Neuropsychologia 43 (2005) 807–814

Overt sentence production in event-related fMRI

Sven Hallera,∗ , E.W. Radueb , Michael Erbc , Wolfgang Groddc , Tilo Kirchera
a Department of Psychiatry, University of Tübingen, Osianderstr. 24, D-72076 Tübingen, Germany
b Division of Neuroradiology, Department of Diagnostic Radiology, University Hospital Basel, Switzerland
c Department of Experimental MR of the CNS, University of Tübingen, D-72076 Tübingen, Germany

Abstract

The use of syntactic structures on a sentence level is a unique human ability. Functional imaging studies have usually investigated syntax
comprehension. However, language production may be performed by different neuronal resources. We have investigated syntax generation
on a sentence level with functional magnetic resonance imaging (fMRI).
BOLD contrast was measured while subjects articulated utterances aloud. In the active condition ‘sentence generation’ (SG), subjects
had to produce subject verb object (SVO) sentences (e.g. “The child throws the ball”) according to syntactically incomplete stimuli (e.g.
“throw ball child”) presented visually. In the control condition ‘word reading’ (WR), subjects had to read identical stimuli without completing
the syntactic structure, while in a second control condition ‘sentence reading’ (SR), subjects had to read complete sentences. The semantic
meaning of all expressions was obvious despite the syntactically incomplete structure in conditions SG and WR.
In both contrasts, SG minus WR and SG minus SR, activation was mainly present in the left inferior frontal (BA 44/45) and medial frontal
(BA 6) gyri, the superior parietal lobule (BA 7) and the right insula (BA 13). A region of interest analysis revealed significantly stronger
left-dominant activation in BA 45 compared to BA 44.
Our data illustrates the crucial involvement of the left BA 45 in syntactic encoding and is in line with more recent imaging and brain lesion
data on syntax processing on a sentence level, emphasizing the involvement of a distributed left and right hemispheric network in syntax
generation.
© 2004 Elsevier Ltd. All rights reserved.

Keywords: Speech; Language; Syntax

1. Introduction a further distinction has to be made between production and

Fundamental aspects of language include syntax, i.e.
grammatically well formed words and sentences, and seman-
tics, i.e. the meaning of a word or phrase. However, syntax and
semantics are inseparable in natural language. Compare as an
example the headlines “Dog bites man” and “Man bites dog”.
Regarding the neural processes underlying human language,
Abbreviations: SG, sentence generation; SR, sentence reading; WR,
word reading; SVO, subject verb object; fMRI, functional magnetic reso-
nance imaging; BA, Brodmann’s area; GLM, general linear model; TAL,
Talairach space; FDR, false discovering rate; VOI, volume of interest; WM,
working memory; DLPFC, dorsolateral prefrontal cortex
∗ Corresponding author. Present address: Department of Neuroradiology,
University of Basel, Kantonsspital BS, CH-4056 Basel, Switzerland.
Tel.: +41 61 2652525.
E-mail address: shaller@uhbs.ch (S. Haller).
comprehension. During speech production, a mental concept
is expressed (Kircher et al., 2004). Words with appropriate se-
mantic meaning are retrieved from a store and an expression
is created according to syntactic rules (Levelt, 1989, 1995).
During speech perception, the inverse process is performed.
The final result of the comprehension process is the idea or
the concept of a given expression. There is evidence from
brain lesion studies involving patients with aphasia that dis-
tinct neural networks compute these two processes (Ferstl,
Guthke, & von Cramon, 1999; Munhall, 2001). In particular,
it has been shown that aphasic patients are unable to produce
syntactically correct sentences, while being able to judge the
syntactical correctness of presented sentences (Linebarger,
Schwartz, & Saffran, 1983).
Most previous neuroimaging studies on language pro-
cessing on a sentence level focused on the investigation of

