Area Spt in the Human Planum Temporale Supports Sensory-Motor
Integration for Speech Processing
Gregory Hickok,1
Kayoko Okada,1
and John T. Serences2
1
Department of Cognitive Sciences and Center for Cognitive Neuroscience, University of California, Irvine, 2
Department of Psychology,
University of California, San Diego, California
Submitted 1 October 2008; accepted in final form 11 February 2009
Hickok G, Okada K, Serences JT. Area Spt in the human planum
temporale supports sensory-motor integration for speech processing. J
Neurophysiol 101: 2725–2732, 2009. First published February 18,
2009; doi:10.1152/jn.91099.2008. Processing incoming sensory information
and transforming this input into appropriate motor responses is
a critical and ongoing aspect of our moment-to-moment interaction
with the environment. While the neural mechanisms in the posterior
parietal cortex (PPC) that support the transformation of sensory inputs
into simple eye or limb movements has received a great deal of
empirical attention—in part because these processes are easy to study
in nonhuman primates—little work has been done on sensory-motor
transformations in the domain of speech. Here we used functional
magnetic resonance imaging and multivariate analysis techniques to
demonstrate that a region of the planum temporale (Spt) shows
distinct spatial activation patterns during sensory and motor aspects of
a speech task. This result suggests that just as the PPC supports
sensorimotor integration for eye and limb movements, area Spt forms
part of a sensory-motor integration circuit for the vocal tract.
I N T R O D U C T I O N
Although most research on sensory-motor integration has
focused on the neural mechanisms in posterior parietal cortex
(PPC) that support interactions between vision and motor
behaviors such as reaching and eye movements (Andersen
1997; Colby and Goldberg 1999; Cui and Andersen 2007),
sensory-motor interactions are also critically important for
speech-related behaviors including aspects of speech development
(Doupe and Kuhl 1999; Hickok and Poeppel 2000, 2007;
Hickok et al. 2000), sensory guidance of speech production
(Guenther et al. 1998; Hickok 2000; Hickok and Poeppel 2007;
Hickok et al. 2000; Warren et al. 2005), maintenance of parity
between speech perception and production (Galantucci et al.
2006; Mattingly and Liberman 1988), and verbal working
memory (Buchsbaum and D’Esposito 2008; Hickok et al.
2003; Jacquemot and Scott 2006). A region in the left posterior
planum temporale (PT) in humans, area Spt, has been proposed
as a critical node in a network that supports sensory-motor
integration for speech and other vocal tract behaviors (Buchsbaum
et al. 2001; Hickok and Poeppel 2000, 2004, 2007;
Hickok et al. 2003; Okada and Hickok 2006; Pa and Hickok
2008). Area Spt exhibits properties parallel to those found in
sensory-motor areas in the PPC of monkeys (Andersen 1997;
Colby and Goldberg 1999; Cui and Andersen 2007). First, Spt
has sensory-motor response properties, activating both during
the auditory perception and covert production of speech
(Buchsbaum et al. 2001, 2005b; Hickok et al. 2003; Okada and
Hickok 2006). Second, Spt appears to be selective for the vocal
tract effector system (Pa and Hickok 2008). Third, although
Spt activates during speech functions, it is not speech selective:
it responds to other vocal-tract related behaviors such as the
perception and covert production (humming) of melodies
(Hickok et al. 2003; Pa and Hickok 2008). Fourth, while Spt is
responsive to speech stimulation, it is not critical for speech
recognition (Hickok and Poeppel 2000, 2004, 2007), just as
PPC sensory-motor areas are not critical for object recognition
(Milner and Goodale 1995; Ungerleider and Mishkin 1982).
Fifth, human cytoarchitectonic studies (Galaburda and Sanides
1980) and comparative studies in monkeys (Smiley et al. 2007)
indicate that the posterior PT region is not part of unimodal
auditory cortex. Finally, the posterior PT appears to be multisensory
(Calvert and Campbell 2003; Calvert et al. 1997;
Smiley et al. 2007). These similarities between PPC sensorymotor
areas and Spt have led to the view that Spt supports
some form of sensory-motor interaction that is critical to
behaviors such as speech (Buchsbaum et al. 2001, 2005b;
Hickok and Poeppel 2004; Hickok et al. 2003; Warren et al.
2005) and more precisely for motor behaviors that involve the
vocal tract effector system (Hickok and Poeppel 2007; Pa and
Hickok 2008).
Here we use functional magentic resonance imaging (fMRI)
to test the hypothesis that subpopulations of neurons in Spt
play specialized roles in sensory-motor integration, just as PPC
sensory-motor areas contain subpopulations of cells with different
sensory-motor weightings (e.g., ϳ33% were motor dominant,
26% visual dominant, and 37% sensory-motor in one
study) (Sakata et al. 1995). Traditional fMRI analyses may
miss such fine-grained distinctions because high-dimensional
data sets are distilled into univariate estimates of the average
response amplitude across all voxels within the region of
interest (ROI). However, recently developed multivariate pattern
classification methods exploit the fact that if some voxels
contain more neurons of a particular persuasion (e.g., sensory
vs. motor), they may exhibit a weak response preference. By
pooling the output of many weakly selective voxels, it is
possible to use machine-learning algorithms to distinguish
characteristic voxel-by-voxel patterns of activation associated
with different sensory-motor functions (Haxby et al. 2001;
Kamitani and Tong 2005; Norman et al. 2006; Serences and
Boynton 2007a,b). Establishing the existence of distinct sensory
versus motor activation patterns would establish that
Address for reprint requests and other correspondence: G. Hickok, Center
for Cognitive Neuroscience, Department of Cognitive Sciences, University of
California, Irvine, CA 92697 (E-mail: greg.hickok@uci.edu) or to J. Serences,
Department of Psychology, University of California, San Diego, CA 92093
(E-mail: jserences@ucsd.edu).
