match the structure of the
 EPISODIC MEMORY FOR MUSIC
context. We test this possibility in the following
experiment, as an example of the role that
prosodic features play in music perception.
EXPERIMENT 3: ROLE OF MELODIC
CONTEXT
The first two experiments indicate that listen-
ers of a wide range of ages and musical experi-
ence were able to encode prosodic features of
short musical sequences (shorter than most
melodies), which were chosen to allow compar-
isons among infants and adults. Are these
prosodic cues useful in larger musical contexts?
Findings in speech perception suggest that lis-
teners as young as 6 months old can use
prosodic information to encode clausal units,
and later recognize those units in new contexts
(Nazzi et al., 2000; Soderstrom et al., 2000). We
test a similar segmentation issue in this study:
whether listeners can identify the short se-
quences embedded in larger melodic contexts,
on the basis of their prosodic features.
The perceptual salience of musical events de-
pends largely on the regularities of the context
in which they appear. Western tonal music con-
tains a complex system of structural regularities;
both music-theoretic and psychological ap-
proaches suggest that listeners generate expecta-
tions for future musical events based on those
regularities, that influence the perception of mu-
sical events (Jones & Boltz, 1989; Meyer, 1956;
Narmour, 1990; Palmer & Krumhansl, 1990).
One of the most regular dimensions of Western
tonal music is its meter, an alternating pattern of
strong and weak beats; metrical regularity influ-
ences listeners’ perception of and memory for
music (Palmer & Krumhansl, 1990; Yee,
Holleran, & Jones, 1994). We take advantage of
this fact in Experiment 3 by placing each short
sequence in two different metrically regular
contexts, one that matches and one that mis-
matches the prosodic cues in the short sequence.
We hypothesize that performances whose
prosodic features do not match the metrically
regular context may be more salient perceptu-
ally (when expectations are inconsistent) than
those performances whose features do match the
metrical context (when expectations are consis-
tent). For example, the short sequences that
537
were originally performed in a 3/4 metrical con-
text may be more salient when they are embed-
ded in a 4/4 context than when they are embed-
ded in a 3/4 context. In addition, we hypothesize
that listeners will identify all performances
heard at familiarization (whether consistent or
inconsistent with the melodic context) more
often than performances not heard at familiar-
ization, based on the prior two experiments.
The experimental task is similar to that of the
previous experiments. Adult listeners are famil-
iarized with particular performances of the same
short sequences used in the first two experi-
ments; at test, they are presented with the se-
quences embedded in larger melodies and asked
to identify those performances they heard at fa-
miliarization. The melodic contexts are con-
structed by computer and contain constant (un-
varying) prosodic cues: all intensities (hammer
velocities), interonset intervals, and articulation
values are the same across all events in the
melodic context. Only the short sequences con-
tain prosodic cues. Because of the increased dif-
ficulty of the task (listeners must find the short
sequence in the melodic context and decide
whether its prosodic features match those heard
at familiarization), only musically trained listen-
ers were recruited for the experiment.
We also address the relationship between the
prosodic cues and their structural correlates. It is
possible, for example, that listeners remember
the metrical structure or other structural corre-
lates of meter and do not retain the subcategori-
cal acoustic cues, as Raffman (1993) suggests.
In order to claim that listeners encode perform-
ance-specific details of the excerpts, we address
the possibility that listeners could recognize the
familiarized short sequences based solely on an
abstract representation of metrical structure. For
this purpose, we include a control condition in
which a different group of musically trained lis-
teners attempted to identify the meter of each
short sequence (heard out of context), to test
whether an abstract representation of meter
could be formed from the short sequences alone.
Method
Participants. Sixty-four musically trained lis-
teners from the Columbus, Ohio, community
538 PALMER, JUNGERS, AND JUSCZYK
participated in this study. Listeners had between
4 and 15 years of musical training, with a mean
of 7.6 years of private lessons on a musical in-
strument. None of the listeners reported any
hearing problems. Listeners either received
course credit or were paid a nominal fee for
their participation. None of the listeners had
participated in the previous experiments.
Materials. The same sequences from Experi-
ment 1 were used in this experiment, as well as
the musical melodies from which they were
originally excised. Two different melodic con-
texts were created for each short sequence, a
context that matched the prosodic cues of the se-
quence (“matched context”) and a context that
did not match the prosodic cues of the sequence
(“mismatched context”). The matched context is
one in which both the short sequence and the
melodic context share the same meter (binary or
ternary); the mismatched context is one in
which the short sequence and melodic context
differ in meter.
The melodic contexts were generated from
computerized (mechanical) versions of both
melodies for each short sequence, in which no
prosodic cues were present (all event interonset
intervals and intensities were equivalent across
the melody). The intensities and interonset in-
tervals were set equal to the mean values in the
original performed melodies of Experiment 1,
from which the short sequences were taken.
