Brain & Language 121 (2012) 47–57

Contents lists available at SciVerse ScienceDirect

Brain & Language

journal homepage: www.elsevier .com/locate /b&l

Regular Article

The contribution of the inferior parietal cortex to spoken language production

Fatemeh Geranmayeh ⇑, Sonia L.E. Brownsett, Robert Leech, Christian F. Beckmann, Zoe Woodhead,Richard J.S. WiseDivision of Experimental Medicine, Imperial College London, Hammersmith Hospital Campus, Du Cane Road, London W12 0NN, UKMRC Clinical Sciences Centre, Imperial College London, Hammersmith Hospital Campus, Du Cane Road, London W12 0NN, UK

a r t i c l e i n f o

Article history:Accepted 3 February 2012Available online 29 February 2012

Keywords:fMRISpeech productionParietal lobeIndependent component analysis

0093-934X/$ - see front matter � 2012 Elsevier Inc. Adoi:10.1016/j.bandl.2012.02.005

⇑ Corresponding author at: Division of ExperimentLondon, Hammersmith Hospital Campus, Du Cane Roa+44 (0)20 7594 8921.

E-mail address: fatemeh.geranmayeh00@imperial.

a b s t r a c t

This functional MRI study investigated the involvement of the left inferior parietal cortex (IPC) in spokenlanguage production (Speech). Its role has been apparent in some studies but not others, and is not con-vincingly supported by clinical studies as they rarely include cases with lesions confined to the parietallobe. We compared Speech with non-communicative repetitive tongue movements (Tongue). The datawere analyzed with both univariate contrasts between conditions and probabilistic independent compo-nent analysis (ICA). The former indicated decreased activity of left IPC during Speech relative to Tongue.However, the ICA revealed a Speech component in which there was correlated activity between left IPC,frontal and temporal cortices known to be involved in language. Therefore, although net synaptic activitythroughout the left IPC may not increase above baseline conditions during Speech, one or more local sys-tems within this region are involved, evidenced by the correlated activity with other language regions.

� 2012 Elsevier Inc. All rights reserved.

1. Introduction

Indefrey and Levelt (2004) were the first to publish a meta-anal-ysis of individual functional imaging experiments (over 80) onspeech production. These studies had employed ‘subtractive’ con-trasts between behavioral conditions using univariate statisticalmethods. Activated regions within the left frontal and temporalcortices, but not the inferior parietal cortex (IPC), were reportedto be related to processing stages from lexical retrieval throughto articulation (Levelt, 2001). One possibility for the apparent ‘null’result in IPC was because the studies included in the meta-analysishad employed only single word tasks. This can be contrasted with alarge lesion-deficit study, relating impaired aspects of speech pro-duction to lesion location identified on anatomical images (Borov-sky, Saygin, Bates, & Dronkers, 2007). In this study, semanticinformation conveyed by utterances was reduced in those patientsin whom infarction included, among other regions, the left inferiorparietal lobe (specifically, the angular gyrus). This study had ana-lyzed discourse level speech production. However, of the fifty pa-tients studied, only one had infarction largely confined to the leftparietal lobe. In all other cases with parietal involvement therewas also infarction of the posterior temporal and/or frontal lobe;and presumably fairly extensive involvement of long white mattertracts, disconnecting remote left hemisphere cortical and subcorti-

ll rights reserved.

al Medicine, Imperial Colleged, London W12 0NN, UK. Fax:

ac.uk (F. Geranmayeh).

cal regions (Catani & ffytche, 2005). Large lesions involving bothgray and white matter are not ideal (but, in human pathology, allthat are commonly available for research) when attempting to as-cribe cognitive functions to particular cortical regions. In anotherlesion-deficit study, Fridriksson and colleagues (2010), suggestedthat hypo perfusion of the inferior supramarginal gyrus is associ-ated with speech repetition deficits. Again in this study, lesionswere not confined to the IPC.

Relatively small numbers of functional imaging studies haveinvestigated discourse level spoken language production ratherthan single word or single sentence level production (Awad, War-ren, Scott, Turkheimer, & Wise, 2007; Blank, Scott, Murphy, War-burton, & Wise, 2002; Braun, Guillemin, Hosey, & Varga, 2001;Brownsett & Wise, 2010; Haller, Radue, Erb, Grodd, & Kircher,2005; Tremblay & Small, 2011). The majority of these studies usedpositron emission tomography (PET), in which movement-relatedand other artifacts are less evident than with conventional func-tional magnetic resonance imaging (fMRI), based on measuringthe blood oxygen-level dependent (BOLD) signal (Gracco, Tremb-lay, & Pike, 2005). The PET studies related spoken language produc-tion networks with those involved in spoken languagecomprehension (Awad et al., 2007), and communication throughsign language (Braun et al., 2001) and writing (Brownsett & Wise,2010). Although none of the PET studies were designed to relatefoci of activation to particular processing stages, they all reporteda role for IPC, in particular the left angular gyrus, in the productionof language, regardless of the output modality.