J Neurophysiol 101: 2725–2732, 2009.
First published February 18, 2009; doi:10.1152/jn.91099.2008.
27250022-3077/09 $8.00 Copyright © 2009 The American Physiological Societywww.jn.org
distinct subpopulations of neurons in Spt support sensorymotor
integration for speech-related functions.
M E T H O D S
Subjects
Twenty-two participants (10 females) between 18 and 35 yr of age
were recruited from the University of California, Irvine (UCI), and
received monetary compensation for their time. The volunteers were
right-handed, native English speakers with normal or corrected-tonormal
vision, no known history of neurological disease, and no other
contraindications for MRI. Informed consent was obtained from each
participant prior to participation in the study in accordance with UCI
Institutional Review Board guidelines.
Materials and procedure
The data reported in this study were part of a larger experiment
aimed at mapping responses to a range of sensory stimuli, including
melodic sequences, noise bursts, auditory speech, and visual speech,
each presented in blocks of 15 s. All stimulus types were randomly
intermixed across the study and presented in equal ratios across the
experiment and within each session (run).
Here we focus only on the auditory speech conditions. The stimuli
were a set of “jabberwocky” sentences—sentences in which content
words were replaced with nonsense words (e.g., “It is the glandor in
my nedderop”)—taken from a previous study (Hickok et al. 2003).
There were three experimental conditions each presented in 15-s
blocks (trials), “continuous speech”: 15 s of listening to continuous
speech (sets of sentences), “listenϩrest”: 3 s of listening to speech
followed by 12 s of rest, “listenϩrehearse”: 3 s of listening to speech
followed by 12 s of covert (subvocal) rehearsal of the heard stimuli
(see Fig. 1). In addition, we included a null (rest) condition of the
same 15-s duration. Each of six sessions (runs) contained two trials
(15-s blocks) of each condition, including the null (rest) condition. A
visual cue distinguished between the conditions: a fixation cross for
the continuous listen condition that cued the subject to simply fixate
and listen, a picture of an ear for the listenϩrest condition that cued the
subject to listen and not rehearse after the offset of the auditory stimulus,
and a picture of a mouth for the listenϩrehearse condition that cued the
subject to listen and then covertly rehearse the auditory stimulus until the
end of the trial. These cues remained on the screen for the duration of the
trial. The listen-rehearse paradigm has been used in previous experiments
(Buchsbaum et al. 2001, 2005a,b; Hickok et al. 2003) and has been
shown to drive activity in posterior planum temporale (Spt), which
responds characteristically to both the listen and rehearse phases of the
trial. This line of investigation assumes that covert rehearsal is a valid
proxy for speech production. Evidence supporting this assumption comes
from studies of speech production in more conventional naming tasks that
demonstrate activity in this region (Graves et al. 2008; Levelt et al. 1998;
Okada and Hickok 2006; Okada et al. 2003) and that show sensory
(listening to words) and motor (naming pictures) overlap (Okada and
Hickok 2006).
The experiment started with a short exposure session to familiarize
subjects to all of the different experimental stimuli. Subjects were
scanned during the exposure session to ensure they could comfortably
hear the stimuli through the scanner noise and to acclimatize them to
the fMRI environment. This was followed by five experimental
sessions (runs). Each experimental session contained an equal number
of trials (blocks) of each condition and a single scanning session was
ϳ6 min long. Auditory stimuli were presented through an MR
compatible headset, and stimulus delivery and timing were controlled using
Cogent software (http://www.vislab.ucl.ac.uk/cogent_2000.php) implemented
in Matlab 6 (Mathworks). To monitor subjects’ attentiveness
during the scans we presented occasional “oddball” stimuli (e.g., a
speech stimulus presented in a female rather than male voice) to
which subjects made a button press. These trials (3/session, 13%)
were excluded from the subsequent analysis.
Scanning parameters
MRIs were obtained in a Philips Achieva 3T MR scanner (Philips
Medical Systems, Andover, MA) fitted with an eight-channel RF
receiver head coil, at the Research Imaging Center, UCI. We first
collected a total of 620 EPI volumes over 5 sessions using Fast Echo
EPI (sense reduction factor ϭ 2.0, FOV ϭ 220 ϫ 180, matrix ϭ
FIG. 1. Schematic diagram of the 3 conditions
analyzed in the present study. 1: the “continuous
listen” condition. 2: the “listenϩrest”
condition. And 3: the “listenϩrehearse condition.
2726 G. HICKOK, K. OKADA, J. T. SERENCES
J Neurophysiol • VOL 101 • MAY 2009 • www.jn.org
112 ϫ 112 mm , TR ϭ 3.0s, TE ϭ 25 ms, flip angle ϭ 70°, size ϭ
1.95 ϫ 1.95 ϫ 2 mm). After the functional scans, a high-resolution
anatomical image was acquired with an MPRAGE pulse sequence in
axial plane (matrix ϭ 256 ϫ 256 mm, TR ϭ 8 ms, TE ϭ 3.7 ms, flip
angle ϭ 8°, size ϭ 1 ϫ 1 ϫ 1 mm).
