Alma Mater Studiorum University of Bologna, August 22-26 2006
Jazz, blues and the language of harmony: Flexibility in
online harmonic processing
Elizabeth Lhost
Cognitive Science Program
Northwestern University
Evanston, IL 60208, USA
e-lhost@northwestern.edu
ABSTRACT
Using ideas from linguistic research and music theory, this
study examines online harmonic processing, focusing on
the effects of contextual cues and schematic expectation.
In this study, participants listened to songs in the 12-bar
jazz-blues style, which we manipulated to fulfill graded
categories of appropriateness. Target chords occurred in
three locations (early, middle and late) and fit, fit moder-
ately well, or did not fit within the context based on theo-
retical assumptions about and a statistical analysis of songs
in this style. Evidence from judgments of fit shows an
ability to differentiate between appropriate and inappropri-
ate chords, thus suggesting the influence of harmonic ex-
pectation during listening. Additionally, longer latencies
for responses to ambiguous chords suggest an attempt to
integrate anomalies into an appropriate hierarchical struc-
ture. These results support the hypothesis that people ac-
commodate certain chord substitutions within the jazz-
blues form while rejecting others, and take longer to decide
when chord category is more ambiguous. These results
provide insight into harmonic processing and online devel-
opment of expectations and strengthen the comparison be-
tween language and music.
In: M. Baroni, A. R. Addessi, R. Caterina, M. Costa (2006) Proceedings
of the 9th International Conference on Music Perception & Cognition
(ICMPC9), Bologna/Italy, August 22-26 2006.©2006 The Society for
Music Perception & Cognition (SMPC) and European Society for the
Cognitive Sciences of Music (ESCOM). Copyright of the content of an
individual paper is held by the primary (first-named) author of that pa-
per. All rights reserved. No paper from this proceedings may be repro-
duced or transmitted in any form or by any means, electronic or me-
chanical, including photocopying, recording, or by any information
retrieval systems, without permission in writing from the paper's primary
author. No other part of this proceedings may be reproduced or transmit-
ted in any form or by any means, electronic or mechanical, including
photocopying, recording, or by any information retrieval system, without
permission in writing from SMPC and ESCOM.
ISBN 88-7395-155-4 © 2006 ICMPC
Richard Ashley
School of Music
Northwestern University
Evanston, IL 60208, USA
r-ashley@northwestern.edu
Keywords
Music; Language; Harmony; Perception; Processing; Ex-
pectation.
INTRODUCTION
Harmonic structure in music parallels syntactic struc-
ture in language (Bernstein, 1973; Besson & Schön, 2003;
Sloboda 1985), and recent studies in music perception sug-
gest processing music is analogous to processing speech.
Both modalities utilize rule-based systems with hierarchical
structures to generate complex acoustic information. In
language, rules of syntax govern speech production; har-
monic rules perform the same role in music (Lerdahl &
Jackendoff, 1983; Sloboda, 1985). Because this analogous
relationship exists, evidence from linguistic research pro-
vides a useful scaffold for the study of music perception.
In fact, recent studies in music perception that use this
analogy improved our understanding of not only musical
processing but also the connections between these do-
mains. Besson, Faïta, Peretz, Bonnel and Requin (1998)
conducted a study comparing physiological brain responses
elicited by semantic incongruities (i.e., inappropriate
words) to those elicited by harmonic incongruities (i.e., out
of key chords). Their study questions whether the same
cognitive systems process the lyrics (i.e., words) and tune
(i.e., music notes) of a song. Their results reveal separate
processing systems for speech and music but illustrate
similarities between those systems. In a study examining a
similar question, Patel, Gibson, Ratner, Besson, and Hol-
comb (1998) demonstrated cognitive parallels in the proc-
essing of syntax in language and harmony in music. Com-
paring neurophysiological responses to syntactic and har-
monic violations revealed a similar time-course for and
location of brain activity.
