Content uploaded by Frank Russo
Author content
All content in this area was uploaded by Frank Russo on Aug 30, 2015
Content may be subject to copyright.
EFFECTS OF EMERGENT-LEVEL STRUCTURE ON MELODIC
PROCESSING DIFFICULTY
FRANK A. RUSSO
Ryerson University, Toronto, Canada
WILLIAM FORDE THOMPSON
Macquarie University, Sydney, Australia
LOLA L. CUDDY
Queen’s University, Kingston, Canada
F
OUR EXPERIMENTS ASSESSED THE INFLUENCE
of emergent-level structure on melodic processing dif-
ficulty. Emergent-level structure was manipulated
across experiments and defined with reference to the
Implication-Realization model of melodic expectancy
(Narmour, 1990, 1992, 2000). Two measures of melodic
processing difficulty were used to assess the influence of
emergent-level structure: serial-reconstruction and
cohesion ratings. In the serial-reconstruction experi-
ment (Experiment 1), reconstruction was more efficient
for melodies with simple emergent-level structure. In
the cohesion experiments (Experiments 2-4), ratings
were higher for melodies with simple emergent-level
structure, and the advantage was generally greater in
the presence of simple surface-level structure. Results
indicate that emergent-level structure as defined by the
model can influence melodic processing difficulty.
Received: November 22, 2014, accepted April 13, 2015.
Key words: melody, hierarchical structure, memory,
cohesion, expectancy
T
HERE IS A RICH MUSIC
-
THEORETIC TRADITION
of describing hierarchical structure in music
(e.g., Forte, 1977; Lerdahl, 1988, 1989; Lerdahl
& Jackendoff, 1983; Meyer, 1973; Narmour, 1983, 1990;
Schenker, 1969, 1979). In the case of melody, this nor-
mally includes some description of the melodic surface
and emergent structures that are composed of salient
events (see Narmour, 1983, for a useful review). From
a cognitive perspective, an important question that
arises from this work is whether listeners are sensitive
to these descriptions and whether they may relate in
some manner to melodic processing difficulty. In other
words, does ease of processing depend in some manner
on emergent-level structure defined by theory? The
current study investigates whether melodic processing
difficulty varies with respect to music-theoretic descrip-
tions of emergent-level structure derived from the
Implication-Realization (I-R) model (Narmour, 1990,
1992).
Two leading cognitive approaches to understanding
melodic complexity include information-theoretic and
dynamic attending models. Information-theoretic mod-
els have focused on the development of coding systems
(Cuddy, Cohen, & Mewhort, 1981; Deutsch, 1980; Leeu-
wenberg, 1969; Restle, 1970; Simon, 1972). A hierarchi-
cal melody with surface- and emergent-level structure
can be described economically using nested codes that
exploit redundancies. The codes are assumed to capture
important aspects of mental representation, and empir-
ical studies have found that melodies with shorter codes
are easier to process (Boltz & Jones, 1986; Deutsch &
Feroe, 1981). Dynamic attending theory has focused on
the role of attention in the mental representations of
melody (Jones, 1987, 1993; Jones & Boltz, 1989). Inter-
nal oscillators are presumed to entrain to levels of oscil-
latory structure that are defined by rhythmic and
melodic accents (Large & Jones, 1999). The joint accent
structure hypothesis posits that melodic processing is
facilitated when rhythmic and melodic accents are in
phase (Boltz & Jones, 1986; Jones, 1987; Jones & Pfor-
dresher, 1997; Jones & Ralston, 1991). This approach
has received extensive empirical support and provides
more flexibility than coding systems.
The I-R model makes predictions regarding proces-
sing difficulty and provides additional flexibility regard-
ing the instantiation of hierarchical structure (Narmour,
1990, 1991, 2000). This is accomplished by considering
expectancy at different levels of structure. Two tones in
sequence at any level of structure are said to be impli-
cative, leading to bottom-up and top-down expectancies
for the next note to follow. Bottom-up expectancies
are Gestalt-like and proposed to be innate.
1
Top-down
1
Narmour’s innate proposal is called into question by Pearce and
Wiggins (2006, 2012) work on the IDyoM model, which demonstrates
that the bottom-up expectancies can be simulated by corpus-based sta-
tistical learning.
Music Perception,VOLUME 33, ISSUE 1, PP. 96–109, ISSN 0730-7829, ELECTRONIC ISSN 1533-8312. ©2015 BY THE REGENTS OF THE UNIVERSITY OF CALIFORNIA ALL
RIGHTS RESERVED.PLEASE DIRECT ALL REQUESTS FOR PERMISSION TO PHOTOCOPY OR REPRODUCE ARTICLE CONTENT THROUGH THE UNIVERSITY OF CALIFORNIA PRESS’S
RIGHTS AND PERMISSIONS WEBSITE,HTTP://WWW.UCPRESSJOURNALS.COM/REPRINTINFO.ASP. DOI: 10.1525/MP.2015.33.1.96
96 Frank A. Russo, William Forde Thompson, & Lola L. Cuddy
expectancies are acquired through statistical learning
(extra-opus) and through rule iteration (intra-opus).
Complexity at any given level of structure is thought
to depend on the extent to which it fulfills expectancy
(Narmour, 1990, 1992; also see Meyer, 1956, pp. 138-
139). Consistent with this view, Rohrmeier and Cross
(2013) recently reported that implicit learning of melo-
dies is impeded when surface-level events frequently
deny bottom-up expectancies described in the I-R
model (Rohrmeier & Cross, 2013). Similarly, Loui
(2012) found that statistical learning of an artificial
grammar is impaired after small intervals are removed
from melodies (denial of a bottom-up principle of
expectancy referred to as pitch proximity). No empirical
evidence exists to date regarding whether bottom-up
principles of expectancy may influence melodic proces-
sing difficulty in hierarchically structured melodies.