When the short sequence was placed in the
matched context (the melody from which it
originated), the event onsets and offsets at the
boundaries were the same as in the original per-
formance. When the short sequence was placed
in the mismatched context (the alternate
melody), the event onsets and offsets at the short
sequence boundaries were aligned with the
times at which the original short sequence (as
performed in its original melodic context) began
and ended. That is, the bracketed excerpts A⬘
and B⬘ in Fig. 1 were exchanged between the
melodic contexts A and B. Thus, performances
A-A⬘-A and B-B⬘-B reflect matched contexts,
and A-B⬘-A and B- A⬘-B reflect mismatched
contexts.
Metrical control condition. To ascertain
whether listeners could form an abstract repre-
sentation of the metrical structure from the short
sequences alone, a control experiment was first
conducted in which 14 different listeners (not
included in the experiment) were asked to iden-
tify the meter of each excerpt. The musically
trained listeners (4–14 years of training, mean
of 6.8 years) were first given examples of
melodies in a binary meter (defined as a pattern
of alternating accents with a strong accent on
every second beat, as in 2/4 or 4/4 meter) and a
ternary meter (defined as a pattern of alternating
accents with a strong accent on every third beat,
as in 3/4 or 6/8 meter). These melodies were not
included in the control experiment. Then listen-
ers heard on each test trial two repetitions of a
short sequence from Experiment 1, and they in-
dicated on paper whether the short sequence
would fit or sound best in a binary or a ternary
meter. The control experiment included five rep-
etitions of each of the four short sequences from
Experiment 1 (two from a binary meter and two
from a ternary meter), totaling 20 trials. Listen-
ers’ percentage of correct responses across trials
indicated no significant differences from chance
(mean percentage correct ⫽ 54%, p ⬎ .70); re-
sponses were equally (in)accurate for sequences
that were originally performed in binary meter
(52% correct) and in ternary meter (55% cor-
rect). Thus, it is unlikely that listeners simply
formed an explicit abstract representation of
meter for the short sequences.
Design and procedure. The familiarization
stage was identical to that of Experiment 1.
During familiarization, two of the four short se-
quences (one from each metrical context) were
presented, 12 times each in succession with
750 ms of intervening silence. During the test
stage, all four sequences were heard in each
block of trials in random order, each one pre-
sented within a melodic context. Half of the lis-
teners heard the sequences presented only in
matched melodic contexts, and half heard the
sequences presented only in mismatched
melodic contexts. The four sequences were pre-
sented in random order within each of nine
blocks, yielding a total of 36 trials. Half of the
trials contained sequences that were heard at
familiarization and half of the trials con-
tained sequences not heard at familiarization.
 EPISODIC MEMORY FOR MUSIC
Thus, familiarization of the test sequences
(present/absent at familiarization) was a within-
subjects factor and was crossed with melodic
context (matched meter/mismatched meter),
which was a between-subjects factor.
On each test trial a melody was sounded
twice, with 1 s of silence between repetitions.
Instructions to listeners were the same as in-
structions in Experiment 1, with the following
exceptions. Listeners were told that each
melody at test would contain an embedded short
sequence with the same pitch/duration pattern
as one they heard during the familiarization
stage of the experiment. They were asked to
respond yes or no as to whether the embedded
sequence was identical to one of the perform-
ances they heard during the familiarization
stage. As in Experiment 1, listeners were told
that some of the test sequences would have the
same notes as the familiarization sequences, but
they would not be identical because they were
from a different performance. The additional
musical examples played for nonmusician lis-
teners in Experiment 1 were included in this
study as well, to reduce the increased task diffi-
culty in this experiment (listeners must find the
embedded sequence as well as judge its famil-
iarity). Listeners also marked their degree of
confidence in their decision on a three-point
scale (1 ⫽ not confident, 3 ⫽ confident).
Results and Discussion
An ANOVA on the percentage of identifica-
tion (“yes”) responses by familiarization condi-
tion (was test sequence familiarized/not familiar-
ized) and melodic context (matched/mismatched
meter) indicated a significant effect of familiar-
ization, F (1,62) ⫽ 21.0, p ⬍ .01. As shown in
Fig. 4, listeners were able to identify those per-
formances they heard at familiarization cor-
rectly, even when they were embedded in a
melodic context. Listeners’ responses to the fa-
miliarized and unfamiliarized sequences differed
significantly within both the matched and mis-
matched conditions (p ⬍ .05). In addition, identi-
fication was improved for the mismatched con-
text over the matched context; the interaction
between familiarization and melodic context ap-
proached significance, F (1,62) ⫽ 3.56, p ⫽ .06.