0028-3932/$ – see front matter © 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.neuropsychologia.2004.09.007
808 S. Haller et al. / Neuropsychologia 43 (2005) 807–814
comprehension (Cabeza & Nyberg, 2000; Indefrey & Lev-
elt, 2000; Kaan & Swaab, 2002). Many of these studies used
sentences with spelling errors or wrong word order (Sakai,
Hashimoto, & Homae, 2001) to test for different aspects of
syntactic processing. Subjects usually performed plausibil-
ity judgments, i.e. they had to decide whether a sentence
was correct or not (Embick, Marantz, Miyashita, O’Neil,
& Sakai, 2000). In other studies, subjects had to decide as
to whether or not two sentences were similar (Dapretto &
Bookheimer, 1999). Language production should ideally be
investigated by analyzing natural speech (Kircher, Brammer,
Williams, & McGuire, 2000). However, it is difficult to con-
trol for the underlying cognitive processes involved in the
generation of longer utterances. Therefore, the majority of
previous functional neuroimaging studies on language pro-
duction reduced verbal output to single words. Particularly
well investigated are verbal fluency tasks, where subjects
are instructed to produce words starting with a particular
letter or of a semantic category (Cabeza & Nyberg, 2000).
However, higher language properties such as syntax or the
meaning of a phrase cannot be covered by these paradigms.
Consequently, little is known about the cerebral substrates
of sentence-level production processes (Indefrey & Levelt,
2000). In the only systematic imaging study (using positron
emission tomography, PET) on syntax during language pro-
duction on a sentence level (Indefrey, Brown, et al., 2001),
subjects had to generate utterances of different syntactic com-
plexity that described the motion of visually presented ob-
jects. The authors provided evidence that a region caudally
adjacent to Broca’s area is involved in both sentence-level
and local (phrase-level) syntactic encoding during speak-
ing.
In the present study, we investigated syntax production
on a sentence level. Subjects had to produce simple subject
verb object (SVO) sentences like e.g. “The child throws the
ball”. We systematically varied the demand on syntax produc-
tion in three conditions, while subjects spoke overtly. We hy-
pothesed increasing signal changes in the left inferior frontal
gyrus (IFG) with increased syntactic production demand.
2. Methods
2.1. Subjects
Fifteen right-handed (score ranging from 14 to 20 in
the online version of the Edinburgh Handedness Inventory
(Oldfield, 1971), http://airto.loni.ucla.edu [scores range from
−20 (left handed) to +20 (right handed)]) male subjects (aged
19–44, mean 26.2, S.D. 6.5 years) gave their written informed
consent prior to inclusion in the study. Subjects had no his-
tory of medical, neurological or psychiatric disorders. All
subjects were native German speakers. The study has been
approved by the local ethics committee, and has therefore
been performed in accordance with the ethical standards laid
down in the 1964 Declaration of Helsinki.
2.2. Selection of stimuli
A set of 90 simple subject verb object sentences was de-
rived from a training program for aphasic patients (LOGMA,
CH-8044 Zürich, Switzerland). Ninety stimuli were pseudo-
randomly divided into the three experimental conditions con-
sisting of 30 stimuli each. In order to avoid putative system-
atic effects of uncontrolled variables, such as word frequency,
imaginability, word length, etc. between the conditions, we
created three different experimental sets. Taking together
all sets, each sentence was presented once in each condi-
tion. Therefore, any putative systematic differences between
the different conditions should have been counter-balanced
across all sets.
In the active condition ‘sentence generation’ (SG), three
words were presented visually, adapted from simple subject
verb object sentences without additional components like
auxiliary verb, prepositions, etc. The task consisted in syntac-
tically completing the sentence (rearranging the word order,
declination of the verb) and vocalizing it aloud using two
definite articles (e.g. stimulus “throw ball child”, expected
response “The child throws the ball”). All subjects reported
in the training sessions (see below) that they could easily
understand the meaning of these stimuli despite their syntac-
tically incomplete form.
In the control condition ‘word reading’ (WR), equivalent
stimuli to condition SG were presented. Subject had to read
these words aloud (e.g. stimulus “throw ball child”, expected
response “throw ball child”) (Table 1).
One of our major concerns was that involuntary or auto-
matic syntactic production might occur in the WR condition,
i.e. subjects automatically produce a complete sentence men-
tally “The child throws the ball”, although it is not necessary
for the task. Therefore, the word order of the stimuli in WR
and SG conditions was pseudo-randomized such that words
were never presented at the position of the expected target ex-
pression, hoping to reduce this potential automatic syntactic
production. Additionally, there was a second control condi-
tion ‘sentence reading’ (SR), where complete sentences were
presented and the task was to read these sentences aloud (e.g.
stimulus “The child throws the ball”, expected response “The
child throws the ball”). Since the syntactic structure is already
complete in this condition, there is no need for syntactic pro-
duction. Further, this condition served as control condition
in addition to WR to account for the higher number of words
produced in SG as compared to WR.