In the current study, hierarchical structure was pri-
marily established through the manipulation of bottom-
up principles of expectancy. Note-to-note transitions
within groups generally fulfilled expectancy, while
note-to-note transitions between groups denied expec-
tancy. The denials were achieved by following a small
interval (three semitones or less) with a large interval
(six semitones or greater). These expectancy denials
occurred with temporal regularity so as to create
surface-level groups of regular size with the first note
of each group rising to the emergent level (see Figure 1).
The type of expectancy denial implemented in this
study has been judged unexpected across a variety of
contexts (Cuddy & Lunney, 1995; Krumhansl, 1995;
Schellenberg, 1996, 1997). Fujioka, Trainor, Ross,
Kakigi, and Pantev (2004) found that this type of
denial elicits a magnetic type of mismatch negativity
(MMNm), suggesting that it is encoded preattentively
and automatically.
Emergent-level structure was further clarified using
two devices. First, under certain conditions, familiar
melodic patterns served as surface-level groups (e.g.,
major triad). Familiar patterns facilitate the identifica-
tion of groups, which should support events rising to
the emergent level. Second, under certain conditions
surface-level groups were repeated under simple trans-
position. Repetition should further clarify grouping
structure and draw attention to the emergent level
(Deutsch & Feroe, 1981; Meyer, 1956; Margulis, 2013;
Narmour, 1990, 1992, 1999, 2000).
Meyer (1973, p. 53) states that ‘‘on the hierarchic level
where repetition is immediate, it [repetition] tends to
separate events. But on the next level – where similar
events are grouped together as part of some larger unit –
repetition tends to create coherence.’’ Similarly, the I-R
model states that all other things being equal, the extent
of repetition at the surface will influence perceived com-
plexity and that this complexity is inversely related to
attention at the emergent level (Narmour, 2000, Table 1).
Hence, an immediate exact repetition of form (A
0
,A
0
)is
more expected and will emphasize emergent-level struc-
ture more than an immediate near repetition of form
(A
0
,A
1
), which in turn will be more expected and
emphasize the emergent-level more than an immediate
contrast in form (A, B). See Figure 2 for examples of
different types of surface-level repetition.
Structure at the emergent level was categorically
labeled as simple or complex. Simple emergent-level
structure involved a sequence of small intervals moving
in the same direction. This type of structure is referred
to as process in the I-R model and is considered highly
expected (Narmour, 1989, 1990, 1999). Similar struc-
tures have been described in other models as highly
expected and even archetypal: inertia (Larson, 2012),
good continuation (Meyer, 1956, 1973), step inertia
(Huron, 2006; von Hippel, 2002) and direction (Margu-
lis, 2005). Complex emergent-level structure also
involved a sequence of small intervals but the direction
of intervals varied resulting in a combination of struc-
tures that is less expected.
Two response measures were implemented to evaluate
the effects of emergent-level structure on melodic proces-
sing difficulty: serial reconstruction and perceived cohe-
sion. In the serial reconstruction task adapted from the
jigsaw puzzle procedure designed by Delie
`ge and collea-
gues (Delie
`ge & Me
´len, 1997; Delie
`ge, Me
´len, Stammers,
& Cross, 1996; also see Tillmann, Bigand, & Madurell,
1998), a melody is presented, after which participants are
given randomly arranged segments from the melody and
FIGURE 1. Surface-level grouping is instantiated by realizing a denial of
expectancy on every fourth note. Notes 1, 4 and 7 form a highly expected
emergent-level structure referred to as a
process
.
Effects of emergent-level structure 97
asked to rearrange the order so as to match the original.
In the cohesion task, listeners are asked to judge the
perceived cohesion of melodies.
Eerola, Himberg, Toiviainen, and Louhivuori (2006)
formalized a number of statistical measures to predict
melodic complexity: entropy of pitch-class distribution,
entropy of interval distribution, mean interval size,
entropy of duration distribution, rhythmic variability,
note density, tonal ambiguity, accent incoherence, con-
tour self-similarity, and contour entropy. These measures
were drawn from information-theoretic, music-theoretic,
and dynamic attending approaches to melodic complex-
ity. As we were primarily interested in the influence of
emergent-level structure as defined by the I-R model, test
melodies within each experiment were composed in
a manner that minimized variability in these measures
across levels of emergent structure (i.e., no statistically
significant differences).
Experiment 1
The aim of this experiment was to assess melodic pro-
cessing difficulty in hierarchically structured melodies
using a serial reconstruction task. Melodies were com-
posed to establish simple or complex emergent-level
structure according to principles of the I-R model. At
the surface level, melodies either repeated the same
group under transposition or chained together unre-
lated surface-level groups. The former type of surface-
level structure was referred to as simple and the latter,
complex.
For melodies with simple surface-level structure, each
surface-level group consisted of a major or minor triad.
The group was repeated five times under transposition.
For melodies with complex surface-level structure, the
surface-level groups were more variable, including less
familiar non-triadic sequences. We predicted main
effects of simple- and emergent-level structure, as well
as an interaction, whereby emergent-level differences
would be enhanced when surface-level structure was
simple. Ease of processing was assessed using a serial
reconstruction task.
METHOD
Participants. Twenty-four undergraduate students were
recruited to participate from the Queen’s University
community. Demographic information for each partic-
ipant group (musician/nonmusician) is provided in
Table 1. Participants recruited through the Introductory
Psychology Participant Pool were given course credit for
their participation. These participants included a mix of
musicians and nonmusicians. Additional participants
for the musician group were recruited using posters
displayed around campus. Musicians recruited with
posters were reimbursed with nominal payment.
Apparatus. Participants were individually tested in
a sound-attenuated chamber. Melodies were generated
FIGURE 2. Examples of exact repetition of form (A
0
,A
0
), near repetition
of form (A
a
,A
b
), and contrast in form (A, B) at the surface level.
TABLE 1.