Data analysis
The first three and the last images of each session were discarded
prior to analysis. Preprocessing of the data were performed using
Statistical Parametric Mapping (SPM5; Wellcome Department of
Imaging Neuroscience, London, UK; www.fil.ion.ucl.ac.uk/spm) implemented
in Matlab7 (Mathworks). First, motion correction was
performed by creating a mean image from all of the volumes in the
experiment and then realigning all volumes to that mean image using
a six-parameter rigid-body transformation. Subsequent data analysis
involved two stages. The first stage identified sensory-motor ROIs
within the planum temporale (Spt) in individual subjects that were
subsequently used in the second stage of analysis. The second stage
used multivoxel pattern classification analysis to examine the spatial
distribution of activation within the Spt ROI (see following text).
ROI identification analysis
First-level analysis was performed on each subject using AFNI
software (Cox 1996). Due to the high anatomical variability of the
posterior Sylvian region (Knaua et al. 2006), single-subject analysis
was elected instead of group analysis. Images were smoothed with an
isotropic 4-mm full-width half-maximum (FWHM) Gaussian kernel.
Regression analysis was performed to find parameter estimates that
best explained variability in the data. Each predictor variable representing
the stimulus presentation time course for each event type was
convolved with a standard hemodynamic response function (Boynton
et al. 1996) and entered into the model along with six motion
regressors.
As noted, our goal in this stage of the analysis was to identify area
Spt in the posterior planum temporale. Spt is defined as a region
within the left Sylvian fissure posterior to Heschl’s gyrus that exhibits
both auditory and motor-related response properties (Buchsbaum et al.
2005a,b; Hickok et al. 2003; Okada and Hickok 2006; Pa and Hickok
2008). This activation is strongly left dominant (Hickok et al. 2003).
In the present study, sensory responsivity was measured using the
contrast between the continuous speech condition and rest (the null
rest blocks), while the motor-related response was measured using the
contrast, speechϩrehearse Ͼ speechϩrest (P Ͻ 0.001, uncorrected).
Thus in individual subjects, ROIs were defined by activations reflecting
the conjunction of continuous speech Ͼ null rest blocks, and
speechϩrehearse Ͼ speechϩrest that were located within the left
planum temporale region (within the Sylvian fissure posterior to
Heschl’s gyrus), defined by coregistering each subject’s activation
maps with their own structural MRIs. Using these criteria, significant
activations were found in 20 of the 22 subjects. ROIs in 14 of these
20 participants involved a sufficient number of voxels (Ն10) to allow
for pattern classification analysis (mean number of voxels Ϯ SE: 70 Ϯ
45). The analysis described in the following text was applied to this
resulting dataset; however, a qualitatively (and statistically) similar
pattern of results was obtained if all 20 subjects were included
(see Supplemental Fig. S11
).
Multivoxel pattern analysis
This analysis focused on the continuous speech condition and the
listenϩrehearse condition. The rationale for using these two conditions
is as follows. First, both of these conditions produce robust
activations in the Spt ROI (see following text), and second, the trial
structure is such that different time points within these conditions
yield different predictions regarding processes that should be neurally
distinguishable or not, providing a within-trial control for our classification
analysis. Specifically, for the first 3 s following trial onset, the
two conditions are similar in terms of their acoustic input properties.
They diverge only during the last 12 s of the trial where one condition
continues to involve sensory perception of speech and the other
involves (covert) speech production without sensory speech stimulation
(Fig. 1). Thus we expect increasing pattern discriminability as the
trial progresses.
To test this hypothesis, we first normalized the time series from
each voxel (using unsmoothed data) in each subject’s Spt ROI on a
run-by-run basis using a z-transform (to remove changes in mean
signal intensity that occur between scanning runs). Then we segmented
the normalized time series from each voxel within each Spt
ROI into three separate temporal bins that each contained activation
patterns measured on two successive TRs (0–3, 6–9, and 12–15 s post
stimulus, where each number indicates the onset time of a volume
acquisition). The pattern classification analysis was run separately on
data from each temporal bin. Data from all but one run were extracted
to form a “training” data set for the classification analysis; data from
the remaining run were defined as a “test” set (note that we use the term
“run” to refer to an entire 372-s data collection sequence so the training
and test data sets were always independent). We then trained a support
vector machine (Vapnik 1998) (the OSU-SVM implementation,
downloaded from http://sourceforge.net/projects/svm/) based only on
the training data and then used it to classify the task requirements
(continuous speech versus listenϩrehearse) on each trial in the test set
(see e.g., Kamitani and Tong 2005; Serences and Boynton 2007a,b;
Serences et al. 2009). Applying a SVM to the data set returns a weight
for each voxel, such that the weighted sum of the values across all
voxels is used to make the binary classification decision. Because zero
is the decision boundary, positive sums indicate one choice, and
negative sums the other. In the present experiment, positive weights
were assigned to voxels that responded more during “listen” trials, and
negative weights were assigned to voxels that responded more during
“rehearse” trials; the most discriminating voxels of each class had the
highest absolute values. To ensure that the classification algorithm
was only using information about the spatially distributed voxel-byvoxel
activation pattern—as opposed to amplitude differences between
the conditions—we explicitly subtracted the mean activation
level from each activation pattern before classification (this subtraction
was carried out on a trial-by-trial basis, so the mean of the
activation pattern associated with each trial was removed). This
procedure was repeated using a ‘hold-one-run-out’ cross validation
approach so that data from every scan were used as a test set in turn.