2.3. Instructions
All subjects were familiarized with the task demands us-
ing a training program outside the scanner. All subjects easily
understood the design of the experiment. All subjects per-
formed several cycles of all conditions as practice using dif-
ferent stimuli than in the actual functional magnetic reso-
nance imaging (fMRI) experiment. Subjects were instructed
to speak in their usual way.
 S. Haller et al. / Neuropsychologia 43 (2005) 807–814
Table 1
Overview of the three experimental tasks used in the present study: name of condition, an example of a visually presented stimulus and an example of an
expected overt speech response
Condition
Sentence generation (SG)
Word reading (WR)
Sentence reading (SR)
In German, prepositions and verb endings (indicated in bold) are defined by one of three possible grammatical genders.
2.4. Experimental setup
The stimuli were projected using a video projector onto a
translucent screen mounted to the table of the scanner. Stimuli
were seen via mirrors of the head coil. We used a Sharp video
projector with XGA resolution that was placed in the control
room. Visual angle was approximately 30◦ . The scanner room
was darkened during the experiment. The subject’s head was
immobilized and noise protection was provided. Verbal out-
put was recorded using a portable MiniDisc recorder placed
in the scanner room and a microphone placed close to the
larynx of the subject. The aim of acceptable voice recording
with minimal image artifacts was easily achieved.
2.5. Experimental procedure
The experiment began with presentation of a visual fix-
ation cross for 10 s. Then, the experimental condition was
indicated by presenting capital red letters (e.g. SENTENCE
GENERATION) for 5 s. Thereafter, one stimulus of that con-
dition stimulus was shown for 4 s followed by a fixation cross
of 10.2 s, to scatter image acquisition onset relative to stim-
ulus onset (TR 2 s) and to compensate for BOLD signal de-
lay. Then, the next pseudo-randomly selected stimulus was
presented, etc. One run consisted of 10 stimuli each for con-
ditions SG, WR and SR, and lasted approximately 10 min,
including an additional fixation period of 10 s after the last
stimulus. Each subject performed 3 runs, equivalent to 30
repetitions of each condition. Conditions were presented in
pseudo-randomized order, such that taking together all ex-
perimental sets, each condition was presented equally often
in first, second, third, etc. position.
2.6. Image acquisition
Imaging was performed on a 1.5 T scanner (SIEMENS
Sonata, Siemens Electric, Erlangen, Germany) and func-
tional T2* weighted images were obtained with an echoplanar
single-shot pulse sequence (EPI) using an axial slice orienta-
tion. The matrix size was 64 × 64 (FOV 192 mm × 192 mm)
and 25 slices were acquired (4.5 mm slice thickness, 1 mm
gap) that covered the whole brain. The resulting resolution
was 3 mm × 3 mm × 5.5 mm voxels. Repetition time (TR)
809
Stimulus Expected response
Throw ball child The child throws the ball
Werfen Ball Kind Das Kind wirft den Ball
Throw ball child Throw ball child
Werfen Ball Kind Werfen Ball Kind
The child throws the ball The child throws the ball
Das Kind wirft den Ball Das Kind wirft den Ball
was 2 s, flip angle 90◦ and echo time (TE) 80 ms. The first two
volumes were discarded from further analysis to avoid non-
steady-state effects caused by T1 saturation. After functional
scanning, high-resolution data were acquired via 1 mm Iso-
Voxel T1MPRage (matrix 256 × 256, 176 slices) for cortex
normalization and cortex surface reconstruction.
2.7. Image analysis
The analysis of the fMRI data was performed using
BrainVoyager 2000 and BrainVoyager QX (http://www.
brainvoyager.com). Intra-session 1 mm T1MPRage scans
were used as anatomic reference for Talairach transforma-
tion of functional and anatomical images. Additionally, cor-
tex surface was reconstructed for surface-based statistical
evaluation. Slice scan time of functional EPI images was cor-
rected, motion artifacts were removed and smoothed using a
3 mm Gaussian kernel. EPI images were transformed into Ta-
lairach space (TAL). One predictor was created for each con-
dition. One additional categorical motion predictor was cre-
ated. Head motion occurs while the subjects speak, therefore
this motion predictor is “1” during speech output for two sec-
onds and “0” else. The BOLD signal has its maximum about
6 s after stimulus onset. Thus, it is possible to temporally dis-
sociate motion artifacts and fMRI response. The intention of
this motion predictor was to further improve signal quality
because it can explain signal changes that are caused by head
motion, as long as these signal changes are equivalent during
each speech output. All 45 functional measurement runs (15
subjects × 3 runs each) were combined into one general linear
model (GLM) using separate subject predictors. All activa-
tions reported were calculated using fixed effect models and
threshold of p < 0.05 corrected (dynamic statistical thresh-
olding using the false discovering rate (FDR) approach) and
extend threshold of 200 mm3 . The resulting clusters were
used as basis for volume of interest (VOI) GLM comparison
of the active condition SG and control conditions WR and
SR. Fixed effect GLMs were calculated to obtain a t-value
for the contrast SG minus WR and SG minus SR, for each
cluster.
Additionally, a region of interest (ROI) analysis was per-
formed. Two ROIs per hemisphere were defined as cubes
of 10 mm × 10 mm × 10 mm for BA 45 (center at Talairach
810 S. Haller et al. / Neuropsychologia 43 (2005) 807–814