1282
 ICMPC9 Proceedings
These results demonstrate perceptual attentiveness to
violations of harmonic structure comparable to such viola-
tions in language. This sensitivity suggests a processing
paradigm for music similar to the garden-path theory in
language. When people listen to (or read) sentences, they
use templates derived from past experience to attach new
words to a syntactic structure generated by earlier words as
quickly as possible (Frazier, 1987; 1989). Incongruities
and anomalies interfere with this process because they do
not integrate easily; they disrupt the listening process in
language and music as results from Patel et al. (1998) and
other studies demonstrate. As a result of this work and
other studies (Koelsch, Gunter, Schröger, & Friederici,
2000; Koelsch, Gunter, Yves, Cramon, Zysset, Lohmann
and Friederici, 2002; Koelsch & Mulder, 2002), some of
the cognitive processes formerly argued to be language-
specific no longer appear to be restricted to language but
instead appear to be restricted to the processing of abstract
rules and regularities like those present in language and
music.
Similarities between reactions to violations support
claims for similar processing mechanisms for syntax and
harmony. Studies comparing these reactions suggest that
whether listening to speech or music, people generate hier-
archical models to comprehend complex, rule-based acous-
tic information. As linguistic research shows, syntactic
processing uses templates from past experience and contex-
tual cues to organize input while listening to and creating
expectations for future events. Abnormalities deviate from
existing models and therefore require an adjustment or re-
assessment of the structure. As the garden-path theory
predicts for linguistic examples, revising the initial struc-
ture complicates the integration process (Frazier & Clifton,
1996), but little is known about the process for integrating
harmonic abnormalities in music.
In a study examining this process, Bharucha and
Stoeckig (1986) used a priming paradigm to test harmonic
expectation. For this experiment, participants provided
true/false judgments for target chords that were either
closely or distantly related to the previous priming chord.
Results show faster response times for closely related
chords than for distantly related chords, thus demonstrating
the presence of harmonic expectation during music listen-
ing. Results from a related study further support this con-
clusion. Justus and Bharucha (2001) performed a similar
study in which they examined the automaticity of culture-
based harmonic expectations when those expectations were
in opposition to veridical (i.e., context-based) expectations.
Their results reveal the overwhelming preponderance of
cultural expectations over contextual information, thereby
suggesting the utilization of templates for harmonic struc-
ture derived from past experience while listening to novel
stimuli.
The current study builds upon this past research to
probe the effects of harmonic expectation during online
processing. We devoted particular attention to the context
ISBN 88-7395-155-4 © 2006 ICMPC
in which harmonic abnormalities appear, and chose the
specific harmonic context of the 12-bar jazz-blues form for
our inquiry. This form is unique in that it begins with a
basic linear progression from tonic (I) to subdominant (IV),
back to tonic (I), to dominant (V), and back to tonic (I)
again within the 12-bars, but as Steedman’s (1984) analysis
of this form indicates, the jazz-blues style permits a number
of harmonic transformations. Figure 1 illustrates several of
the transformations this form permits.
Possible chord progressions in the 12-bar form.
a. I IV I I7 IV IV I I V7 V7 I I
a.’ I IV I I 7
IV IV I I 7
IIm V 7
I I
a.’’ I IV I I7 IV IV I VI7 IIm7 V7 I I
7 7 7 7 7
a.’’ I IV I I IV IV IIIm VI IIm V I I
’
---------------------------------etc.------------------------------
Figure 1. Steedman (1996) shows the generation of more
complex 12-bar phrases through the backwards propaga-
tion of the final authentic cadence.
As Figure 1 illustrates, the 12-bar form breaks down into
three four-bar phrases. The first phrase establishes the key
for the tune, typically hovering around the tonic chord.
The second phrase, (ie., the modulation to subdominant),
situates the progression within the jazz-blues style, and the
third phrase marks the cadential conclusion in which the
tune returns to tonic. Steedman’s analysis is interesting in
that it shows a proliferation of possible harmonies as tunes
progress (i.e., there are more permissible deviations toward
the end of the pattern). This proliferation of harmonic pos-
sibility mimics the proliferation of syntactic possibility
following a conjunction while nonetheless describing a
context in which specific expectations for harmonic events
exist at certain locations. However, this proliferation devi-
ates from other contexts in which more possibility for
variation exists at the beginning.