Because each subject completed five runs/session, the overall classification
accuracy for a subject was defined as the average classification
accuracy across all five possible permutations of holding one run
out as the test set and using the remaining runs as a training set. After
computing classification accuracy within each of the three time bins
separately for each of the subjects, we averaged the data across
observers.
As a check to ensure that our analysis path was valid, we also
verified that a data set composed of independent identically distributed
(IID) noise did not yield above-chance classification accuracy
when processed through the same analysis code (because there was no
internally reliable signal). In addition, we repeated the analysis 10,000
times after randomly permuting the condition label assigned to each
pattern to ensure that the code and/or the algorithm did not always
produce above chance accuracy. (See Nichols and Holmes 2002 for a
tutorial on permutation processes.) The dashed blue lines in Fig. 4
(and Supplemental Fig. S1) show classification accuracy on the upper
and lower 5% of the 10,000 repetitions of the analysis using permuted
trial labels, and all of our effects of interest survive these thresholds.1
The online version of this article contains supplemental data.
2727SENSORY-MOTOR ORGANIZATION OF Spt
J Neurophysiol • VOL 101 • MAY 2009 • www.jn.org
R E S U L T S
Overall performance accuracy on the oddball detection task
was 98%. This indicates that subjects were alert and attentive
during the scanning procedure.
The Spt ROI localizer analysis (listenϩrehearse Ͼ listenϩrest
and continuous listen Ͼ null rest blocks) detected reliable
activation within the left posterior PT in 20 of the 22 subjects.
Fourteen of these subjects had ROIs that contained a sufficient
number of voxels (Ն10) to allow for pattern classification
analysis. The results reported here focus on these 14 subjects.
The location of sensory-motor ROIs were confirmed to be
within the planum temporale (area Spt) in each subject based
on coregistration of the functional images with a high-resolution
MRI of each subject’s own brain. The location of these
activations in standardized space is presented in Table 1 and is
displayed on a standardized brain image in Fig. 2 (top row).
Activation foci in two individual subjects is shown in Fig. 2
(bottom row). The location of these ROIs is consistent with
previous studies of area Spt (Buchsbaum et al. 2001, 2005b;
Hickok et al. 2003).
The mean amplitude of the hemodynamic response for the
listenϩrehearse and the continuous listen conditions, collapsed
across all voxels in the Spt ROIs, is presented in Fig. 3. An
ANOVA carried out on these data confirmed that both conditions
produced significant activations in the ROI [main effect
of time point: F(5,65) ϭ 25.17, P ϭ 0.001] and that the
activation levels for the two conditions differed during some
time points [condition ϫ time point interaction: F(5,65) ϭ
3.41, P ϭ 0.008]. During the first two time points, when the
sensory-evoked response dominates the signal, the amplitudes
are statistically identical (P Ͼ 0.59) for the two conditions.
Beginning at time point 3, the curves begin to diverge with the
listenϩrehearse condition yielding greater amplitude [t(13) ϭ
2.49, P ϭ 0.027], presumably because the rehearsal process
produces some additional activation in Spt which sums with the
residual sensory response. The response in the listenϩrehearse condition
peaks and reaches its greatest difference from the continuous
listen condition at time point 4 [t(13) ϭ 3.15, P ϭ
0.008] and then declines to equal the continuous listen condition
in the last two time points [t(13) ϭ 1.19, P ϭ 0.25 and
t(13) ϭ 0.08, P ϭ 0.94, respectively]. Because rehearsal
continues for the duration of the block, this decline in amplitude
presumably reflects the decay of the sensory component of
the BOLD response. Note that our predictions regarding the
evolution of pattern discriminability does not track the observed
amplitude differences. Specifically we expect the
greatest discriminability not at the peak of the amplitude
difference between the two conditions, but at the end of the
trial when the amplitude signals are not statistically different
between the two conditions, i.e., when the signals are
dominated by different functional sources (e.g., sensory vs.
motor). It is this last time bin therefore that represents the
strongest test of our hypothesis.
TABLE 1. Talairach coordinates for the center of mass of the Spt
ROI in each of 14 subjects that were entered into the pattern
classification analysis
ROI. Area Spt
S1 Ϫ47 Ϫ35 18
S2 Ϫ52 Ϫ40 22
S3 Ϫ52 Ϫ43 28
S4 Ϫ53 Ϫ50 15
S5 Ϫ49 Ϫ44 23
S6 Ϫ53 Ϫ39 22
S7 Ϫ51 Ϫ37 18
S8 Ϫ55 Ϫ47 22
S9 Ϫ49 Ϫ35 18
S10 Ϫ40 Ϫ41 14
S11 Ϫ61 Ϫ33 11
S12 Ϫ42 Ϫ32 16
S13 Ϫ40 Ϫ32 26
S14 Ϫ59 Ϫ47 17
ROI, region of interest; Spt, a region in the planum temporale.
FIG. 2. Top row: location of the each subject’s region of the
planum temporale (Spt) region of interest (ROI) presented on a
standardized 3-dimesensional (3D) brain image. Note that all
activation foci were individually confirmed to be inside the
posterior Sylvian fissure based on co-registration of the activation
map onto each subject’s own structural brain image. Bottom
row: location of the Spt ROI (green crosshairs) in 2 subjects
projected onto each subject’s own anatomical magnetic resonance
imaging (MRI) scan.