±50, 20, 5) and BA 44 (center at Talairach ±57, 12, 24). The
maximum BOLD signal change (%) and standard deviation
were calculated for both hemispheres.
2.8. Behavioral data
The voice recordings were analyzed off-line. Assessed
were onset of verbal output relative to onset of stimulus
presentation, and correctness of the response. The voice
recordings were transferred to an Apple Macintosh computer
and digitized, then evaluated using the software Amadeus II
(http://www.hairersoft.com). The statistical analysis of the
behavioral data (i.e. onset of verbal response and correct-
ness) was evaluated using the statistical software Graphpad
Prism (http://www.graphpad.com). Onset times and accuracy
of verbal output in experimental conditions WR, SR and SG
were analyzed using a one-way ANOVA. Additionally, pair-
wise comparison was performed using Bonferroni’s Multiple
Comparison Test.
woman cooks the noodles’), which occurred in 3.3% of tri-
als in condition SG. There was a significant difference in
the group comparison (p < 0.05 one-way ANOVA). Pair-wise
analysis revealed significant difference between SR and SG
(p < 0.05 Bonferroni’s Multiple Comparison Test). No signif-
icant difference was present between SR and WR, WR and
SG.
Mean onset of verbal output relative to the onset of vi-
sual stimulus presentation was 1.51 ± 0.28 s, 1.43 ± 0.24 s
and 1.98 ± 0.42 s for conditions WR, SR and SG, respec-
tively. Comparison of all conditions (WR, SR and SG) re-
vealed a significant difference of means (p < 0.001, one-
way ANOVA). Pair-wise comparison of conditions showed
significant differences between WR and SG (p < 0.01,
Bonferroni’s Multiple Comparison Test) and SR and SG
(p < 0.001, Bonferroni’s Multiple Comparison Test), while
no significant difference was present between WR and
SR.
3.2. Functional MR data