Because the 12-bar form dictates which substitutions
are acceptable, it provides a rule-based yet flexible context,
perfect for examining harmonic processing. Additionally,
its popularity as a form for commercial jazz and blues re-
cordings suggests the possibility that listeners may already
possess a jazz-blues schema that would allow them to proc-
ess typical patterns more quickly than atypical patterns. In
our study we collected fittingness judgments and reaction
times to a series of acceptable and unacceptable 12-bar
patterns. Results reveal the ability to categorically differ-
entiate between acceptable and unacceptable harmonic sub-
stitutions, while response times suggest an attempt to inte-
grate abnormalities within the harmonic framework. These
results reveal graded expectations in accordance with the
theoretical possibilities Steedman (1984) describes, thereby
intimating a garden-path model for online harmonic proc-
essing. Results from our study not only reveal the presence
1283

of harmonic expectations but also reveal similarities be-
tween harmonic and syntactic processing.
Analysis of the jazz-blues corpus
In order to understand processing mechanisms or lis-
tener expectations, we first had to understand the jazz-blues
form, so our study began with a statistical analysis of songs
in this style. The corpus we analyzed consisted of over 100
compositions written in the 12-bar style. We gathered
these songs from jazz and blues fakebooks because such
books contain a broad selection of popular and familiar
tunes, and we identified songs in the 12-bar style by look-
ing for the repetition of a 12-bar harmonic pattern or the
presence of common end-of-phrase markers such as repeat
signs or double bars. We predicted songs would closely
follow the 12-bar pattern Steedman (1984) describes, and
our statistical analysis of the corpus provides results to
support that hypothesis.
The purpose for the statistical analysis of the corpus
was two-fold: (1) to identify important locations within the
form; and (2) to identify the most probable chords at those
locations. In order to determine which measures were im-
portant, we performed a same-different rating of the har-
monies in each measure relative to the previous measure.
A change in harmony from the previous measure received a
judgment of different, which we then used to make com-
parisons between tunes. This analysis revealed measures 5
and 9 as the most important measures; harmonic content
was different for these measures in 91 percent of the tunes.
Measure 7 was also significant with different ratings in 87
percent of the tunes, but a closer analysis of the harmonic
content at that location revealed the preponderance of a
return to tonic in those measures, which was rather uninter-
esting for our purposes. For our third experimental loca-
tion, we chose measure 11 instead. The return to tonic also
characterizes this measure, but because it is at the end of
the 12-bar phrase, we considered this return more meaning-
ful and more interesting. Figure 2 shows the distribution of
harmonic changes for all 12 measures.
Harmonic change by measure
P
e 1
r 0.9
c
0.8
e versity (7 male, 5 female, age range 19-24 years, M = 21.2
n 0.7
t 0.6
0.5
o participation. Seven participants were currently engaged in
f 0.4
0.3
S 0.2
o
n 0.1
g 0
s 2 3 4 5 6 7 8 9 10 11
Measure
Figure 2. Harmonic change by measure.
ISBN 88-7395-155-4 © 2006 ICMPC
ICMPC9 Proceedings
After selecting measures 5, 9, and 11 as our experi-
mental locations, we proceeded to analyze the harmonic
content at those locations. In order to determine which
chords were most likely to occur at those locations, we
performed a between-song comparison, which revealed a
high probability for those chords prescribed by the jazz-
blues form. Subdominant (IV) occurred at measure 5, in 92
percent of the songs. This finding fits nicely with the har-
monic patterns Steedman (1984; 1996) describes. In meas-
ure 9, dominant (V) and supertonic (II) were the most
common chords. Dominant occurred in 33 percent of the
songs and supertonic (II) occurred in 26 percent. This
finding is not surprising considering Steedman (1996)
shows supertonic (II) at measure 9 in 75 percent of the
progressions. Though the relative frequencies of V and II
in measure 9 are similar, by the end of measure 10, almost
half of the songs (49 percent) were dominant. Thus, we
consider the first half of this phrase characterized by domi-
nant and the second half of the phrase as a return to tonic.