2728 G. HICKOK, K. OKADA, J. T. SERENCES
J Neurophysiol • VOL 101 • MAY 2009 • www.jn.org
To critically test the hypothesis that the Spt contains functionally
distinct subdivisions of neurons, we next used a
multivariate pattern classification analysis to determine if the
voxel-by-voxel activation patterns within Spt discriminated
speech perception (continuous listen condition) from speech
production (listenϩrehearse condition). In contrast to the time
course of the mean BOLD amplitudes, which reconverged at
the end of the trial, the activation patterns associated with each
condition became more separable and pattern classification
accuracy increased across the duration of the trial [1-way
repeated-measures ANOVA, F(2,26) ϭ 6.7, P Ͻ 0.01]. Specifically,
classification accuracy was not significantly different
from chance in the first time bin [0–3 s, t(13) ϭ 0.538, P ϭ
0.60] as expected. After this point, classification accuracy
improves to above-chance levels [6–9 s, t(13) ϭ 3.58, P Ͻ
0.0034] and reaches ϳ75% accuracy in the final time bin
[t(13) ϭ 3.717, P Ͻ 0.0004, Fig. 4; Supplemental Fig. S1].
Figure 5 shows a map of SVM weights within the area Spt
ROI for two representative subjects; the relatively interdigitated
distribution of positively and negatively weighted voxels
implies that this region contains a mixture of neurons that are
preferentially responsive to one cognitive operation over the
other.
There are two potential problems with the above analysis.
One is that the voxel selection procedure and the pattern
classification analysis are not completely independent (Vul
et al. 2009). The other is that visual cue for the listenϩrehearse
condition contained a namable object (mouth), whereas the
continuous listen condition involved a simple fixation crosshair.
It is conceivable that this visual cue difference is driving
the pattern discrimination difference. To assess these possibilities,
we used the even runs (sessions) to identify our ROI then
ran the pattern classification analysis on data from only the odd
runs (sessions). This was possible in 14 of our subjects using a
slightly relaxed threshold of 0.005 (using only 2 runs to
identify the ROIs necessarily reduces power). This split plot
approach removes any possible bias. In addition, not only did
we (re)assess pattern classification accuracy on the continuous
listen versus listenϩrehearse, but we also assessed pattern classification
accuracy on the continuous listen versus listenϩrest
condition that served as a control for the effects of the visual
cue as this condition also contained a nameable cue (ear). We
expected that continuous listen versus listenϩrest would not
result in above-chance classification.
The results of these analyses are presented in Fig. 6. Figure
6A shows the average hemodynamic response in the odd runs
across subjects (n ϭ 14) for three conditions, continuous listen,
listenϩrehearse, and listenϩrest. Notice that in the first 6 s of
the trial, which is dominated by the sensory response, signal
amplitude is equivalent across conditions. After this point, the
FIG. 3. Mean blood-oxygen-level-dependent (BOLD) signal amplitude
time course (z-score) for the continuous listen and listenϩrehearse conditions
in the Spt ROI for 6 time points starting with trial onset. Speech stimuli were
presented at time point 0, and trials ended at time point 12; time point 15 is the
same as time point 0 in the next (random) trial and is included to show the
continued evolution of the response function for these conditions.
FIG. 4. A: pattern classification accuracy as a function of time within the
trial. Data are organized into 3 bins corresponding to early, middle, and late
portions of the trial. Chance classification accuracy ϭ 0.50. Conditions are
classified significantly better than chance in the 2nd and 3rd bins only, two
asterisks, P Ͻ 0.001, three asterisks, P Ͻ 0.0001. Dashed lines indicate the
upper and lower 5% of classification accuracies observed using a permutation
test (see METHODS). B: BOLD amplitude for the 2 test conditions organized into
the same time bins used in the pattern classification analysis for comparison.
Note that classification accuracy does not track with amplitude differences
between the 2 conditions.
2729SENSORY-MOTOR ORGANIZATION OF Spt
J Neurophysiol • VOL 101 • MAY 2009 • www.jn.org
response in the three conditions diverge. In the continuous
listen condition, the signal saturates and remains roughly the
same for the balance of the trial. In the listenϩrest condition,
the signal drops off reflecting a return to baseline after speech
stimulation is ended. In the listenϩrehearse condition, the
signal continues to increase, reaching its peak at 9 s presumably
because the sensory response sums with the motor-related
rehearsal response during this time window, then falls off
during the final two time points, presumably reflecting the
decay of the sensory response; the signal remains well above a
resting baseline, however, due to the rehearsal. Again it is in
this final phase of the trial when we expect to see maximal
discriminability between the continuous listen and listenϩrehearse
conditions because the source of the activation, although
equivalent in amplitude, derives from different underlying
sources: sensory versus motor. Patten classification in the fully
unbiased (odd run) dataset again confirmed this prediction
(Fig. 6B): classification accuracy was significantly above
chance at the final time point [t(13) ϭ 3.017, P ϭ 0.01,
2-tailed, 0.03 after Bonferoni correction]. The same analysis
carried out on the continuous listen versus listenϩrest condition
yielded no above-chance classifications (all P values Ͼ
0.19, 2-tailed, Ͼ0.50 after Bonferroni correction; Fig. 6C). We
conclude from these analyses that pattern classification accuracy
in the continuous listen versus listenϩrehearse conditions
is not an artifact of selection bias nor attributable to visual cue
differences.
D I S C U S S I O N
A region in the posterior planum temporale, area Spt, is
activated both by the perception and production of speech
(Buchsbaum et al. 2001, 2005a,b; Hickok et al. 2003; Okada
and Hickok 2006). The present study showed that the pattern of
FIG. 5. Activity pattern within area Spt.