Demographic Information
Experiment 1 Experiment 2 Experiment 3 Experiment 4
Musicians
Mean (SE) Points
þ
12.09 (1.03) 13.13 (0.88) 10.03 (0.59) 10.95 (0.59)
Female / Male 9 / 3 13 / 2 13 / 3 12 / 3
Mean Age (years) 21.1 19.1 21.3 20.5
Nonmusicians
Mean (SE) Points 1.58 (0.68) 1.87 (0.34) 1.67 (0.38) 1.81 (0.69)
Female / Male 6 / 6 13 / 2 13 / 3 10 / 5
Mean Age (years) 22.0 19.3 20.2 21.2
þ
All musicians continued to be musically active, whereas nonmusicians were not. One point was awarded for each year of private instruction and a half point for each year of
group instruction.
98 Frank A. Russo, William Forde Thompson, & Lola L. Cuddy
from a Roland SoundCanvas tone generator, set to
‘‘Piano,’’ under the control of a Power Mac computer
running MusicShop software. Melodies were played
through a single Fostex 6301 speaker monitor situated
approximately one foot in front of the listener and set to
a comfortable listening level. Icons were presented on
the screen to represent chunks of the melody. Partici-
pants were able to rearrange the order of icons except
the first using a computer mouse.
Stimuli. Eight melodies were composed to encompass
all combinations of two binary factors. First, melodies
possessed either simple or complex surface-level struc-
ture. Simple surface-level structure involved the repeti-
tion of a familiar melodic group (a major or minor
triad). Complex surface-level structure involved a variety
of melodic groups. Second, melodies possessed either
simple or complex emergent-level structure. In melo-
dies with simple emergent-level structure, the first note
of each surface-level group formed a process at the
emergent level. In melodies with complex emergent-
level structure, the first note of each surface-level group
formed a combination of structures at the emergent
level that involved a contour change. An inverted coun-
terpart of each melody was generated, resulting in a total
of eight test melodies (ascending and descending var-
iants of each binary combination of factors).
The music notation for each test melody is provided
in Figure 3. All test melodies shared the following char-
acteristics: 15 tones (9 tones occurring once and 3 tones
occurring twice), 8 contour changes, and a pitch range
of 11 semitones. None of the melodies implied a tradi-
tional Western tonal key as determined by the
Krumhansl-Schmuckler key-finding algorithm (see
Krumhansl, 1990), i.e., no significant correlation with
any of the 24 tonal hierarchies. Melodies were isometric
and isochronous, with an interonset interval of 250 ms.
The surface-level groups (as defined by the I-R model)
were consistently three tones in length, leading to an
implied triple meter.
Procedure. On each trial, participants were presented
with one of the eight test melodies. Following presenta-
tion of the melody, the participant was provided with
five icons corresponding to a temporal sequence. The
first icon was identified with the letter ‘‘T’ and was
always associated with the first three notes of the mel-
ody. The remaining icons (identified with letters ‘‘N,’’
‘‘P,’’ ‘‘V,’’ and ‘‘R’’) were associated with a unique three-
note sequence drawn from the melody. Icons could be
arranged in any sequence and played back at will. The
initial presentation order of icons was randomized with
the provision that at least one icon move was required
before reconstructing the melody. Once participants
believed they had correctly rearranged the icons, they
were required to transcribe the letter tags onto an
answer sheet. Trial orders were independently random-
ized for each participant.
RESULTS AND DISCUSSION
Reconstruction accuracy was at ceiling (95.8%)and
thus not interpretable, but variability was present in:
(1) number of icon moves; (2) number of replays; and
(3) response latency (i.e., the time from the end of the
initial presentation to the final transcription). A mixed
analysis of variance was conducted on each of these
three measures with emergent-level structure and
surface-level structure as the within-subjects variables,
and musicianship as the between subjects variable. An
alpha-level of .05 was used for all statistical tests.
For each measure, the main effects of emergent-level
structure and surface-level structure were significant.
There were no significant main effects or interactions
involving musicianship. Figure 4 displays the mean
number of icon moves, replays, and response times,
collapsed across musicianship.
Number of icon moves. Melodies with simple emergent-
level structure received fewer icon moves than melodies
with complex emergent-level structure, F(1, 24) ¼
16.66, p< .001. Melodies with simple surface-level
structure received fewer icon moves than melodies with
complex surface-level structure, F(1, 24) ¼9.89, p< .01.
Replay. Melodies with simple emergent-level structure
were replayed fewer times than melodies with complex
emergent-level structure, F(1, 24) ¼7.21, p< .05. Mel-
odies with simple surface-level structure were replayed
fewer times than melodies with complex surface-level
structure, F(1, 24) ¼4.82, p< .05.
Response time. Melodies with simple emergent-level
structure led to shorter response times than melodies
with complex emergent-level structure, F(1, 24) ¼5.64,
p< .05. Melodies with simple surface-level structure led
to shorter response times than melodies with complex
surface-level structure, F(1, 24) ¼4.39, p< .05.
The results of these analyses support our prediction
that melodic processing is facilitated by complexity at
both emergent-level and surface-level of structure as
defined by the I-R model. Although the predicted inter-
action between surface and emergent-level structure
was not found, a consistent trend may be observed in
Figure 4, wherein the advantage of simple emergent-
level structure appears to be more present in melodies
with simple surface-level structure.
Effects of emergent-level structure 99
Experiment 2
Much like the shape of a visually presented object,
melodies are usually perceived in a ‘‘Gestalt’’ manner
(von Ehrenfels, 1937). It is possible that the serial
reconstruction task employed in Experiment 1 some-
how altered this normative mode of listening. In Exper-
iment 2, we adopted a cohesion-rating task where
cohesion was defined as ‘‘the extent to which the tones
of a melody sound as though they create an organized
FIGURE 3. Musical notation for ascending (original) and descending (inverted) melodies used in Experiments 1 and 2.
100 Frank A. Russo, William Forde Thompson, & Lola L. Cuddy
whole.’’ The operative assumption here is that judgments
of cohesion will be influenced by ease of processing. Mel-
odies were identical to those used in Experiment 1.