3. Results 3.2.1. Movement

3.1. Behavioral data
The correct response was given in 99.2, 100 and 95.8%
of trials for conditions WR, SR and SG, respectively. The
most common incorrect response was neglect of the sec-
ond article (e.g. ‘The woman cooks noodles’ instead of ‘The
The absolute maxima of head motion were calculated in
all 45 runs (15 subjects × 3 runs, duration 11 min each). The
maximum translation in x y z dimensions were 0.32 ± 0.25
(mean ± S.D. [mm]), 0.94 ± 0.37 and 0.80 ± 0.62, respec-
tively. Analogous, the maximum rotation in the x y z di-
mension was 0.66 ± 0.69 (mean ± S.D. [◦ ]), 0.30 ± 0.19 and
0.27 ± 0.16, respectively.

Fig. 1. This figure displays significant activations for the comparison of the active condition SG minus control condition WR. Calculations are based on a
cortex-based fixed effects GLM with threshold of p < 0.05 corrected (FDR) and an extent threshold of 200 mm3 . Significant activations are color-coded and
superimposed on axial slices of an NMI standard brain transformed into Talairach space. The strongest activation (largest volume of activation cluster and
strongest t-value) was present in left inferior frontal gyrus, BA 44/45. The left side on the image is the right side of the brain.
 S. Haller et al. / Neuropsychologia 43 (2005) 807–814 811

Table 2
Significant activations for the comparison of the active condition SG minus control condition WR based on a cortex-based fixed effects GLM with a threshold
of p < 0.05 corrected (FDR) and an extent threshold of 200 mm3
Region Size (mm3 ) Side BA t-value TAL, x TAL, y TAL, z
Inferior frontal gyrus 7497 L 44/45 7.81 −43 17 11
Inferior parietal lobule 822 L 40 4.86 −29 −55 38
Superior parietal lobule 506 L 7 4.76 −21 −66 42
Precentral gyrus 501 L 6/44 5.76 −62 0 21
Superior frontal gyrus 246 L 6 3.79 −6 0 65
Inferior frontal gyrus 200 L 9 5.73 −57 11 30
Medial frontal gyrus 1106 R 6 5.44 1 13 43
Insula 571 R 13 4.94 36 12 −1
Inferior frontal gyrus 257 R 45 4.12 36 27 16
Anatomic region in Talairach space (reference), side, BA: Brodmann’s area, size of activation cluster in mm3 , t-value for the comparison SG minus WR based
on VOI GLM and TAL coordinates.