539
The d-prime values confirmed listeners’ in-
creased identification accuracy for performances
in mismatched (d⬘ ⫽ 0.868) relative to matched
(d⬘ ⫽ 0.358) melodic contexts (nonparametric
Mann–Whitney U ⫽ 724, p ⬍ .01); only the
d-primes for mismatched contexts differed from
zero (p ⬍ .01). Neither the sequences that were
included at familiarization nor the random order
in which subjects heard trials was a significant
factor. Analyses indicated higher percentages of
“no” responses in block 1, F (8,496) ⫽ 3.2,
p ⬍ .01, but block effects disappeared when the
first block of trials was removed, possibly indi-
cating an initial difficulty of locating the embed-
ded sequence. Confidence ratings did not differ
significantly across familiarization or context
variables.
Two additional analyses were conducted to
ascertain whether listeners’ musical training in-
fluenced the differences found between the
matched versus the mismatched melodic context
conditions, which was a between-subjects vari-
able. First, the musical training of the 32 listen-
ers in each between-subjects group (matched
and mismatched contexts) was compared; there
were no significant differences in number of
years of musical training between the matched
context group (mean ⫽ 7.4 years) and the mis-
matched context group (mean ⫽ 7.8 years). Sec-
ond, listeners were divided into two groups
within each familiarization by melodic context
condition, based on a median split on amount of
musical experience. Analyses on mean re-
sponses by familiarization condition, melodic
context, and musical training indicated no ef-
fects of musical training or interactions with the
other variables. Thus, it is unlikely that the find-
ings are based on different degrees of musical
training.
These results indicate that prosodic features
for short melodic sequences can be used to iden-
tify those sequences in larger melodic contexts.
Listeners were able to locate the familiarized se-
quences in a larger melodic context and identify
whether they contained the same prosodic cues
as heard earlier. Furthermore, listeners’ sensitiv-
ity to prosodic features was influenced by the
metrical context in which they appeared; short
sequences presented in mismatched metrical
540 PALMER, JUNGERS, AND JUSCZYK
FIG. 4. Experiment 3: adult listeners’ mean percentage of yes responses by familiarization condition (famil-
iarized/not familiarized) and melodic context (matched/mismatched prosodic cues), with standard error bars.
contexts were easier to recognize than se-
quences in matched contexts. However, it is
unlikely that identification judgments were
based solely on an abstract metrical structure, as
indicated by listeners’ inability to identify ex-
plicitly the meter of the short sequences out of
context. Together, these findings suggest that
listeners identified the embedded performances
based partially on performance-specific (non-
structural) features and partially on acoustic fea-
tures that correlated with an abstract metrical
structure.
GENERAL DISCUSSION
Several experiments demonstrated that listen-
ers’ memory for musical sequences can incorpo-
rate detailed, instance-specific acoustic features
that differentiate one performance from another.
Listeners were able to identify a particular per-
formance from a set of performances which re-
tained the same melodic and rhythmic structure
(i.e., the same musical composition) but differed
in expressive features. This finding disconfirms
the view that only structural categories in music,
such as pitch and duration categories, can be re-
tained in memory. More consistent with these
findings is an episodic view of memory for
music, in which individual representations for
acoustic events encode stimulus-specific fea-
tures in addition to abstract structural features.
This view is also consistent with recent perspec-
tives on speech perception and recognition that
consider stimulus-specific nonlinguistic (indexi-
cal) properties to be perceived and encoded
episodically along with abstract linguistic prop-
erties (Goldinger, 1997; Jusczyk, 1993; Luce &
Lyons, 1998; Nygaard, Burt, & Queen, 2000;
Nygaard, Sommers, & Pisoni, 1995; Pisoni,
1997).
How is memory for specific music perform-
ances related to memory for abstract musical
structure? This question raises the relationship
between episodic and generic (abstract) memo-
ries. Episodic memory has been defined as a
recollection of a specific experience that pre-
serves spatial and/or temporal properties of the
experience (Tulving & Thompson, 1973).
Whether young children have episodic memory
is in debate; episodic memory usually entails
conscious awareness of the experience (Tulv-
ing, 1985). Repeated exposure to exemplars
may produce not only traces of the individual
events in episodic memory but also a single ab-
stract representation in a functionally (and pos-
sibly anatomically) separate memory system
(cf. Tulving, 1983). Multiple-trace theories pro-
pose in contrast that only traces of individual
episodes are stored; traces acting in concert at
retrieval represent the abstract memory (cf.
Hintzman, 1986). We do not distinguish be-
 EPISODIC MEMORY FOR MUSIC
tween these alternatives here; rather, we posit
that both infants and adults can form memories
that encode subcategorical, temporally specified
acoustic properties of specific musical experi-
ences.
Do listeners encode only those acoustic fea-
tures that are correlated with abstract structure
(such as meter in these experiments)? Although
the acoustic features of the short performances
were not manipulated separately, the experi-
mental evidence suggests the answer is no.