METHOD
Participants. Thirty undergraduate students were
recruited to participate from the Queen’s University
Psychology Participant Pool. Demographic information
is provided in Table 1. All participants were given course
credit for their participation.
Apparatus. The cohesion-rating task was conducted in
a sound-attenuated chamber with groups of 1 to 3 par-
ticipants. Melodies were generated and presented over
a Roland FP1 digital piano, set to ‘‘Piano 1,’’ under the
control of a Power Mac Computer running MusicShop
software. Response sheets were used to record cohesion
ratings.
Stimuli. Test melodies were identical to the eight melo-
dies used in Experiment 1.
Procedure. Eighteen randomized orders of melody pre-
sentation were constructed, one for each of 18 test
groups. Participants in each test group were asked to
rate the cohesion of the eight melodies. Each melody
was presented twice in succession with a 2-s pause
between presentations. The cohesion of each melody
was rated using a 7-point scale (1 ¼not cohesive,7¼
very cohesive). Participants were encouraged to use the
full range of the scale. To familiarize participants with
the nature of the melodies and the rating scale, partici-
pants were given two practice trials. The melodies in
these practice trials were randomly selected from the
set of test melodies.
RESULTS AND DISCUSSION
A mixed analysis of variance was conducted with
emergent-level structure (simple vs. complex) and
surface-level structure (simple vs. complex) as the
within-subjects factors and musicianship as the
between-subjects factor. Consistent with Experiment 1,
main effects of emergent-level and surface-level structure
were both significant. The interaction between emergent-
level and surface-level structure was also significant.
Figure 5 displays cohesion ratings collapsed across
musicianship. Melodies with simple emergent-level
structure yielded higher cohesion ratings than melodies
with complex emergent-level structure, F(1, 28) ¼
77.96, p< .0001. Melodies with simple surface-level
structure yielded higher cohesion ratings than melodies
with complex surface-level structure, F(1, 28) ¼91.37,
p< .0001. The advantage of simple emergent-level
structure was amplified for melodies with simple
FIGURE 4. Mean number of icon moves, replays, and response times in
Experiment 1 as a function of eme rgent-level complexity (simple vs.
complex) and surface-level complexity (simple vs. complex).
FIGURE 5. Mean cohesion ratings in Experiment 2 as a function of
emergent-level complexity (simple vs. complex) and surface-level
complexity (simple vs. complex).
Effects of emergent-level structure 101
surface-level structure, F(1, 28) ¼10.31, p< .003.
Although this interaction did not reach significance in
the serial reconstruction data (Experiment 1), similar
trends were apparent.
Although none of the melodies tested in the first two
experiments were associated with a major or minor key,
they all began with a prototypical diatonic pattern (major
or minor triads) that was immediately repeated in trans-
position (A
0
,A
0
). The familiarity of these patterns may
have helped to instantiate an implied triple meter, sup-
porting perception of the emergent-level structure that
was otherwise defined by systematic placement of expec-
tancy denials. In addition, this implied meter may have
been reinforced due to the lack of variability in the length
of surface-level groups. Experiment 3 was conducted to
explicitly control for these factors that may have contrib-
uted to the emergent-level findings.
Experiment 3
The results of Experiments 1 and 2 suggest that relations
between emergent-level tones can influence the manner
in which listeners perceive and remember melodies.
However, the melodies used in these experiments con-
tained structural cues beyond regular expectancy
denials that may have reinforced emergent-level struc-
ture. Experiment 3 was conducted to determine whether
sensitivity to emergent-level structure would persist
when groups were not reinforced by these other cues.
METHOD
Participants. Thirty-two undergraduate students were
recruited to participate from the Queen’s University
Psychology Participant Pool. Demographic information
is provided in Table 1. Participants were given course
credit for their participation.
Apparatus. The experiment was conducted in a sound-
attenuated chamber at Queen’s University. A Power
Mac computer running Experiment Creator Software
was used to present melodies and collect responses.
Melodies were realized as MIDI performances using
a piano patch, with sound output over Sennheiser
HD280 headphones.
Stimuli. Sixteen melodies were composed for this exper-
iment and presented in original and inverted form to
create 32 test melodies (see Figure 6). None of the mel-
odies contained prototypic triadic patterns. Melodies
varied in emergent-level complexity (simple vs. com-
plex), the number of tones in each surface level group
(3 or 4 tones), and surface-level complexity (4 levels). As
in prior experiments, emergent-level structure involved
a sequence of small intervals forming a process (simple)
or a combination of structures (complex). The four levels
of surface-level complexity were created by manipulating
the degree of redundancy between surface-level groups.
At the simplest level (1), a single surface-level group was
repeated (in transposition) throughout the melody.
Higher levels of complexity progressively reduced the
extent of repetition. The highest level of complexity (4)
contained no repetition.
Eight dummy melodies were interspersed among the
twenty-four test melodies. Dummy melodies were
composed with surface-level groups that were five
tones in length. The purpose of the dummy melodies
was to reduce the likelihood that listeners would carry
over expectations about group length from earlier
trials. The resultant 32 melodies (24 test melodies and
8 dummy melodies) possessed the same number of
surface-level groups (5), but varied in number of tones
because of differences in group length (triple, quadru-
ple, quintuple).
All melodies had a pitch range of 11 semitones with
a frequency distribution of pitches that did not clearly
imply a traditional Western tonal key as determined by
the Krumhansl-Schmuckler key-finding algorithm (as
described in Krumhansl, 1990). Melodies were isometric
and isochronous, with an interonset interval of 250 ms.
Procedure. The procedure was identical to that
described in Experiment 2 except that participants were
run individually.
RESULTS AND DISCUSSION
A mixed analysis of variance was conducted with
emergent-level structure (simple vs. complex), surface-
level structure (4 levels of complexity), and group length
(3 or 4 tones) as the within-subjects variables and musi-
cianship as the between subjects variable. Figure 7 dis-
plays cohesion ratings collapsed across musicianship.