Analysis of harmonic content at Measure 11 showed a re-
turn to tonic in 66 percent of the tunes.
From the important locations and most typical (i.e.,
expected) chord/s at these locations, we gained a better
understanding of listener expectations. In addition to re-
vealing the most typical chord at these locations, our analy-
sis also revealed the least likely chords at these locations.
This aspect of the analysis was important in that it suggests
graded harmonic potentials, which later helped guide our
manipulation of the stimuli. The results from our analysis
not only correspond with theoretical predictions for the 12-
bar form but they also reflect the harmonic content most
often encountered during natural listening. These findings
suggest that before entering the experimental setting, ex-
perienced listeners may already possess a specific jazz-
blues schema, which would enable faster judgments. We
predicted participants would make faster judgments for
chords that were either expected or unacceptable and
slower judgments for songs that were acceptable due to the
higher number of possible chords in this category.
METHODS
Participants
Twelve undergraduate students at Northwestern Uni-
years, SD = 1.2 years) participated in the study. Musical
experience and normal hearing were both prerequisites for
music study or performance; 5 participants were currently
majoring in music at Northwestern. All participants were
native speakers of English and received compensation for
their participation.
12
Materials
From the songs in the analysis corpus, we chose 33 of
the most prototypical tunes for our experimental stimuli.
These were tunes with the fewest deviations from the basic
1284
 ICMPC9 Proceedings

12-bar form, which for the purposes of our study possessed
a harmonic rhythm of one chord per measure. We manipu-
lated these tunes at each of the three locations (measure 5,
measure 9, and measure 11) for each type of violation (ex-
Ex-
pected, acceptable and unacceptable) to produce nine ver-
pected
sions of each tune, thus yielding 297 stimuli.
V II
We manipulated stimuli according to predictions about
II
the likelihood of certain chords predicted by our analysis of
II
the corpus. All of our manipulations used chords in the
V IV
tune’s original key. This type of manipulation differs from
past studies because most studies test harmonic expectation
using in-key and out-of-key chords (See, e.g., Patel et al.,
1998; Koelsch, Gunter, Schröger & Friederici, 2003;
Koelsch & Mulder, 2002). We decided in-key substitu-
tions were more appropriate for testing the aspects of har-
monic processing in which we were interested. We wanted
our stimuli to correspond as closely as possible with har-
monic progressions used in real-world jazz and blues com-
positions and therefore preserved the original key.
In measure 5, subdominant (IV), as statistically most
probably chord, represents the “expected” category. We
chose dominant (V) for the “acceptable” category because
it is likely to occur within the tune, though it is not likely to
occur at this location. Additionally, the progression from
tonic (I) to dominant (V) is not only acceptable but also
common within the style, though it typically occurs at
measure 9. For the “unacceptable” category, we chose
supertonic (II) because it occasionally follows tonic (I) in
measure 9 and also because it is harmonically related to
subdominant (IV), though again, it is unlikely to occur at
this location. We replicated this reasoning when manipu-
lating the harmonies at measures 9 and 11 (See Table 1).
As accurately as possible, we integrated manipulations
within the harmonic context of the tune. That is, we ad-
justed manipulations to account specifically for each tune
using the preceding and succeeding chords to determine
which chords fulfilled the three categories of manipulation
for each location.
We created stimuli using music composition software,
and then exported these compositions to Macintosh Quick-
Time Pro 6.5 to control instrumentation and sound quality.