Applying a SVM to the data set returns a
weight for each voxel such that the weighted
sum of the values across all voxels is used to
make the binary classification decision. Because
0 is the decision boundary, positive
sums indicate one choice, and negative sums
the other. In the present experiment, positive
weights were assigned to voxels that responded
more during “listen” trials, and negative
weights were assigned to voxels that
responded more during “rehearse” trials; the
most discriminating voxels of each class had
the highest absolute values. Figure shows a
map of SVM weights within the area Spt
ROI for 2 representative subjects.
FIG. 6. BOLD amplitude and pattern
classification accuracy in the split-plot
analysis. A: BOLD amplitude in area Spt
for the 3 conditions, continuous listen,
listenϩrehearse, and listenϩrest, in odd
runs only when Spt was defined using independent
data from even runs. B: pattern
classification accuracy for continuous listen
vs. listenϩrehearse as a function of
time within the trial. C: pattern classification
accuracy for continuous listen versus
listenϩrest as a function of time within the
trial. Chance classification accuracy ϭ 0.50,
and horizontal lines indicate the upper and
lower 5% of classification accuracies observed
using a permutation test (see METHODS).
Asterices, P Ͻ 0.001.
2730 G. HICKOK, K. OKADA, J. T. SERENCES
J Neurophysiol • VOL 101 • MAY 2009 • www.jn.org
activity across voxels within Spt is different during speech
perception compared with speech production-related processes
(covert rehearsal in the present study). Pattern classification
analysis revealed that the pattern of activity in Spt correctly
predicted whether a trial involved speech perception or speech
production at better than chance levels (ϳ70% accuracy). This
was particularly true in the final portion or the trial when the
source of the signals was maximally different (sensory vs.
motor-related). This classification cannot be attributed to general
amplitude differences between the conditions because the
amplitude was normalized across the conditions for the analysis
and because even unnormalized amplitude was not significantly
different during the final portion of the trial when
discrimination accuracy was maximal. Instead the result demonstrates
that the spatial pattern of activation across the ROI
differs for speech perception and production, which may result
from different spatial distributions of sensory and motor
(and/or sensory-motor) responsive cell types across voxels in
the ROI. The distinction in activation pattern for sensory
versus motor conditions in Spt also argues against the possibility
that motor response properties of Spt result from auditory
imagery as well as the possibility that the sensory responses in
Spt result from subvocal rehearsal: if sensory and motor
activations result in distinguishable activation patterns, they
are unlikely to be a consequence of the same process.
It is worth noting that the sensory-motor nature of the
response of Spt appears to be reflected as well in the BOLD
amplitude response, particularly in the fact that the listenϩrehearse
condition yields greater activation than the continuous speech
condition in the middle time points of the trial, but then falls
back to the level of the continuous speech condition by the end
of the trial (Figs. 3, 4B, and 6A). If the response in Spt was
purely sensory (e.g., if the “motor” response were simply
auditory imagery), there is no explanation for why the signal is
greater in the listenϩrehearse condition where there is less
auditory input. If, on the other hand, there are distinct populations
of cells that are sensory- versus motor-weighted, then
the increased amplitude for the listenϩrehearse condition in
the middle time bin can be explained as the summed hemodynamic
response resulting from activity of these different cell
types which would be evident in the mid-trial phase. By the
end of the trial, the sensory contribution to the summed activity
will have decayed in the listenϩrehearse condition resulting in
a drop in the overall activation level as observed. The
listenϩrest condition, which does not involve rehearsal, does
not show this pattern but instead results in similar activity in
the middle time points of the trial, and far less still in the final
time points of the trial (Fig. 6A). Thus the pattern of BOLD
amplitude activity is readily explainable on a sensory-motor
account of Spt function consistent with our interpretation of the
pattern classification results.
If our conclusion is correct that Spt contains both sensoryweighted
and motor-weighted classes of cell types, it would
indicate a further parallel between area Spt and sensory-motor
integration areas in the posterior parietal lobe. Like parietal
lobe sensory-motor integration areas, Spt exhibits both sensory
and motor response properties (Buchsbaum et al. 2001;
Dhankhar et al. 1997; Hickok et al. 2003), is relatively selective
for motor modality (vocal tract) (Pa and Hickok 2008),
appears to be multi-sensory (Dhankhar et al. 1997), is functionally
connected to frontal motor areas (BA 44 in particular)
(Buchsbaum et al. 2001), and appears to contain both sensoryand
motor-weighted cell types (present study). This set of
properties suggests that area Spt is a part of the collection of
regions in the posterior parietal cortex that supports sensorymotor
integration (Andersen 1997) with Spt tied to actions
associated with the vocal tract (Hickok and Poeppel 2007; Pa
and Hickok 2008).
Damage to tissue in the vicinity of area Spt has been
associated with conduction aphasia (Damasio and Damasio
1983, 1980). Just as disruption of sensory-motor areas in the
PPC result in disrupted motor function while sparing sensory
recognition abilities (Milner and Goodale 1995; Ungerleider
and Mishkin 1982), conduction aphasia is primarily a deficit of
speech production while sparing speech recognition (Benson
et al. 1973; A. R. Damasio 1991; Damasio and Damasio 198,
1980; Goodglass 1992, 1993; Goodglass and Kaplan 1983).