First, infants as well as musically trained and
untrained adults, whose familiarity and training
with abstract musical structures varied widely,
were able to identify familiarized performances
in Experiments 1 and 2. The familiarized per-
formances were balanced within and across in-
dividuals so that metrical structure alone could
not be a defining feature of which perform-
ances were heard earlier. Second, musically
trained listeners were unable to identify explic-
itly the metrical structure of the musical se-
quences in a control experiment, although im-
plicit knowledge may be present (cf. Reber,
1989). Third, listeners were able to identify all
familiarized sequences when they were placed
in metrical contexts in Experiment 3—not just
those sequences that were placed in a mis-
matched metrical context, as might be expected
if the metrical structure alone were retained.
The fact that identification was aided by a mis-
matched metrical context rules out the possibil-
ity that the context was insufficient to induce a
perception of meter. Thus, listeners appear to
be identifying music performances based both
on performance-specific (nonstructural) fea-
tures and on abstract structural features.
Musical training or experience does not ap-
pear to be a prerequisite for listeners’ ability to
encode and recall features of particular music
performances. Experiment 1 demonstrated that
both musically trained and untrained adult lis-
teners could identify particular performances of
the musical sequences with high accuracy. Ex-
periment 2 demonstrated that 10-month-old in-
fants were able to recognize and distinguish spe-
cific music performances they had heard earlier.
Although some work suggests that sensitivity to
complex abstract musical structures, such
541
as tonality and implied harmony, changes from
infancy to adulthood (Cuddy & Badertscher,
1987; Krumhansl & Keil, 1982; Lynch & Eilers,
1992; Trainor & Trehub, 1992), other findings
suggest that children’s basic auditory perception
abilities function along principles similar to
those of adults (Baruch & Drake, 1997; De-
many, 1982; Drake & Penel, 1999; Trehub,
2000; Trehub, Bull, & Thorpe, 1984). Analogies
between musical motion and principles of phys-
ical motion also suggest that musical training
may not be necessary for sensitivity to acoustic
features of performances. For example, the rate
of slowing at phrase boundaries in music resem-
bles deceleration patterns of physical motion, as
in running and walking (Kronman & Sundberg,
1987). The similarity between infants’ and
adults’ identification of music suggests that an
ability to encode stimulus-specific acoustic fea-
tures at least for short musical sequences is pres-
ent early in life—by 10 months of age. This
functionality is important if the acoustic fea-
tures serve to mark salient units in music,
speech, and other auditory domains (Drake &
Penel, in press; Trehub, 2000); an ability to en-
code the acoustic features which mark units is
necessary for the recognition of those units.
One limitation of the current studies is that in-
dividual acoustic features were not isolated in
stimulus presentation; whether one acoustic cue
is more memorable than another is unknown.
Further research addresses the extent to which
one acoustic dimension dominates another in
memory for these performances (Jungers &
Palmer, 2000). Another limitation is the reduced
length and number of musical sequences, which
were chosen to allow comparisons among infant
and adult perception. Finally, stimulus variabil-
ity in these experiments was reduced on several
acoustic dimensions, including pitch, timbre,
and time; durational categories and tempi were
kept constant, in order to keep the total stimulus
duration equivalent across performances. Many
studies indicate that variability in event timing
in music performance correlates with musical
structure (Palmer & Kelly, 1992; Sloboda, 1983;
Todd, 1985), and listeners are sensitive to that
variability (Juslin, 1997; Palmer, 1996b);
whether memory for auditory features of music
542 PALMER, JUNGERS, AND JUSCZYK
would differ in the presence of these forms of
stimulus variability is unknown.
Why would episodic representations for
music be useful? One reason is its bootstrap-
ping potential in perceptual learning. A com-
mon view in speech is that prosodic features
provide a low-level cue to aid segmentation,
particularly during acquisition of knowledge
about the units of speech (Gleitman & Wanner,
1982; Hirsh-Pasek et al., 1987; Jusczyk &
Kemler Nelson, 1996; Morgan, 1986; Peters,
1983). Listeners’ abilities to identify smaller
musical units may serve to bootstrap their abil-
ity to apprehend higher-order relationships
among those units. This research indicates that
listeners can incorporate prosodic features in
memory for music in the absence of extensive
musical experience. Other studies have shown
that acoustic features of music can influence in-
fants’ perception of musical units (Jusczyk &
Krumhansl, 1993; Krumhansl & Jusczyk,
1990). However, this study is the first to
demonstrate infants’ ability to retain in memory
those acoustic features—an important prerequi-
site for a bootstrapping explanation of listeners’
acquisition of complex musical relationships.