Melodies with simple emergent-level structure
yielded higher cohesion ratings than melodies with
complex emergent-level structure, F(1, 29) ¼11.53,
p< .01. Lower levels of surface-level complexity also led
to higher cohesion ratings, F(1, 29) ¼18.23, p< .0001.
This finding is compatible with the main effect of
surface-level structure revealed in Experiments 1 and 2.
We predicted that lower levels of surface-level com-
plexity would emphasize emergent-level structure and
amplify the advantage of simple emergent-level struc-
ture. Although the interaction between emergent-level
structure and surface-level structure was not significant,
F(3, 87) ¼1.44, p ¼.24, the advantage of simple
emergent-level structure was significant at all levels of
102 Frank A. Russo, William Forde Thompson, & Lola L. Cuddy
surface-level complexity (all pvalues < .05), except for
the highest level, F(1, 29) < 1. This finding suggests that
it is not necessary to have strict repetition of the same
surface-level group (in transposition) in order to
observe effects of emergent-level structure.
There was an effect of group length, F(1, 29) ¼8.46,
p< .01. Melodies with groups that were four tones in
length were perceived as more cohesive than melodies
with groups that were three tones in length. One possi-
bility is that this finding reflects a cultural bias in favor
FIGURE 6. Musical notation for test melodies used in Experiment 3 (inverted melodies not shown). Each melody is labeled as having simple or complex
emergent-level structure. The first number in brackets represents the number of notes in surface-level groups (3 or 4), while the second number
represents the extent of surface-level complexity (1 ¼low and 4 ¼high).
Effects of emergent-level structure 103
of duple time (see e.g., Smith & Cuddy, 1989; Trainor &
Corrigall, 2010). Another possibility is that listeners’
ratings were partly influenced by the absolute length
of surface-level groups, with longer groupings deemed
to be more cohesive.
Although melodies in this experiment were composed
without the use of overly familiar surface-level groups,
the first surface-level group was always repeated in trans-
position. The I-R model suggests that this initial repeti-
tion (A
0
,A
0
) should facilitate the perception of
emergent-level structure (Narmour, 2000). Experiment
4 was designed to assess whether sensitivity to complexity
of emergent-level structure could be observed using mel-
odies that do not repeat the initial surface-level group.
Experiment 4
Experiments 1-3 revealed that ease of melodic proces-
sing depends on surface and emergent-level grouping.
Surface-level groups were defined through the use of
regularly occurring expectancy denials and by an imme-
diate repetition of the initial surface-level group. The
question addressed in Experiment 4 was whether this
initial repetition is essential for the observed effect of
emergent-level structure.
METHOD
Participants. Thirty undergraduate students were
recruited to participate from the University of Toronto
Community. Demographic information for each partic-
ipant group (musician/nonmusician) is provided in
Table 1. Participants recruited through the Introductory
Psychology Participant Pool were given course credit for
their participation. These participants included a mix of
musicians and nonmusicians. Additional participants
for the musician group were recruited using posters
displayed around campus. Participants recruited using
posters were reimbursed with nominal payment.
Apparatus. The experiment was conducted in a sound-
attenuated chamber at the University of Toronto, Mis-
sissauga. The equipment used to present stimuli and
collect data was identical to that described in Experi-
ment 3.
Stimuli. Twelve melodies were composed and presented
in both original and inverted forms to create 24 test
melodies. Melodies varied in the number of tones in
each surface-level group (3 or 4 tones), emergent-level
structure (simple vs. complex), and surface-level com-
plexity (3 levels). As may be seen in Figure 8, increasing
levels of surface-level complexity were associated with
lower levels of redundancy between surface-level
groups, but in no case did this redundancy involve an
immediate repetition of a melodic group (A
0
,A
0
). At
the simplest level (1), two surface-level groups that
formed a near repetition were alternated (A
a
,A
b
,A
a
,
A
b
,A
a
). At the most complex level (4), there was almost
no repetition present across surface-level groups (A
a
,B,
A
b
, C, D). As in Experiment 3, eight dummy melodies
with surface-level groups of 5 tones were interspersed
among the test melodies in order to minimize any
carry-over effect of group length. All melodies pos-
sessed the same number of surface-level groups (5), but
melodies varied in length from 15-25 tones because of
the variable length of groups. All other aspects of the
melodies were consistent with test melodies used in
Experiment 3.
Procedure. The procedure was identical to that
described in Experiment 3.
RESULTS AND DISCUSSION
A mixed analysis of variance was conducted with
emergent-level structure (simple vs. complex), surface-
level structure (3 levels of complexity), and group length
(3 or 4 tones) as the within-subjects variables and musi-
cianship as the between subjects variable. The main
effect of musicianship and its interactions were not
significant.
Figure 9 displays mean cohesion ratings across levels
of surface-level and emergent-level complexity. The
main effect of emergent-level structure did not reach
significance, F(1, 28) ¼1.87, p¼.18. Thus, eliminating
FIGURE 7. Mean cohesion ratings in Experiment 3 as a function of
emergent-level complexity (simple vs. complex) and surface-level
complexity (1 ¼low and 4 ¼high).
104 Frank A. Russo, William Forde Thompson, & Lola L. Cuddy
repetition of the first surface-level group reduced the
likelihood that listeners would perceive the emergent-
level structure. However, the interaction between
emergent-level structure and surface-level structure
was significant, F(2, 56) ¼4.01, p< .05. Orthogonal
contrasts revealed that while there was no effect of
emergent-level structure in melodies with high (3) and
intermediate (2) surface-level complexity, F(1, 28) < 1,
the emergent-level effect was significant in melodies
with low (1) surface-level complexity, F(1, 28) ¼
5.75, p< .05. For melodies with low surface-level com-
plexity, melodies with simple emergent-level structure
were judged as more cohesive than melodies with com-
plex emergent-level structure. Thus it seems that at
least for nontonal melodies, some surface repetition
may be necessary to perceive emergent-level effects.