Patients with conduction aphasia make frequent sound-based
(phonemic) errors in their speech output and relatively few
meaning-based errors, which are more prevalent in aphasic
syndromes, such as Wernicke’s aphasia, associated with more
inferior temporal lobe damage (A. R. Damasio 1992; H.
Damasio 1991; Hillis 2007). Conduction aphasics have difficulty
in the verbatim repetition of speech, particularly with
low-frequency utterances or pseudowords (Goodglass 1992).
Verbatim repetition requires reference to a sensory-phonological
trace for accurate reproduction—a requirement that is
exaggerated with unfamiliar items or pseudowords—and therefore
is a behavior that would be substantially impacted by
damage to a speech-related sensory-motor integration system
(Hickok 2000; Hickok et al. 2000). For these reasons, conduction
aphasia has been interpreted as a syndrome that results
from damage to the sensory-motor integration network in the
posterior planum temporale, area Spt (Hickok 2001, 2000;
Hickok and Poeppel 2004; Hickok et al. 2000, 2003). Thus the
clinical effects of lesions involving area Spt are consistent with
the proposed sensory-motor functions that region.
R E F E R E N C E S
Andersen R. Multimodal integration for the representation of space in the
posterior parietal cortex. Philos Trans R Soc Lond B Biol Sci 352: 1421–
1428, 1997.
Benson DF, Sheremata WA, Bouchard R, Segarra JM, Price D, Geschwind
N. Conduction aphasia: a clincopathological study. Arch Neurol 28: 339–346,
1973.
Boynton GM, Engel SA, Glover GH, Heeger DJ. Linear systems analysis of
functional magnetic resonance imaging in human V1. J Neurosci 16:
4207–4221, 1996.
Buchsbaum B, Hickok G, Humphries C. Role of left posterior superior
temporal gyrus in phonological processing for speech perception and production.
Cogn Sci 25: 663–678, 2001.
Buchsbaum BR, D’Esposito M. The search for the phonological store: from
loop to convolution. J Cogn Neurosci 20: 762–778, 2008.
Buchsbaum BR, Olsen RK, Koch P, Berman KF. Human dorsal and ventral
auditory streams subserve rehearsal-based and echoic processes during
verbal working memory. Neuron 48: 687–697, 2005a.
Buchsbaum BR, Olsen RK, Koch PF, Kohn P, Kippenhan JS, Berman KF.
Reading, hearing, and the planum temporale. Neuroimage 24: 444–454,
2005b.
Calvert GA, Bullmore ET, Brammer MJ, Campbell R, Williams SCR,
McGuire PK, Woodruff PWR, Iversen SD, David AS. Activation of
auditory cortex during silent lipreading. Science 276: 593–596, 1997.
Calvert GA, Campbell R. Reading speech from still and moving faces: The
neural substrates of visible speech. J Cogn Neurosci 15: 57–70, 2003.
Colby CL, Goldberg ME. Space and attention in parietal cortex. Annu Rev
Neurosci 22: 319–349, 1999.
2731SENSORY-MOTOR ORGANIZATION OF Spt
J Neurophysiol • VOL 101 • MAY 2009 • www.jn.org
Cox RW, AFNI: Software for analysis and visualization of functional magnetic
resonance neuroimages. Comput Biomed Res 29: 162–173, 1996.
Cui H, Andersen RA. Posterior parietal cortex encodes autonomously selected
motor plans. Neuron 56: 552–559, 2007.
Damasio AR. Signs of aphasia. In: Acquired Aphasia, edited by Sarno MT.
San Diego, CA: Academic, 1991, p. 27–43.
Damasio AR. Aphasia. N Engl J Med 326: 531–539, 1992.
Damasio H. Neuroanatomical correlates of the aphasias. In: Acquired aphasia,
edited by Sarno M. San Diego, CA: Academic, 1991, p. 45–71.
Damasio H, Damasio AR. The anatomical basis of conduction aphasia. Brain
103: 337–350, 1980.
Damasio H, Damasio AR. Localization of lesions in conduction aphasia. In:
Localization in Neuropsychology, edited by Kertesz A. San Diego, CA:
Academic, 1983, p. 231–243.
Dhankhar A, Wexler BE, Fulbright RK, Halwes T, Blamire AM, Shulman
RG. Functional magnetic resonance imaging assessment of the human brain
auditory cortex response to increasing word presentation rates. J Neurophysiol
77: 476–483, 1997.
Doupe AJ, Kuhl PK. Birdsong and human speech: common themes and
mechanisms. Annu Rev Neurosci 22: 567–631, 1999.
Galaburda A, Sanides F. Cytoarchitectonic organization of the human auditory
cortex. J Comp Neurol 190: 597–610, 1980.
Galantucci B, Fowler CA, Turvey MT. The motor theory of speech perception
reviewed. Psychon Bull Rev 13: 361–377, 2006.
Goodglass H. Diagnosis of conduction aphasia. In: Conduction Aphasia,
edited by Kohn SE. Hillsdale, NJ: Erlbaum, 1992, p. 39–49.
Goodglass H. Understanding Aphasia. San Diego, CA: Academic, 1993.
Goodglass H, Kaplan E. The assesasment of apahasia and Related disordears.
Philadelphia: Lea and Febiger, 1983.
Graves WW, Grabowski TJ, Mehta S, Gupta P. The left posterior superior
temporal gyrus participates specifically in accessing lexical phonology. J
Cogn Neurosci 20: 1698–1710, 2008.