A related reason that episodic memory for
music performances would be useful is the re-
lationship that structural ambiguity holds with
acoustic variability in music. Music is often
ambiguous in structure. Each musical piece al-
lows multiple interpretations of phrasing, voic-
ing, and segment boundaries, and musicians
typically use multiple acoustic features to mark
those structures. Performances of the same
music can differ greatly, even when performed
by the same musician. These studies and many
others show that prosodic features of perform-
ances are correlated with performers’ structural
intentions, and listeners are sensitive to that re-
lationship. Thus, prosodic features may be
necessary for a memory representation that dis-
tinguishes one structural interpretation of a mu-
sical piece from another. This view is consis-
tent with Raffman’s (1993) perspective that
expressive features help listeners to distinguish
among multiple structural representations for a
given musical piece; we posit further that the
memory representations encode at least some
of the performance-specific features. Last,
episodic memory for music performances may
enable listeners to identify individual perform-
ers in much the same way that voice character-
istics allow identification of individual talkers.
Music, like speech, is a domain in which
stimulus variability offers a primary resource
for aiding memory and learning. This perspec-
tive is supported by several points. First, most
musical instruments permit variability on multi-
ple dimensions, and the inherent ambiguities in
musical structure not only allow but require per-
formers to make use of acoustic variability to
encode structure. Sources of stimulus variabil-
ity, such as the performers’ emotional state, pro-
duction rate, and syntactic/interpretive effects,
produce large changes in the acoustic signal.
Second, communication of structural informa-
tion in music is typically maximized through re-
dundancy of cues; multiple acoustic cues carry
information about the same structural content.
For example, reductions in tempo and amplitude
often mark phrase boundaries. These cues may
provide reliable information to facilitate com-
munication when the signal is presented under
degraded conditions (cf. Palmer, 1996b). Fi-
nally, music is multidimensional and, like
speech, it has a complex mapping among its dif-
ferent instantiations in the physical domain, the
production domain, and the perceptual domain.
Its complexity is evidenced by the difficulty of
devising rule-based systems that map acoustic
cues onto structural categories. Episodic mem-
ory for at least some acoustic features may be
necessary to facilitate the recognition of struc-
ture in auditory sequences of this complexity.
REFERENCES
Baruch, C., & Drake, C. (1997). Tempo discrimination in in-
fants. Infant Behavior and Development, 20, 573–577.
Bengtsson, I., & Gabrielsson, A. (1983). Analysis and syn-
thesis of musical rhythm. In J. Sundberg (Ed.), Studies
of music performance (pp. 27–60). Stockholm: Royal
Swedish Academy of Music.
Bradlow, A. R., Nygaard, L. C., & Pisoni, D. B. (1999). Ef-
fects of talker, rate, and amplitude variation on recogni-
tion memory for spoken words. Perception & Psy-
chophysics, 61, 206–219.
Chang, H. W., & Trehub, S. E. (1977). Auditory processing
of relational information by young infants. Journal of
Experimental Child Psychology, 24, 324–331.
 EPISODIC MEMORY FOR MUSIC
Clynes, M. (1986). When time is music. In J. R. Evans & M.
Clynes (Eds.), Rhythm in psychological, linguistic, and
musical processes (pp. 169–224). Springfield, IL:
Thomas.
Clynes, M. (1995). Microstructural musical linguistics:
Composers’ pulses are liked most by the best musi-
cians. Cognition, 55, 269–310.
Cooper, W. E., & Sorensen, J. (1977). Fundamental fre-
quency contours at syntactic boundaries. Journal of the
Acoustical Society of America, 62, 683–692.
Cuddy, L. L., & Badertscher, B. (1987). Recovery of
the tonal hierarchy: Some comparisons across age
and levels of musical experience. Perception & Psy-
chophysics, 41, 609–620.
Cutler, A., Dahan, D., & van Donselaar, W. (1997). Prosody
in the comprehension of spoken language: A literature
review. Language and Speech, 40, 141–201.
Demany, L. (1982). Auditory stream segregation in infancy.
Infant Behavior and Development, 5, 261–276.
Dowling, W. J., & Harwood, D. L. (1986). Music cognition.
Orlando: Academic Press.
Drake, C., & Penel, A. (1999). Learning to play music:
Rhythm and timing. Parole, 9, 49–62.
Fellowes, J. M., Remez, R. E., & Rubin, P. E. (1997). Per-
ceiving the sex and identity of a talker without natural
vocal timbre. Perception & Psychophysics, 59,
839–849.
Gabrielsson, A., & Juslin, P. N. (1996). Emotional expres-
sion in music performance: Between the performer’s
intention and the listener’s experience. Psychology of
Music, 24, 68–91.
Gerstman, L. (1968). Classification of self-normalized vow-
els. IEEE Transactions on Audio and Electroacoustics
(ACC-16), 78–80.