The effects of surface-level complexity and group
length were all consistent with those observed in Exper-
iment 3. Ratings of cohesion were higher for melodies
with simple surface-level structure, F(2, 56) ¼41.01, p<
.0001. This finding was predicted and is attributable to
the reduction in complexity that is conferred by increas-
ing pattern repetition. Ratings were higher for melodies
with surface-level groups that were four tones in length,
F(1, 28) ¼16.60, p<.0001.AssuggestedinExperiment3,
possible explanations for this finding include a cultural
bias in favor of duple time or an effect of absolute length
of surface-level groups. Although this issue is beyond the
FIGURE 8. Musical notation for test melodies used in Experiment 4 (inverted melodies not shown). Each melody is labeled as having simple or complex
emergent structure. The first numbers in brackets represents the number of notes in surface-level groups (3 or 4), while the second number represents
the extent of surface-level complexity (1 ¼low and 3 ¼high).
Effects of emergent-level structure 105
scope of this study, future research might contrast
these two interpretations by including melodies with
even longer groupings that are not in duple time (e.g.,
five-note groups).
General Discussion
Serial-reconstruction and cohesion-rating tasks were
administered to assess processing difficulty in hierarchi-
cally structured melodies as defined by the I-R model.
There were three main findings. First, melodies with
simple emergent-level structure were easier to process
than melodies with complex emergent-level structure.
Second, melodies with simple surface-level structure were
easier to process than melodies with complex surface-
level structure. Third, sensitivity to emergent-level struc-
ture generally increased with increasing simplicity at the
surface level. None of the experiments yielded main
effects or interactions involving musicianship.
The processing advantage for melodies with simple
emergent-level structure has been characterized using
music-theoretic descriptions of emergent-level structure
derived from the Implication-Realization model. In such
melodies, a single emergent-level structure connected
together non-adjacent tones that were the first tones of
surface-level groups. In the terms of the I-R model, the
emergent-level structure formed a process, involving
sequential pitch intervals composed of non-adjacent
tones moving in a common direction. A process is con-
sidered a highly expected structure by Narmour (1989,
1990, 1999), as well as other theorists (Huron, 2006;
Larson, 2012; Margulis, 2005; Meyer, 1956, 1973).
On the other hand, the processing advantage of mel-
odies with simple emergent-level structure could also be
argued from an information-theoretic perspective. Mel-
odies that can be described with relatively short codes
are easier to process than melodies with longer codes.
However, while it is true that an emergent-level process
can be captured by a short code, coding systems do not
provide a mechanism for the establishment of
emergent-level structure in the absence of exact repeti-
tion at the surface level.
Although an expectancy-based explanation for the
emergent-level findings appears likely, there were no
direct tests of expectancy conducted in this study. Both
experimental tasks required that listeners make retro-
spective judgments (serial reconstruction and cohe-
sion). Hence, it remains remotely possible that some
account of simplicity that does not depend upon expec-
tation per se was responsible for the pattern of judg-
ments observed. It would be valuable for future studies
of hierarchical melodic structure to combine prospec-
tive tasks that tap expectancy alongside of retrospective
tasks. It would also be useful to incorporate melodies
that fulfill emergent-level expectancy in a manner that is
distinct from process (e.g., a Narmourian reversal).
Sensitivity to emergent-level structure became more
apparent when multiple structural cues were available.
The presence of familiar patterns (Experiment 1, 2), an
immediate repetition (Experiment 3), and/or minimal
variation in surface-level groups (Experiment 4; Surface
Level Complexity ¼1), appear to have been instrumen-
tal in yielding effects of emergent-level structure. We do
not presume that this is an exhaustive list of criteria but
it seems that a listener’s attention will tend to remain at
the melodic surface in the absence of substantive evi-
dence reinforcing a possible emergent-level structure.
Other studies have also found evidence for sensitivity
to emergent-level structure in melody. Memory for
short and simple melodic structures appears to preserve
emergent-level structure (Bigand, 1990, Exp. 3; Deutsch
& Feroe, 1981; Sloboda & Parker, 1985). Listeners prefer
correct over incorrect melodic reductions (Dibben,
1994; Serafine et al., 1989), and are able to perceive tonal
tensioninnon-adjacentdependencies(Lerdahl&
Krumhansl, 2007; see also Cuddy & Smith, 2000; Smith
& Cuddy, 2003).
Statistical learning studies have also revealed sensitiv-
ity to non-adjacent dependencies. By manipulating
pitch proximity between the odd and even tones in
a melody, Creel, Newport, and Aslin (2004) were able
to affect grouping such that relationships between non-
adjacent tones were readily learned. On the basis of this
finding and those of the present study, it seems reason-
able to predict that statistical learning of non-adjacent
dependencies will be harder in melodies that deny
FIGURE 9. Mean cohesion ratings in Experiment 4 as a function of
emergent-level complexity (simple vs. complex) and surface-level
complexity (1 ¼low and 3 ¼high).
106 Frank A. Russo, William Forde Thompson, & Lola L. Cuddy
emergent-level expectancy. In this way, the bottom-up
principles may form a sort of bottleneck for the proces-
sing of melodic structures (Rohrmeier & Cross, 2013).
Other studies have cast doubt on listeners’ sensitivity
to emergent-level structure. Cook (1987) investigated
listeners sensitivity to tonal closure. Listeners were
asked to indicate their preference between two versions
of the same piano excerpt, only one of which involved
a return to the initial key. Results showed little effect of
tonal closure on judgments. A similar result was
obtained by Marvin and Brinkman (1999) when they
explicitly asked participants about whether excerpts
ended in the key with which they started.