3.2.2. Comparison of ‘word reading’ and ‘sentence
reading’
The comparison of the two control conditions ‘word read-
ing’ and ‘sentence reading’ did not show any significant dif-
ferences.
3.2.3. Comparison of ‘sentence generation’ and ‘word
reading’
The comparison between the active condition SG and the
control condition WR was associated with significant activa-
tions dominantly in the left hemisphere. The maximum acti-
vation (largest size of activation cluster and highest t-value)
was found in left inferior frontal gyrus BA 44/45. Further
signal changes were present in the left medial frontal (BA 6)
gyrus, the superior parietal lobule (BA 7) and the right insula
(BA 13) (see Fig. 1 and Table 2).
3.2.4. Comparison of ‘sentence generation’ and
‘sentence reading’
The comparison between the active condition SG and the
second control condition SR was again associated with signif-
icant activations dominantly in the left hemisphere. Equiva-
lent to the comparison of SG and WR, the strongest activation
(size of activation cluster and t-value) was present in left in-
ferior frontal gyrus BA 44/45 Further signal changes were
present in the left medial frontal (BA 6) gyrus, the superior
parietal lobule (BA 7) and the right insula (BA 13) (see Fig. 2
and Table 3).
3.2.5. ROI analysis
Additionally, the maximum BOLD signal change was as-
sessed for BA 44 and BA 45 of both hemispheres for all ex-
perimental conditions, and illustrated in Fig. 3. The strongest

Fig. 2. This figure displays significant activations for the comparison of the active condition SG minus the second control condition SR. Calculation and
illustration equivalent to Fig. 1. The strongest activation (largest volume of activation cluster and strongest t-value) was again found in left inferior frontal gyrus,
BA 44/45.
812 S. Haller et al. / Neuropsychologia 43 (2005) 807–814

Table 3
Significant activations for the comparison of the active condition SG minus the second control condition SR are listed based on a cortex-based fixed effects
GLM with a threshold of p < 0.05 corrected (FDR) and an extent threshold of 200 mm3
Region Size (mm3 ) Side BA t-value TAL, x TAL, y TAL, z
Inferior frontal gyrus 16762 L 44/45 9.37 −42 15 14
Medial frontal gyrus 4923 L 6 7.21 −1 10 45
Middle temporal gyrus 359 L 21 4.39 −53 −45 5
Superior parietal lobule 310 L 7 4.35 −20 −66 44
Insula 1099 R 13 6.07 40 10 8
Postcentral gyrus 531 R 3 5.58 25 −31 69
Postcentral gyrus 286 R 1 4.20 45 −31 58
Anatomic region in TAL space, BA: Brodmann’s area, size of activation cluster in mm3 , t-value for the comparison SG minus WR as determined using VOI
GLM, and TAL coordinates.

BOLD signal change over all was found for condition SG
in left BA 45. Conditions WR and SR evoked lesser BOLD
signal changes compared to condition SG in BA 44 and 45 bi-
laterally. The size of maximum BOLD signal in condition SG
was significantly stronger in BA 45 compared to BA 44 bi-
laterally, with a left dominance (p < 0.01 left, p < 0.05 right).
The amplitude of maximum BOLD change was larger in the
left hemisphere.
4. Discussion
In the present study, subjects had to produce simple SVO
sentences. The demand on syntactic production load was var-
ied between three different conditions. In line with our predic-
tions, activation was mainly observed in left inferior frontal
gyrus/Broca’s area for the higher syntactic demand condition.
In particular, a region of interest analysis revealed, that
within the IFG, the maximum BOLD signal change of the
active condition was significantly stronger in BA 45 com-
pared to BA 44. Further, significant however less pronounced
activation was also present in right IFG, again with a signif-
icant stronger BOLD change in BA 45 compared to BA 44.
This supports the crucial involvement of BA 45 in syntactic
encoding.
The present activations of the left IFG for syntactic pro-
cessing are in accordance with brain lesion and functional
imaging data (see e.g. the review (Kaan & Swaab, 2002)).
In line with the present results, dominant activation in left
IFG was also found in a PET study on syntax in sentence
generation (Indefrey, Brown, et al., 2001). In this restrictive
scene description study, subjects had to produce sentences
like “The red square launches the blue ellipse” according to
visually presented objects. This sentence production condi-
tion was compared to single word utterance condition (i.e.
“square red, ellipse blue, launch”). The maximum difference
between these two conditions was found in left inferior frontal
gyrus BA 44 (16.3% of activated volume) and caudally ad-
jacent to BA 44 (83.7% of activated volume) most probably
in Rolandic part of BA 6. Interestingly, the reported activa-
tion largely overlaps with the presented results despite the
differences in tasks. While subjects in the present study were
asked to complete and produce “every day” sentences accord-
ing to visually presented words, in the Indefrey et al. study
subjects had to produce sentences describing three colored
objects (circle, square ellipse), which could perform one of
two actions (go next to, launch). Thus, the semantic com-
ponents in the two studies are different. Further, the control
conditions also differ. Due to the large overlap in activation
(in left IFG) between the Indefrey et al. and our study, we