Gleitman, L. R., & Wanner, E. (1982). The state of the state
of the art. In E. Wanner & L. R. Gleitman (Eds.), Lan-
guage acquisition: The state of the art (pp. 3–48).
Cambridge: Cambridge Univ. Press.
Goldinger, S. D. (1997). Words and voices: Perception and
production in an episodic lexicon. In K. Johnson & J.
W. Mullennix (Eds.), Talker variability in speech pro-
cessing (pp. 33–66). San Diego: Academic Press.
Henderson, M. T. (1936). Rhythmic organization in artistic
piano performance. In C. E. Seashore (Ed.), Objective
analysis of musical performance, Vol. 4 (pp. 281–305).
Iowa City: Univ. of Iowa Press.
Hintzman, D. L. (1986). “Schema abstraction” in a multiple-
trace memory model. Psychological Review, 93,
411–428.
Hirsh-Pacek, K., Kemler Nelson, D. G., Jusczyk, P. W.,
Wright-Cassidy, K., Druss, B., & Kennedy, L. (1987).
Clauses are perceptual units for young infants. Cogni-
tion, 26, 269–286.
Houston, D. M., & Jusczyk, P. W. (2000). The role of talker-
specific information in word segmentation by infants.
Journal of Experimental Psychology: Human Percep-
tion and Performance, 26, 1570–1582.
Jones, M. R., & Boltz, M. (1989). Dynamic attending and
responses to time. Psychological Review, 96, 459–491.
543
Jungers, M. K., & Palmer, C. (2000). Episodic memory for
music performance. Presented at the Meeting of the
Psychonomic Society, New Orleans, Nov.
Jusczyk, P. W. (1993). From language-general to language-
specific properties: The WRAPSA model of how
speech perception develops. Journal of Phonetics, 21,
3–28.
Jusczyk, P. W. (1997). The discovery of spoken language.
Cambridge, MA: MIT Press.
Jusczyk, P. W. (1998). Using the headturn preference pro-
cedure to study language acquistion. In C. Rovee-
Collier, L. P. Lipsitt, & H. Hayne (Eds.), Advances in
infancy research (Vol. 12, pp. 188–204). Stamford,
CT: Ablex.
Jusczyk, P. W., Hohne, E. A., Jusczyk, A. M., & Redanz, N.
J. (1993). Do infants remember voices? Paper pre-
sented at the 125th Meeting of the Acoustical Society
of America, Ottawa, Canada, May.
Jusczyk, P. W., & Hohne, E. A. (1997). Infants’ memory for
spoken words. Science, 277, 1984–1986.
Jusczyk, P. W., Houston, D. M., & Newsome, M. (1999).
The beginnings of word segmentation in English-learn-
ing infants. Cognitive Psychology, 39, 159–207.
Jusczyk, P. W., & Kemler Nelson, D. G. (1996). Syntactic
units, prosody, and psychological reality during in-
fancy. In J. L. Morgan & K. Demuth (Eds.), Signal to
syntax (pp. 389–408). Mahwah, NJ: Erlbaum.
Jusczyk, P. W., & Krumhansl, C. L. (1993). Pitch and
rhythm patterns affecting infants’ sensitivity to musical
phrase structure. Journal of Experimental Psychology:
Human Perception and Performance, 19, 627–640.
Juslin, P. N. (1997). Emotional communication in music
performance: A functionalist perceptive and some data.
Music Perception, 14, 383–418.
Kemler Nelson, D. G., Jusczyk, P. W., Mandel, D. R.,
Myers, J., Turk, A., & Gerken, L. A. (1995). The Head-
turn Preference Procedure for testing auditory percep-
tion. Infant Behavior and Development, 18, 111–116.
Kendall, R. A., & Carterette, E. C. (1990). The communica-
tion of musical expression. Music Perception, 8,
129–163.
Kronman, U., & Sundberg, J. (1987). Is the musical ritard an
allusion to physical motion? In A. Gabrielsson (Ed.),
Action and perception in rhythm and music (no. 55, pp.
57–68). Stockholm: Royal Swedish Academy of
Music.
Krumhansl, C. L., & Jusczyk, P. W. (1990). Infants’ percep-
tion of phrase structure in music. Psychological Sci-
ence, 1, 70–73.
Krumhansl, C. L., & Keil, F. C. (1982). Acquisition of the
hierarchy of tonal functions in music. Memory & Cog-
nition, 10, 243–251.
Large, E. W., Palmer, C., & Pollack, J. B. (1995). Reduced
memory representations for music. Cognitive Science,
19, 53–96.
Lehiste, I. (1970). Suprasegmentals. Cambridge, MA: MIT
Press.