In addition, Bigand and Parncutt (1999) found no
evidence for an influence of hierarchical structure in
chord sequences on judgments of tension. This finding
stands in contrast with tension data reported by Lerdahl
and Krumhansl (2007). One explanation for this dis-
crepancy is that tension was defined differently across
the two studies. Bigand and Parncutt (1999) defined
tension as a feeling of non-closure in the sense that ‘‘there
must be a continuation of the sequence’’ (p. 242). Lerdahl
and Krumhansl (2007) did not emphasize this non-
closural aspect in their definition of tension. It seems
that there are many instances in which these two defini-
tions may lead to unique modes of listening and hence
contradictory outcomes – e.g., the initial tonic chord of
a harmonic sequence where there is no implication of
closure.
The picture that is developing from various studies of
emergent-level structure is that listeners can be rather
sensitive to theoretical descriptions of structure beyond
the surface, but that the extent of this sensitivity
depends greatly on the cues supporting the hierarchy
and on the listening mode. The current investigation
adds to existing research on melodic perception by
demonstrating that emergent-level structure contributes
to processing difficulty and that emergent-level struc-
ture may be instantiated in the absence of exact repeti-
tion at the surface.
Author Note
This research was supported by independent discovery
grants awarded to each author from the Natural
Sciences and Engineering Research Council of Canada.
We thank Eugene Narmour, Daniel Levitin, and Anir-
uddh Patel for helpful suggestions.
Correspondence concerning this article should be
addressed to Frank Russo, Department of Psychology,
Ryerson University, Toronto, Ontario, M5B 2K3,
Canada. E-mail: russo@ryerson.ca
References
BIGAND, E. (1990). Abstraction of two forms of underlying
structure in a tonal melody. Psychology of Music, 18, 45-59.
BIGAND,E.,&PARNCUTT, R. (1999). Perceiving musical
tension in long chord sequences. Psychological Research, 62,
237-254.
BOLTZ,M.,&JONES, M. R. (1986). Does rule recursion make
melodies easier to reproduce? If not, what does? Cognitive
Psychology,18, 389-431.
COOK, N. (1987). The perception of large-scale tonal closure.
Music Perception, 5, 197-206.
CREEL,S.C.,NEWPORT,E.L.,&ASLIN, R. N. (2004). Distant
melodies: Statistical learning of nonadjacent dependencies in
tone sequences. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 30, 1119-1130.
CUDDY,L.L.,COHEN,A.J.,&MEWHORT, D. J. K. (1981).
Perception of structure in short melodic sequences. Journal of
Experimental Psychology: Human Perception and Performance,
7, 869-883.
CUDDY,L.L.,&LUNNEY, C. A. (1995). Expectancies
generated by melodic intervals: Perceptual judgments of
melodic continuity. Perception and Psychophysics, 57,
451-462.
CUDDY,L.L.,&SMITH, N. A . (2000). Perception of tonal pitch
space and tonal tension. In D. Greer (Ed.), Musicology and
sister disciplines: Past, present, future (pp. 47-59). Oxford:
Oxford University Press.
DELIE
`GE,I.,&ME
´LEN, M. (1997). Cue abstraction in the rep-
resentation of musical form. In I. Delie
`ge & J. A. Sloboda
(Eds.), Perception and cognition of music (pp. 387-412).
London: Lawrence Erlbaum Associates.
DELIE
`GE,I.,&ME
´LEN,M.,STAMMERS,D.,&CROSS, I. (1996).
Musical schemata in real time listening to a piece of music.
Music Perception, 14, 117-160.
DEUTSCH, D. (1980). The processing of structured and
unstructured tonal sequences. Perception and Psychophysics,
28, 381-389.
DEUTSCH,D.,&FEROE, J. (1981). The internal representation of
pitch sequences in tonal music. Psychological Review, 88, 503-522.
DIBBEN, N. (1994). The cognitive realityof hierarchic structure in
tonal and atonal music. Music Perception, 12, 1-25.
EEROLA,T.,HIMBERG,T.,TOIVIAINEN,P.,&LOUHIVUORI,J.
(2006). Perceived complexity of Western and African folk
melodies by Western and African listeners. Psychology of
Music, 34, 337-371.
Effects of emergent-level structure 107
FORTE, M. R. A . (1977). The structure of atonal music.New
Haven, CT: Yale University Press.
FUJIOKA,T.,TRAINOR,L.J.,ROSS,B.,KAKIGI,R.,&PANTEV,C.
(2004). Musical training enhances automatic encoding of
melodic contour and interval structure. Journal of Cognitive
Neuroscience, 16, 1010-1021.
HURON, D. B. (2006). Sweet anticipation: Music and the psy-
chology of expectation. Cambridge, MA: MIT Press.
JONES, M. R. (1987). Dynamic pattern structure in music: Recent
theory and research. Perception and Psychophysics,41(6),
621-634.
JONES, M. R. (1993). Dynamics of musical patterns: How do
melody and rhythm fit together. In W. J. Dowling & T. J. Tighe
(Eds.), Psychology and music: The understanding of melody and
rhythm (pp. 67-92). Hillsdale, NJ: Lawrence Erlbaum.
JONES,M.R.,&BOLTZ, M. (1989). Dynamic attention and
responses to time. Psychological Review, 96, 459-491.
JONES,M.R.,&PFORDRESHER, P. Q. (1997). Tracking musical
patterns using joint accent structure. Canadian Journal of
Experimental Psychology/Revue canadienne de psychologie
expe
´rimentale,51(4), 271-291.
JONES,M.R.,&RALSTON, J. T. (1991). Some influences of accent
structure on melody recognition. Memory and Cognition,
19(1), 8-20.
KRUMHANSL, C. L. (1990). Cognitive foundations of musical pitch.
New York: Oxford University Press.
KRUMHANSL, C. L. (1995). Music psychology and music theory:
Problems and prospects. Music Theory Spectrum, 17, 53-80.
LARGE,E.W.,&JONES, M. R. (1999). The dynamics of attend-
ing: How people track time-varying events. Psychological
Review, 106(1), 119-159.