Fig. 3. The maximum BOLD signal change (%) for conditions WR, SR and SG in BA 44 (grey columns) and 45 (black columns), defined by cubic ROIs of
10 mm with the center at TAL ±50, 20, 5 and ±57, 12, 24, respectively.
 S. Haller et al. / Neuropsychologia 43 (2005) 807–814
conclude that activation in left IFG can best be attributed to
syntactic encoding.
Besides syntax processing, the left IFG has also been
linked to verbal working memory. It could thus be argued that
our active condition SG might differ from the control condi-
tions not only in terms of syntactic operations (constructing
the canonical word order as well as tempus, gender, etc.) but
also in terms of memory load (keeping non-integrated ma-
terial active while processing other words) (Martin, 2003).
Concerning working memory (WM), a distinction can be
made between syntax-specific WM involved in the genera-
tion of a syntactic structure (e.g. maintenance of the numerous
of the subject until the selection of the correct verb-form in
the sentence) (Kolk, 1995), and task-independent WM (e.g.
maintenance and processing of the stimulus) (Indefrey, Ha-
goort, Herzog, Seitz, & Brown, 2001). Involvement of the
syntax-specific WM is obviously inevitable, but is regarded
as part of syntax production. Task-independent WM might
theoretically also differ between our production and control
conditions. It is, however, most challenging if not impossi-
ble to create control conditions (of different cognitive de-
mands) without having a theoretical confound manifesting
as alteration in the WM load. Despite a wide agreement in
the crucial role of dorsolateral prefrontal cortex (DLPFC)
in working memory, there are substantial differences in dif-
ferent functional imaging studies investigating this process
regarding the exact location of activation, probably because
the particular location is task dependent. Although the role
of working memory in comprehension has been studied ex-
tensively, relatively little attention has been paid to its role
in production (Martin, 2003). Chein, Fissell, Jacobs, and
Fiez (2002) performed a meta-analysis of 30 fMRI studies
on verbal working memory and identified two sub-regions
in left inferior frontal cortex. In particular, a dorsally lo-
cated focus was found in verbal working memory studies in
which the level of difficulty/performance was manipulated
(TAL −44, 7, 26) while a ventrally located focus was found
in studies that contrasted word and pseudo-word process-
ing (TAL −41, 13, 7). The peak activation in our study is
close to the dorsal focus in the Chein et al. meta-analysis,
which suggests that it is related to the non-semantic, i.e. syn-
tactic aspects of working memory during sentence produc-
tion.
One of our major concerns was that the incomplete syn-
tactic structure of the stimuli in conditions SG and WR
might evoke an involuntary or automatic syntactic comple-
tion in condition WR, although it is not necessary for the
completion of the task. Consequently, expected differences
in the neuronal activation between SG and WR might be
blurred. Therefore, in a second control condition SR, com-
plete target sentences were presented. Since the presented
sentences are grammatically complete, there is no need for
syntax production in this condition. If automatic syntactic
production occurred in WR, one would expect differences
in signal changes between WR and SR, which was how-
ever not observed. Additionally, there was no significant
813
difference on the behavioral level in terms of response la-
tency between conditions WR and SR, while there were dif-
ferences between WR and SG as well as between SR and
SG.
Areas of activation for the sentence generation condi-
tion other than the IFG included the left medial frontal (BA
6) gyrus, the superior parietal lobule (BA 7) and the right
insula (BA 13). These areas have previously been impli-
cated in sentence (Kircher, Brammer, Tous Andreu, Williams,
& McGuire, 2001) and syntax processing (Gernsbacher
& Kaschak, 2003; Kaan & Swaab, 2002). The involvement
of these areas is in line with more recent models on syntax
encoding, implying not only the left IFG but a more dis-
tributed cortical network (Bookheimer, 2002; Dick et al.,
2001; Grodzinsky, 2000; Kaan & Swaab, 2002; Munhall,
2001).
A point of controversial discussion in fMRI studies on lan-
guage production is the use of overt or covert (silent) speech
(Huang, Carr, & Cao, 2002). We used overt language produc-
tion in order to monitor subjects performing the task. While
this is probably not always necessary in healthy subjects, it is
certainly appropriate in patients (Rosen, Ojemann, Ollinger,
& Petersen, 2000) who we shall investigate. Also, it has been
shown that activations during overt speech differ in some
aspects from covert speech (Barch et al., 1999; McCarthy,
Blamire, Rothman, Gruetter, & Shulman, 1993; Palmer et
al., 2001; Price, Moore, & Frackowiak, 1996; Riecker, Ack-
ermann, Wildgruber, Dogil, & Grodd, 2000; Rosen et al.,
2000; Zelkowicz, Herbster, Nebes, Mintun, & Becker, 1998).
The major disadvantages of overt speech are head movement
and susceptibility artifacts (Birn, Bandettini, Cox, Jesman-
owicz, & Shaker, 1998). We tried to minimize these prob-
lems by using only single sentences requiring approximately
2 s of verbal output. While head movement and resulting
artifacts are present while speaking, the BOLD effect has
its peak only after a delay of approximately 6 s. Thus, it
is possible to temporally dissociate motion artifacts and the
BOLD signal. Additionally, we measured head movement
and this was minimal in all conditions. Visual inspection of
the raw data revealed only negligible susceptibility artifacts.
Additionally, event-related paradigms are less sensitive to
noise and signal drifts (Kang, Constable, Gore, & Avrutin,
1999).
5. Conclusions
In conclusion, we could demonstrate that language pro-
duction on a sentence level can be investigated in fMRI using
overt speech. In particular, activation associated with syntac-
tic encoding during overt sentence production was predom-
inant in the left inferior frontal gyrus and other left hemi-
spheric areas. We emphasize the need to further investigate