Four beats of ride cymbal preceded each progression, and
the tempo for each tune was standardized at 120 beats per
minute. We played the stimuli on the QuickTime general
MIDI instruments and converted stimuli to audio files for
experimental presentation. The electronic bass played the
bass line for each tune and an electric harpsichord played
the upper harmonies. This instrumentation mirrors but
does not replicate common instrumentation in natural lis-
tening settings. Due to the size of the stimuli set, partici-
pants heard either 36 or 45 tunes during the experimental
session. Each block contained a random sample of the
available stimuli, which represented an equal number of
songs in each condition (i.e., location and type of viola-
tion). Stimuli appeared in a random sequence to avoid or-
dering effects.
Table 1. Sample chord substitutions for measures 9
and 11.
Measure 9 Measure 11
Accept- Unaccept- Ex- Accept- Unaccept-
able able pected able able
III I IV VI
IV VII I V VI
VI IV I IV VI
III I IV VI
Apparatus
We presented stimuli and collected data on a Macin-
tosh iBook laptop computer, using Psyscope 1.2.5 under
Macintosh OS 9.2. A Psyscope button box (New Micros
Inc.) controlled timing and collected responses, thereby
guaranteeing millisecond-accurate timings. Participants
listened to stimuli in a sound-attenuated booth over Alesis
Point Seven studio monitors at a comfortable volume.
Procedure
A brief training session with five sample stimuli pre-
ceded the experimental session in order to acclimate par-
ticipants to the experimental apparatus and the task. After
the training period, participants were permitted to ask the
experimenter questions about the task but not about the
content of the stimuli. The task was self-paced, and par-
ticipants controlled rate of presentation by making a key-
stroke on the computer when they were ready for the next
example. Additionally, participants could rest between
examples if necessary in order to avoid fatigue. During the
experiment, participants responded to target chords match-
ing the conditions described above by pressing the button
corresponding to their judgment. We labeled these buttons
“does not fit,” “fits moderately well,” and “fits well” to
underscores the importance of fit within the context and
also to avoid attempts to guess right or wrong chords. At
the onset of the target chord, lights on the button box illu-
minated to signal the target chord. We chose this visual cue
to avoid interference from an aural cue and to focus par-
ticipants’ attention on the button box during listening. The
12-bar pattern continued to play after the target chord to
emphasize continuity between chords before and after the
target chord, and as response times indicate, this continua-
tion had an important influence on participant judgments.
We instructed participants to make their judgments about
the target chord’s appropriateness as quickly as possible.
We were interested in their first response but wanted some-
thing more substantial than a true/false judgment. Having
three response choices helped eliminate this true/false im-
pulse while stressing the decision-making aspect of the
task.
RESULTS
Data analyses excluded the scores form one participant
whose experimental trial was interrupted by a computer

ISBN 88-7395-155-4 © 2006 ICMPC 1285

 ICMPC9 Proceedings
error, which caused data to be lost. We performed all sta-
tistical procedures using an alpha level of .05 (α = .05).
Response Choice
Analysis of response choice reveals an ability to dif-
ferentiate between chord categories. Participants decided
target chords of the three types (expected, acceptable and
unacceptable) “fit well” in 72, 41 and 20 percent of the
trials, respectively. Similarly, participants judged 8 percent
of the expected chords, 25 percent of the acceptable chords
and 53 percent of the unacceptable chords as chords that
“did not fit.” These results show an ability to differentiate
between acceptable and unacceptable chords at levels
above chance and suggest a connection between graded
harmonic potentials in the analysis corpus and graded ex-
pectations during listening, thereby supporting our argu-
ment for the presence and influence of context-specific
harmonic expectations during listening. Response choice
for target chords in the acceptable category reveal a ten-
dency to accept these chords, but the distribution of re-
sponse choices for chords in this category is distributed
fairly evenly across response choice. Participants chose
“fits well” in response to 42 percent of the stimuli, “fits
moderately well” in response to 31 percent and “does not
fit” in response to 26 percent of the stimuli in this category.
Response times by location
Response times also vary by location, thus further sup-
porting our hypothesis. Longer response times to the early
target location suggest the activation of multiple models of
harmonic structure (M = 2863.4, SD = 2419.3). This result
is not surprising because more possibilities may be active
earlier in processing than later. Accordingly, response
times decreased proportionally relative to target location.