Guenther FH, Hampson M, Johnson D. A theoretical investigation of
reference frames for the planning of speech movements. Psychol Rev 105:
611–633, 1998.
Haxby JV, Gobbini MI, Furey ML, Ishai A, Schouten JL, Pietrini P.
Distributed and overlapping representations of faces and objects in ventral
temporal cortex. Science 293: 2425–2430, 2001.
Hickok G. Speech perception, conduction aphasia, and the functional neuroanatomy
of language. In: Language and the Brain, edited by Grodzinsky Y,
Shapiro L, Swinney D. San Diego, CA: Academic, 2000, p. 87–104.
Hickok G. Functional anatomy of speech perception and speech production:
psycholinguistic implications. J Psycholinguistic Res 30: 225–234, 2001.
Hickok G, Buchsbaum B, Humphries C, Muftuler T. Auditory-motor
interaction revealed by fMRI: speech, music, and working memory in area
Spt. J Cogn Neurosci 15: 673–682, 2003.
Hickok G, Erhard P, Kassubek J, Helms-Tillery AK, Naeve-Velguth S,
Strupp JP, Strick PL, Ugurbil K. A functional magnetic resonance
imaging study of the role of left posterior superior temporal gyrus in speech
production: implications for the explanation of conduction aphasia. Neurosci
Lett 287: 156–160, 2000.
Hickok G, Poeppel D. Towards a functional neuroanatomy of speech perception.
Trends Cognit Sci 4: 131–138, 2000.
Hickok G, Poeppel D. Dorsal and ventral streams: a framework for understanding
aspects of the functional anatomy of language. Cognition 92:
67–99, 2004.
Hickok G, Poeppel D. The cortical organization of speech processing. Nat Rev
Neurosci 8: 393–402, 2007.
Hillis AE. Aphasia: progress in the last quarter of a century. Neurology 69:
200–213, 2007.
Jacquemot C, Scott SK. What is the relationship between phonological
short-term memory and speech processing?. Trends Cogn Sci 10: 480–486,
2006.
Kamitani Y, Tong F. Decoding the visual and subjective contents of the
human brain. Nat Neurosci 8: 679–685, 2005.
Knaua TA, Bollich AM, Corey DM, Lemen LC, Foundas AL. Variability
in perisylvian brain anatomy in healthy adults. Brain Language 97: 219–
232, 2006.
Levelt WJM, Praamstra P, Meyer AS, Helenius P, Salmelin R. An MEG
study of picture naming. J Cogn Neurosci 10: 553–567, 1998.
Mattingly IG, Liberman AM. Specialized perceiving systems for speech an
other biologically significant sounds. In: Auditory Function: Neurological
Bases of Hearing, edited by Edelman GMG, Gall WE, Cowan WM. New
York: Wiley, 1988, p. 775–793.
Milner AD, Goodale MA. The Visual Brain in Action. Oxford, UK: Oxford
Univ. Press, 1995.
Nichols TE, Holmes AP. Nonparametric permutation tests for functional
neuroimaging: a primer with examples. Hum Brain Mapp 15: 1–25, 2002.
Norman KA, Polyn SM, Detre GJ, Haxby JV. Beyond mind-reading:
multi-voxel pattern analysis of fMRI data. Trends Cogn Sci 10: 424–430,
2006.
Okada K, Hickok G. Left posterior auditory-related cortices participate both
in speech perception and speech production: neural overlap revealed by
fMRI. Brain Lang 98: 112–117, 2006.
Okada K, Smith KR, Humphries C, Hickok G. Word length modulates
neural activity in auditory cortex during covert object naming. Neuroreport
14: 2323–2326, 2003.
Pa J, Hickok G. A parietal-temporal sensory-motor integration area for the
human vocal tract: evidence from an fMRI study of skilled musicians.
Neuropsychologia 46: 362–368, 2008.
Sakata H, Taira M, Murata A, Mine S. Neural mechanisms of visual
guidance of hand action in the parietal cortex of the monkey. Cereb Cortex
5: 429–438, 1995.
Serences JT, Boynton GM. Feature-based attentional modulations in the
absence of direct visual stimulation. Neuron 55: 301–312, 2007a.
Serences JT, Boynton GM. The representation of behavioral choice for
motion in human visual cortex. J Neurosci 27: 12893–12899, 2007b.
Serences JT, Ester E, Vogel E, Awh E. Stimulus-specific delay activity in
human primary visual cortex. Psychological Science 20: 201–214, 2009.
Smiley JF, Hackett TA, Ulbert I, Karmas G, Lakatos P, Javitt DC,
Schroeder CE. Multisensory convergence in auditory cortex. I. Cortical
connections of the caudal superior temporal plane in macaque monkeys.
J Comp Neurol 502: 894–923, 2007.
Ungerleider LG, Mishkin M. Two cortical visual systems. In: Analysis of
Visual Behavior, edited by Ingle DJ, Goodale MA, Mansfield RJW. Cambridge,
MA: MIT Press, 1982, p. 549–586.
Vapnik VN. Statistical Learning Theory. New York: Wiley, 1998.
Vul E, Harris C, Winkilman P, and Pashler H. Voodoo correlations in
social neuroscience. Perspect Psychol Sci In press.
Warren JE, Wise RJ, Warren JD. Sounds do-able: auditory-motor transformations
and the posterior temporal plane. Trends Neurosci 28: 636–643,
2005.
2732 G. HICKOK, K. OKADA, J. T. SERENCES
J Neurophysiol • VOL 101 • MAY 2009 • www.jn.org