Lehiste, I., Olive, J. P., & Streeter, L. (1976). The role of du-
ration in disambiguating syntactically ambiguous sen-
544 PALMER, JUNGERS, AND JUSCZYK
tences. Journal of the Acoustical Society of America,
60, 1199–1202.
Lerdahl, F., & Jackendoff, R. (1983). A generative theory of
tonal music. Cambridge, MA: MIT Press.
Luce, P. A., & Lyons, E. A. (1998). Specificity of memory
representations for spoken words. Memory & Cogni-
tion, 26, 708–715.
Lynch, M. P., & Eilers, R. E. (1992). A study of perceptual
development for musical tuning. Perception & Psy-
chophysics, 52, 599–608.
Mandel, D. R., Jusczyk, P. W., & Kemler Nelson, D. G.
(1994). Does sentence prosody help infants organize
and remember speech information? Cognition, 53,
155–180.
Meyer, L. B. (1956). Emotion and meaning in music.
Chicago: Univ. of Chicago Press.
Morgan, J. L. (1986). From simple input to complex gram-
mar. Cambridge, MA: MIT Press.
Nakamura, T. (1987). The communication of dynamics
between musicians and listeners through musical
performance. Perception & Psychophysics, 41,
525–533.
Narmour, E. (1990). The analysis and cognition of basic
melodic structures: The implication-realization model.
Chicago: Univ. of Chicago Press.
Nazzi, T., Kemler Nelson, D. G., Jusczyk, P. W., & Jusczyk,
A. M. (2000). Six-month-olds’ detection of clauses em-
bedded in continuous speech: Effects of prosodic well-
formedness. Infancy, 1, 124–147.
Nygaard, L. C., Burt, S. A., & Queen, J. S. (2000). Surface
form typicality and asymmetric transfer in episodic
memory for spoken words. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 26,
1228–1244.
Nygaard, L. C., Sommers, M. S., & Pisoni, D. B. (1994).
Speech perception as a talker-contingent process. Psy-
chological Science, 5, 42–46.
Nygaard, L. C., Sommers, M. S., & Pisoni, D. B. (1995). Ef-
fects of stimulus variability on perception and repre-
sentation of spoken words in memory. Perception &
Psychophysics, 57, 989–1001.
Palmer, C. (1989). Mapping musical thought to musical
performance. Journal of Experimental Psychology:
Human Perception and Performance, 15, 331–346.
Palmer, C. (1996a). Musical communication and theories
of the stimulus. Paper presented at the 131st Meeting
of the Acoustical Society of America, Indianapolis,
April.
Palmer, C. (1996b). On the assignment of structure in music
performance. Music Perception, 14, 23–56.
Palmer, C. (1997). Music performance. Annual Review of
Psychology, 48, 115–138.
Palmer, C., & Kelly, M. H. (1992). Linguistic prosody and
musical meter in song. Journal of Memory and Lan-
guage, 31, 525–542.
Palmer, C., & Krumhansl, C. L. (1990). Mental representa-
tions for musical meter. Journal of Experimental Psy-
chology: Human Perception and Performance, 16,
728–741.
Palmeri, T. J., Goldinger, S. D., & Pisoni, D. B. (1993).
Episodic encoding of voice attributes and recognition
memory for spoken words. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 19,
309–328.
Peters, A. (1983). The units of language acquisition. Cam-
bridge: Cambridge Univ. Press.
Pisoni, D. B. (1997). Some thoughts on “normalization” in
speech perception. In K. Johnson & J. W. Mullennix
(Eds.), Talker variability in speech processing (pp.
9–32). San Diego: Academic Press.
Price, P. J., Ostendorf, M., Shattuck-Hufnagel, S., & Fong,
C. (1991). The use of prosody in syntactic disambigua-
tion. Journal of the Acoustical Society of America, 90,
2956–2970.
Raffman, D. (1993). Language, music, and mind (Bradford
Book ed.). Cambridge, MA: MIT Press.
Reber, A. S. (1989). Implicit learning and tacit knowledge.
Journal of Experimental Psychology: General, 118,
219–235.
Remez, R. E., Fellowes, J. M., & Rubin, P. E. (1997). Talker
identification based on phonetic information. Journal
of Experimental Psychology: Human Perception and
Performance, 23, 651–666.
Repp, B. H. (1989). Expressive microstructure in music: A
preliminary perceptual assessment of four composers’
“pulses.” Music Perception, 6, 243–274.
Scott, D. R. (1982). Duration as a cue to the perception of a
phrase boundary. Journal of the Acoustical Society of
America, 71, 996–1007.
Seashore, C. E. (1938). Psychology of music. New York:
McGraw–Hill.
Serafine, M. L., Glassman, N., & Overbeeke, C. (1989). The
cognitive reality of hierarchic structure in music. Music
Perception, 6, 347–430.
Shankweiler, D., Strange, W., & Verbrugge, R. (1977).