Participants made faster judgments when the target chord
occurred in measure 11, (i.e., near the end of the 12-bar
progression) (M = 1736.2, SD = 7772.0). Response times
not only decreased toward the end of the progression but
variation between response times also decreased. An analy-
sis of variance reveals a statistically significant difference
between means for response time over location (F (1, 2) =
28.01, p < .001). These data suggest harmonic expecta-
tions become increasingly narrow with more contextual
evidence, thereby enabling faster response times as the
harmonic structure becomes more specific (see Table 2).
Response times by location and type
Overall analysis of variance between chord category
(i.e., expected, acceptable, and unacceptable) and target
location (i.e., measures, 5, 9, and 11) revealed statistically
significant differences for multiple combinations of loca-
tion and variation (F (1, 8) = 6.2796, p < .0001). A re-
peated measures analysis revealed statistically significant
differences for response times in the expected categories
for measures 5 (t = 2.77, p < .0059), 9 (t = -2.42, p <
.0158), and 11 (t = -3.23, p < .0013) and in the unaccept-
able categories for measures 9 (t = -2.42, p < .0158) and 11
ISBN 88-7395-155-4 © 2006 ICMPC
(t = -2.13, p < .0337). Response times to acceptable chords
in measure 5 also revealed a statistically significant differ-
ence (t = 4.00, p < .0001). These results show the influ-
ence of target location and chord type on response times,
though these influences do not appear to be uniform across
type or location. That is, both factors influenced response
times, but this influence was not uniform for all conditions.
Table 2. Average response time by target location.
Average Response Time (ms)
Target
Location Mean Standard Devia-
tion
Measure 5 2863.379 2419.348
Measure 9 1948.256 1123.396
Measure 11 1736.247 772.0454
Response times by target category
The average response time for all chords in all loca-
tions was much higher than expected (M = 2203.8, SD =
1704.9), which suggests that most participants made judg-
ments after the onset of the chord following the target
chord. With the tempo for each tune standardized at 120
beats per minute, there was a 2000 ms delay between
chords, so response times above 2000 ms suggest partici-
pants made judgments after the start of the succeeding
measure. In support of our hypothesis, participants made
faster judgments for chords in the expected category (M =
2096.2, SD = 1713.14). They did not, however, make simi-
larly rapid judgments for chords in the unacceptable cate-
gory (M = 2443.8, SD = 1912.8). Table 3 shows the aver-
age response times and standard deviations by target chord
category. However, an analysis of variance did not reveal
a statistically significant difference between response times
by category (F (1, 2) = 1.85, p > .05), but this figure ap-
proached significance when we collapsed chords from the
acceptable and unacceptable categories into a more general
“not expected” category (F (1, 1) = 3.68, p < .0555).
Table 3. Average response time by manipulation type.
Average Response Time (ms)
Manipulation Standard
Type Mean Deviation
Expected 2096.2471 1713.1466
Acceptable 2419.7818 1799.1169
Unaccept- 2443.7622 1912.7719
able
These results suggest the influence of chord category
on response time latencies, though the results are not statis-
tically significant. We suspect the lack of statistically sig-
nificant results reflects both the difficulty of the task (e.g.,
participants properly identified chords in the acceptable
category accurately 30 percent of the time), the ambiguity
of specific chord category, and the tendency to make judg-
1286
 ICMPC9 Proceedings
ments after the onset of the next chord. Though participants
were able to differentiate between acceptable and unac-
ceptable chords at levels above chance, they exhibited
more caution before rejecting a chord than we predicted.
An analysis of average response time by decision
choice shows a statistically significant interaction between
response choice and reaction time (F (1,2) = 6.61, p <
.0015). These results support the hypothesis that people
make faster judgments for expected and unacceptable
chords and slower judgments for acceptable chords but
suggest our chord categories do not necessarily correspond
with harmonic expectation either in all categories or in all
stimuli. That is, we suspect differences between response
times for target chords in each category are not statistically
significant because our stimuli are not categorized appro-
priately. We predict, that after recalibrating non-
categorical stimuli that response time variation will corre-
spond to target category more closely.