LARSON, S. (2012). Musical forces: Motion, metaphor, and
meaning in music. Bloomington, IN: Indiana University Press.
LEEUWENBERG, E. L. (1969). Quantitative specification of infor-
mation in sequential patterns. Psychological Review, 76,
216-220.
LERDAHL, F. (1988). Tonal pitch space. Music Perception, 5,
315-350.
LERDAHL, F. (1989). Atonal prolongational structure.
Contemporary Music Review, 4, 65-87.
LERDAHL,F.,&JACK ENDOFF, R. (1983). A generative theory of
tonal music. Cambridge, MA: MIT Press.
LERDAHL,F.,&KRUMHANSL, C. L. (2007). Modeling tonal ten-
sion. Music Perception, 24, 329-366.
LOUI, P. (2012). Learning and liking of melody and harmony:
Further studies in artificial grammar learning. Topics in
Cognitive Science, 4, 1-14.
MARGULIS, E. H. (2005). A model of melodic expectation. Music
Perception, 21, 663-714.
MARGULIS, E. H. (2013). On repeat: How music plays the mind.
New York: Oxford University Press.
MARVIN,E.W.,&BRINKMAN, A . (1999). The effect of modu-
lation and formal manipulation on perception of tonic closure
by expert listeners. Music Perception, 16, 389-407.
MEYER, L. B. (1956). Emotion and meaning in music. Chicago, IL:
University of Chicago Press.
MEYER, L. B. (1973). Explaining music. Berkeley, CA: University
of California Press.
NARMOUR, E. (1983). Some major theoretical problems con-
cerning the concept of hierarchy in the analysis of tonal music.
Music Perception,1, 129-199.
NARMOUR, E. (1989). The ‘‘genetic code’’ of melody: Cognitive
structures generated by the implication-realization model.
Contemporary Music Review, 4, 45-63.
NARMOUR, E. (1990). The analysis and cognition of basic melodic
structures: The implication-realization model. Chicago, IL:
University of Chicago Press.
NARMOUR, E. (1991). The top-down and bottom-up systems of
musical implication: Building on Meyer’s theory of emotional
syntax. Music Perception, 9, 1-26.
NARMOUR, E. (1992). The analysis and cognition of melodic
complexity: The implication-realization model. Chicago, IL:
University of Chicago Press.
NARMOUR, E. (1999). Hierarchical expectation and musical style.
In D. Deutsch (Ed.), The psychology of music (2nd ed., pp. 441-
472). San Diego, CA: Academic Press.
NARMOUR, E. (2000). Music expectation by cognitive rule-
mapping. Music Perception, 17, 329-398.
PEARCE,M.T.,WIGGINS, G. A. (2006). Expectation in melody:
The influence of context and learning. Music Perception, 23,
377-405.
PEARCE,M.T.,WIGGINS, G. A. (2012). Auditory expectation:
The information dynamics of music perception and cognition.
Topics in Cognitive Science, 4, 625-652.
ROHRMEIER,M.,&CROSS, I. (2013). Artificial grammar
learning of melody is constrained by melodic inconsistency:
Narmour’s principles affect melodic learning. PloS One, 8,
e66174.
RESTLE, F. (1970). Theory of serial pattern learning: Structural
trees. Psychological Review,77(6), 481.
SCHELLENBERG, E. G. (1996). Expectancy in melody: Tests of the
implication-realization model. Cognition, 58, 75-125.
SCHELLENBERG, E. G. (1997). Simplifying the implication-
realization model of melodic expectancy. Music Perception, 14,
295-318.
SCHENKER, H. (1969). Five graphic musical analyses. New York:
Dover Publications.
SCHENKER, H. (1979). Free composition [Der freie Satz] (E. Oster,
Trans. & Ed.). New York: Pendragon Press.
SERAFINE,M.L.,GLASSMAN,N.,OVERBEEKE, C. (1989). The
cognitive reality of hierarchic structure in music. Music
Perception, 6, 397-430.
108 Frank A. Russo, William Forde Thompson, & Lola L. Cuddy
SIMON, H. A. (1972). Complexity and the representation of
patterned sequences of symbols. Psychological Review, 79,
369-382.
SLOBODA,J.A.,&PARKER, D. H. H. (1985). Immediate recall of
melodies. In P. Howell, I. Cross, & R. West (Eds.), Musical
structure and cognition (pp. 143-167). London: Academic
Press.
SMITH,K.C.,&CUDDY, L. L. (1989). Effects of metric and
harmonic rhythm on the detection of pitch alterations in
melodic sequences. Journal of Experimental Psychology:
Human Perception and Performance, 15, 457-471.
SMITH,N.A.,&CUDDY, L. L . (2003). Perceptions of musical
dimensions in Beethoven’s Waldstein sonata: An application of
tonal pitch space theory. Musicae Scientiae,7, 7-34.
TILLMANN,B.,BIGAND,E.,&MADURELL, F. (1998). Local versus
global processing of harmonic cadences in the solution of
musical puzzles. Psychological Research, 61, 157-174.
TRAINOR,L.J.,&CORRIGALL, K. A. (2010). Music acquisition
and effects of musical experience. In M. R. Jones, R. R. Fay, and
A. Popper (Eds.), Music perception (pp. 89-127). New York:
Springer.
VON EHRENFELS, C. (1937). On Gestalt-qualities. Psychological
Review, 44, 521-524.
VON HIPPEL, P. (2002). Melodic-expectation rules as learned
heuristics. In C. Stevens, D. Burnham, G. McPherson, E.
Schubert, J. Renwich (Eds.), Proceedings of the 7th
International Conference on Music Perception and Cognition.
Sydney, Australia: ICMPC.
Effects of emergent-level structure 109
A preview of this full-text is provided by University of California Press.
Content available from Music Perception: an interdisciplinary journal
This content is subject to copyright.