Speech and the problem of perceptual constancy. In R.
Shaw & J. Bransford (Eds.), Perceiving, acting, and
knowing: Toward an ecological psychology (pp.
315–345). Hillsdale, NJ: Erlbaum.
Shattuck-Hufnagel, S., & Turk, A. E. (1996). A prosody
tutorial for investigators of auditory sentence proc-
essing. Journal of Psycholinguistic Research, 25,
193–247.
Siegel, J. A., & Siegel, W. (1977). Categorical perception of
tonal intervals: Musicians can’t tell sharp from flat.
Perception & Psychophysics, 21, 399–407.
Sloboda, J. A. (1983). The communication of musical metre
in piano performance. Quarterly Journal of Experi-
mental Psychology, 35, 377–396.
Sloboda, J. A. (1985). Expressive skill in two pianists:
Metrical communication in real and simulated per-
formances. Canadian Journal of Psychology, 39,
273–293.
Soderstrom, M., Jusczyk, P. W., & Kemler Nelson, D. G.
(2000). Evidence for the use of phrasal packaging by
English-learning 9-month-olds. In S. C. Howell, S. A.
Fish, & T. Keith-Lucas (Eds.), Proceedings of the 24th
Annual Boston University Conference on Language
 EPISODIC MEMORY FOR MUSIC
Development (Vol. 2, pp. 708–718). Somerville, MA:
Cascadilla Press.
Speer, S. R., Crowder, R. G., & Thomas, L. M. (1993).
Prosodic structure and sentence recognition. Journal of
Memory and Language, 32, 336–358.
Streeter, L. A. (1978). Acoustic determinants of phrase
boundary perception. Journal of the Acoustical Society
of America, 64, 1582–1592.
Studdert-Kennedy, M. (1974). The perception of speech.
In T. A. Sebeok (Ed.), Current trends in linguistics
(pp. 2349–2385). The Hague: Mouton.
Studdert-Kennedy, M. (1983). On learning to speak. Human
Neurobiology, 2, 191–195.
Sundberg, J., Askenfelt, A., & Fryden, L. (1983). Musical
performance: A synthesis-by-rule approach. Computer
Music Journal, 7, 37–43.
Thompson, W. F. (1989). Composer-specific aspects of mu-
sical performance: An evaluation of Clynes’ theory of
pulse for performances of Mozart and Beethoven.
Music Perception, 7, 15–42.
Todd, N. P. M. (1985). A model of expressive timing in tonal
music. Music Perception, 3, 33–58.
Trainor, L. J. (1996). Infant preferences for infant-directed
versus non-infant-directed play songs and lullabies.
Infant Behavior and Development, 19, 83–92.
Trainor, L. J., Clark, E. D., Huntley, A., & Adams, B. A.
(1997). The acoustic basis of preferences for infant-
directed singing. Infant Behavior and Development,
20, 383–396.
Trainor, L. J., & Adams, B. (2000). Infants’ and adults’ use
of duration and intensity cues in the segmentation
of tone patterns. Perception & Psychophysics, 62,
333–340.
545
Trainor, L. J., & Trehub, S. E. (1992). A comparison of in-
fants’ and adults’ sensitivity to Western musical struc-
ture. Journal of Experimental Psychology: Human
Perception and Performance, 18, 394–402.
Trehub, S. E. (2000). Human processing predispositions and
musical universals. In N. L. Wallin, B. Merker, & S.
Brown (Eds.), The origins of music. Cambridge, MA:
MIT Press.
Trehub, S. E., Bull, D., & Thorpe, L. A. (1984). Infants’ per-
ception of melodies: The role of melodic contour. Child
Development, 55, 821–830.
Trehub, S. E., & Thorpe, L. A. (1989). Infants’ perception of
rhythm: Categorization of auditory sequences by tem-
poral structure. Canadian Journal of Psychology, 43,
217–229.
Trehub, S. E., & Trainor, L. J. (1998). Singing to infants:
Lullabies and play songs. In C. Rovee-Collier &
L. Lipsitt (Eds.) Advances in infancy research (pp.
43–77). Norword, NJ: Ablex.
Tulving, E. (1983). Elements of episodic memory. Oxford:
Oxford Univ. Press.
Tulving, E. (1985). Memory and consciousness. Canadian
Psychology, 26, 1–12.
Tulving, E., & Thompson, D. M. (1973). Encoding speci-
ficity and retrieval processes in episodic memory.
Psychological Review, 80, 352–373.
Yee, W., Holleran, S., & Jones, M. R. (1994). Sensitivity to
event timing in regular and irregular sequences: Influ-
ences of musical skill. Perception & Psychophysics, 56,
461–471.
(Received July 6, 2000)
(Revision received December 1, 2000)
(Published online June 20, 2001)