DISCUSSION
Although correct responses for expected and unaccept-
able chords reveal an ability to differentiate between ap-
propriate and inappropriate harmonies within the jazz-blues
style, ambiguity within the acceptable category underscores
the abstract nature of chord category. Expected chords
sound good, and therefore participants decided these
chords “fit well.” Unacceptable chords sound bad, so par-
ticipants determined these chords “did not fit.” However,
acceptable chords were more ambiguous, and participants
were not able to either integrate or reject these anomalies
quickly. Categorical decisions reveal a tendency to accept
chords in this category, but the relatively equal distribution
of response choices to chords in this category suggests an
inability to make precise judgments about these chords
either quickly or consistently.
In most studies examining harmonic perception, par-
ticipants choose between two categories (e.g., true or
false). We wanted to probe the possibility of graded expec-
tations and therefore used a middle category and a middle
response choice. The results are inconclusive but promis-
ing. The comparison of response time versus location sup-
ports our hypothesis that people generate hierarchical mod-
els for understanding harmonic structure during listening.
Shorter response times for later target chords suggest deci-
sions becomes easier as the model for harmonic structure
becomes more specific and that this specificity emerges in
relation to the amount of input received. Furthermore,
these data support our hypothesis that contextual informa-
tion affects judgments of fit and that these judgments are
not universal or schematic for Western tonal music in gen-
eral but rather they correspond to veridical (i.e., context-
based) expectations.
As results from this study suggest, people generate rich
mental models to organize and integrate harmonic material.
These results, however, do not provide a clear picture of
how that structure develops. We suspect based on the dif-
ferences in response times across location, that this process
ISBN 88-7395-155-4 © 2006 ICMPC
mirrors, if it does not replicate, the process used to generate
similar syntactic structures for and to gather meaning from
linguistic input. It would be interesting to compare behav-
ioral as well as neurophysiological responses to harmonic
targets against similar data from linguistic input. Results
from this study nonetheless demonstrate the perceptual
relevance of harmonic violations in conjunction with re-
sults from previous experiments examining similar ques-
tions about harmonic processing. The subtlety of the har-
monic manipulations we used in this study reveal the pres-
ence of veridical and possibly schematic expectations with
more precision than basic in-key and out-of-key judgments
suggest. This sensitivity warrants future inquiry into the
origins of harmonic expectation and the effect of contex-
tual cues.
In general, results from this study show that harmonic
structure is complex and abstract, and underscore some of
the difficulties encountered when attempting to categorize
artistic decisions. Chords that are acceptable in one context
may not be acceptable in another context despite schematic
similarities between these contexts, which makes it difficult
to judge chord appropriateness categorically. The jazz-
blues form determines which chords are acceptable within
this style, but the line separating acceptable from unaccept-
able is still ambiguous. Results from this study support the
hypothesis that while listening, multiple harmonic expecta-
tions exist and that these expectations become more refined
as listeners integrate more information into their mental
model of harmonic structure. These results also suggest the
influence of past experience on harmonic judgments, which
limits the listener’s ability to accommodate certain anoma-
lous harmonies but does not limit that ability completely.
Harmonic structures maintain some flexibility, which de-
creases but does not disappear entirely during listening.
While jazz-blues musicians familiar with this form may
reject certain chord substitutions within this style, artistic
inclination toward variety and experimentation may entice
listeners to temporarily accept marginally appropriate
chords. Though formal substitution rules prevent certain
substitutions for compositions in this style, curiosity and
intrigue create greater flexibility during experimental lis-
tening. As with all judgments of artistic appropriateness,
there is always room for personal taste.