speech production directly, despite difficulties of control-
ling conceptual processing underlying the generation of sen-
tences.
814 S. Haller et al. / Neuropsychologia 43 (2005) 807–814
Acknowledgement
We thank Peter Indefrey for helpful comments on earlier
versions of the manuscript.
References
Barch, D. M., Sabb, F. W., Carter, C. S., Braver, T. S., Noll, D. C.,
& Cohen J.D. (1999). Overt verbal responding during fMRI scan-
ning: Empirical investigations of problems and potential solutions.
NeuroImage, 10, 642–657.
Birn, R. M., Bandettini, P. A., Cox, R. W., Jesmanowicz, A., & Shaker, R.
(1998). Magnetic field changes in the human brain due to swallowing
or speaking. Magnetic Resonance in Medicine, 40, 55–60.
Bookheimer, S. (2002). Functional MRI of language: New approaches to
understanding the cortical organization of semantic processing. Annual
Review of Neuroscience, 25, 151–188.
Cabeza, R., & Nyberg, L. (2000). Neural bases of learning and memory:
Functional neuroimaging evidence. Current Opinion in Neurology, 13,
415–421.
Chein, J. M., Fissell, K., Jacobs, S., & Fiez, J. A. (2002). Functional het-
erogeneity within Broca’s area during verbal working memory. Phys-
iology and Behaviour, 77, 635–639.
Dapretto, M., & Bookheimer, S. Y. (1999). Form and content: Dissoci-
ating syntax and semantics in sentence comprehension. Neuron, 24,
427–432.
Dick, F., Bates, E., Wulfeck, B., Utman, J. A., Dronkers, N., & Gerns-
bacher, M. A. (2001). Language deficits, localization, and grammar:
Evidence for a distributive model of language breakdown in aphasic
patients and neurologically intact individuals. Psychological Review,
108, 759–788.
Embick, D., Marantz, A., Miyashita, Y., O’Neil, W., & Sakai, K. L.
(2000). A syntactic specialization for Broca’s area. Proceedings of
the National Academy of Sciences of the United States of America,
97, 6150–6154.
Ferstl, E. C., Guthke, T., & von Cramon, D. Y. (1999). Change of per-
spective in discourse comprehension: Encoding and retrieval processes
after brain injury. Brain Language, 70, 385–420.
Gernsbacher, M. A., & Kaschak, M. P. (2003). Neuroimaging studies of
language production and comprehension. Annual Review of Psychol-
ogy, 54, 91–114.
Grodzinsky, Y. (2000). The neurology of syntax: Language use without
Broca’s area. The Behavioural and Brain Sciences, 23, 1–21, discus-
sion 21–71.
Huang, J., Carr, T. H., & Cao, Y. (2002). Comparing cortical activations
for silent and overt speech using event-related fMRI. Human Brain
Mapping, 15, 39–53.
Indefrey, P., & Levelt, W. J. M. (2000). In M. Gazzaniga (Ed.), The new
cognitive neurosciences (2nd ed.). Cambridge, MA: MIT Press.
Indefrey, P., Hagoort, P., Herzog, H., Seitz, R. J., & Brown, C. M. (2001).
Syntactic processing in left prefrontal cortex is independent of lexical
meaning. NeuroImage, 14, 546–555.
Indefrey, P., Brown, C. M., Hellwig, F., Amunts, K., Herzog, H., Seitz,
R. J., et al. (2001). A neural correlate of syntactic encoding during
speech production. Proceedings of the National Academy of Sciences
of the United States of America, 98, 5933–5936.
Kaan, E., & Swaab, T. Y. (2002). The brain circuitry of syntactic com-
prehension. Trends Cognitive Sciences, 6, 350–356.
Kang, A. M., Constable, R. T., Gore, J. C., & Avrutin, S. (1999). An
event-related fMRI study of implicit phrase-level syntactic and se-
mantic processing. NeuroImage, 10, 555–561.
Kircher, T. T., Brammer, M. J., Williams, S. C., & McGuire, P. K. (2000).
Lexical retrieval during fluent speech production: An fMRI study.
NeuroReport, 11, 4093–4096.
Kircher, T. T., Brammer, M., Tous Andreu, N., Williams, S. C., &
McGuire, P. K. (2001). Engagement of right temporal cortex during
processing of linguistic context. Neuropsychologia, 39, 798–809.
Kircher, T. T., Brammer, M. J., Levelt, W., Bartels, M., Williams, S. C.
R., & McGuire, P. K. (2004). Pausing for thought: Engagement of left
temporal cortex during pauses in speech. NeuroImage, 21(1), 84–90.
Kolk, H. (1995). A time-based approach to agrammatic production. Brain
Language, 50, 282–303.
Levelt, W. J. M. (1989). Speaking: from intention to articulation. Cam-
bridge, MA: The MIT Press.
Levelt, W. J. M. (1995). The ability to speak: From intentions to spoken
words. European Review, 3, 13–23.
Linebarger, M. C., Schwartz, M. F., & Saffran, E. M. (1983). Sensitivity
to grammatical structure in so-called agrammatic aphasics. Cognition,
13, 361–392.
Martin, R. C. (2003). Language processing: functional organization and
neuroanatomical basis. Annual Review of Psychology, 54, 55–89.
McCarthy, G., Blamire, A. M., Rothman, D. L., Gruetter, R., & Shul-
man, R. G. (1993). Echo-planar magnetic resonance imaging studies
of frontal cortex activation during word generation in humans. Pro-
ceedings of the National Academy of Sciences of the United States of
America, 90, 4952–4956.
Munhall, K. G. (2001). Functional imaging during speech production.
Acta Psychologica (Amsterdam), 107, 95–117.
Oldfield, R. C. (1971). The assessment and analysis of handedness: The
Edinburgh inventory. Neuropsychologia, 9, 97–113.
Palmer, E. D., Rosen, H. J., Ojemann, J. G., Buckner, R. L., Kel-
ley, W. M., & Petersen, S. E. (2001). An event-related fMRI study
of overt and covert word stem completion. NeuroImage, 14, 182–
193.
Price, C. J., Moore, C. J., & Frackowiak, R. S. (1996). The effect of
varying stimulus rate and duration on brain activity during reading.
NeuroImage, 3, 40–52.
Riecker, A., Ackermann, H., Wildgruber, D., Dogil, G., & Grodd, W.
(2000). Opposite hemispheric lateralization effects during speaking
and singing at motor cortex, insula and cerebellum. NeuroReport, 11,
1997–2000.
Rosen, H. J., Ojemann, J. G., Ollinger, J. M., & Petersen, S. E. (2000).
Comparison of brain activation during word retrieval done silently
and aloud using fMRI. Brain and Cognition, 42, 201–217.
Sakai, K. L., Hashimoto, R., & Homae, F. (2001). Sentence processing
in the cerebral cortex. Neuroscience Research, 39, 1–10.
Zelkowicz, B. J., Herbster, A. N., Nebes, R. D., Mintun, M. A., & Becker,
J. T. (1998). An examination of regional cerebral blood flow during
object naming tasks. Journal of the International Neuropsychological
Society, 4, 160–166.