While results from this study underscore the ambiguity
of chord category, they nonetheless reveal more specificity
in harmonic expectation than that which was predicted by
previous studies. By examining responses to in-key chord
substitutions, we were able to demonstrate the complexity
and specificity of harmonic expectation. These results not
only correspond to results from previous studies but also
build upon results from those studies. Although structural
processing of harmony seems similar to syntactic process-
ing, it is not clear at this point which factors specifically
influence or generate harmonic expectations. In all likeli-
hood, harmonic expectation stems from a variety of factors,
some of which we accounted for in this study and others of
1287
 ICMPC9 Proceedings
which we are still uncertain. We suspect that further in-
quiry into the context-specificity as well as the context-
flexibility of harmonic expectation will enable us to gener-
ate a better understanding of those factors that influence
expectations. Once we understand those expectations bet-
ter, we will have a clearer understanding of the cognitive
mechanisms underlying harmonic processing, which may
or may not provide evidence for similarity between har-
monic and syntactic processing.
ACKNOWLEDGMENTS
A Cognitive Science Summer Undergraduate Research
Fellowship and an Undergraduate Research Grant from
Northwestern University supported this research. The au-
thors would like to thank the Cognitive Science Program
and the Music Studies Department at Northwestern Univer-
sity for their continued support. Additionally, the authors
would like to thank their colleagues in Cognitive Science
and Music Cognition at Northwestern University for their
advice, support and encouragement. `
REFERENCES
Bernstein, L. (1973) The unanswered question: Six talks
at Harvard. Cambridge, Mass.: Harvard University Press.
Besson, M., & Schön, D. (2003). Comparison between
language and music. In I. Peretz, & R. Zatorre (Eds.), The
Cognitive Neuroscience of Music (pp. 269-293). Oxford:
Oxford University Press.
Besson, M., Faïta, F., Peretz, I., Bonnel, A.-M., & Requin,
J. (1998). Singing in the brain: Independence of lyrics and
tunes. Psychological Science, 9 (6), 494-498.
Bharucha, J. J., & Stoeckig, K. (1986). Reaction time and
musical expectancy: Priming of chords. Journal of Ex-
perimental Psychology: Human perception and perform-
ance, 12 (4), 403-410.
ISBN 88-7395-155-4 © 2006 ICMPC
Frazier, Lyn. (1989) Against lexical generation of syntax.
In W. Marslen-Wilson (Ed.) Lexical representation and
process. (pp. 505-528). Cambridge, MA: The MIT Press.
Frazier, L. (1987). Sentence processing: A tutorial review.
In M. Coltheart (Ed.), Attention and performance (pp. 559-
586). Hillsdale, NJ: Lawrence Erlbaum Associates.
Frazier, L., & Clifton, C., Jr. (1996). Construal. Cam-
bridge, MA: MIT Press.
Justus, T.C., & Bharucha, J.J. (2001). Modularity in musi-
cal processing: The automaticity of harmonic processing.
Journal of Experimental Psychology; Human perception
and performance 27(4), 1000-1011.
Koelsch, S., Gunter, T., Schröger, E., & Friederici, A. D.
(2003). Processing tonal modulations: An ERP study. Jour-
nal of Cognitive Neuroscience, 15 (8), 1149-1159.
Lerdahl, F., & Jackendoff, R. (1983). A generative theory
of tonal music. Cambridge, MA: MIT Press.
Patel, A.D., Gibson, E., Ratner, J., Besson, M., & Hol-
comb, P. J (1998). Processing syntactic relations in lan-
guage and music: An event-related potential study. Journal
of Cognitive Neuroscience, 10 (6), 717-733.
Sloboda, J. A. (1985). The Musical Mind: The Cognitive
Psychology of Music. Oxford: Clarendon Press.
Steedman, M.J. (1984). A generative grammar of jazz
chord sequences. Music Perception 2 (1), 52-77.
Steedman, M.J. (1996). The blues and the abstract truth:
Music and mental models. In A. Garnham, & J. Oakhill
(Eds.) Mental Models in Cognitive Science (pp. 305-318),
Hove: Psychology Press.
1288