The motivation for these PET studies was to attempt to observethe entire networks engaged by everyday use of language. This is in

48 F. Geranmayeh et al. / Brain & Language 121 (2012) 47–57

contrast to the alternative approaches, which have attempted toattribute one or a few cortical regions to separate language pro-cesses such as phonological, semantic and syntactic levels of pro-cessing. Notwithstanding, the outcome of many of these latterstudies has been to show a considerable degree of overlap acrossthe left cerebral neocortex (Vigneau et al., 2006). A particularexample is the left posterior middle temporal gyrus and the infe-rior frontal gyrus, important components of a semantic networkaccording to Binder and colleagues (2009) but also central to theprocessing of syntax according to others (Papoutsi, Stamatakis,Griffiths, Marslen-Wilson, & Tyler, 2011); one conclusion from thisis that attempting a ‘neophrenology’ of language can only be par-tially successful, and such a complex cognitive process can equallybe represented as a distributed network without attempting de-tailed regional associations between function and anatomy.

The difficulties of separating signal from noise have discouragedthe study of spoken language production with fMRI; although fMRIbased on arterial spin labeling may offer an alternative to rival PETimaging of conversational speech production (Kemeny, Ye, Birn, &Braun, 2005). A number have used brief utterances of syllables andsingle words to investigate motor-sensory aspects of speech pro-duction (for example, Bohland & Guenther, 2006; Guenther, Ghosh,& Tourville, 2005; Riecker et al., 2005). In one MRI study investigat-ing single sentence production (e.g. ‘The child throws the ball’), taskswere confined to epochs of 2 s (Haller et al., 2005). In that study,when compared to sentence reading, sentence production wasassociated with activations of left inferior and medial frontal gyri,right insula and the left superior parietal lobule. In another fMRIstudy that investigated propositional speech production, speech-related activity in the left angular gyrus is apparent in one of thefigures, although this was small in spatial extent relative to exten-sive left inferolateral temporal and dorsolateral frontal activity(Dhanjal, Handunnetthi, Patel, & Wise, 2008). This latter study re-lied on image acquisition with sparse-sampling (Hall et al., 1999),in which functional imaging data were acquired during brief peri-ods when the subject’s spoken output was interrupted to minimizehead movements and other sources of fMRI artifact related tospeech production. The technique of sparse-sampling has been re-cently used to clarify the contribution of the pre-supplementarymotor, ventral premotor, cingulate, and pars triangularis to motorresponse selection in speech production (Tremblay & Small, 2011).

A further issue in language studies relates to the nature of thebaseline task. This can be of particular importance when assessingthe role of IPC when data are analyzed with univariate statisticalmethods. Parts of the IPC form a component of the so-called defaultmode network. Activity in this network increases at ‘rest’ or whena task is not cognitively ‘effortful’, and it has proved to have a veryreproducible distribution (Buckner, Andrews-Hanna, & Schacter,2008; Esposito et al., 2006; Gusnard & Raichle, 2001; Raichleet al., 2001). It has been argued (Binder et al., 2009; McKiernan,D’Angelo, Kaufman, & Binder, 2006) that the default mode networkis associated with the internal generation of spontaneous stimulus-independent thoughts, which will engage the semantic processesalso involved in language processing. Therefore, ‘subtractive’ con-trasts may mask important language-related regions, dependingon the chosen baseline condition. This was the subject of the studyby Seghier and colleagues (2010), in which careful use of contrastsbetween a number of conditions separated single word semanticprocesses from activity related to the default mode network inthe left angular gyrus. It was also the motivation for including acognitively ‘effortful’ numerical task as one of the baseline condi-tions in the language studies of Awad and colleagues (2007) andBrownsett and Wise (2010).

The objective of the present study was to investigate the partic-ipation of the IPC in spoken language production. The study em-ployed a similar design and image acquisition methodology to

those used by Dhanjal and colleagues (2008). However, in additionto univariate contrasts, group temporal concatenation independentcomponent analysis (ICA) was used to investigate correlated activ-ity over time across distributed brain regions (Beckmann & Smith,2004). This approach has two advantages. It is able to separate sig-nals within distributed functional brain networks from noise. This‘noise’ may be created by head movement and physiological fluctu-ations, such as varying arterial carbon dioxide levels that may oc-cur during the breathing control associated with speaking. Mostimportantly, ICA can reveal distinct but anatomically overlappingnetworks, corresponding to functionally different activation timecourses intermixed within a single voxel (for example, see Leech,Braga. R., & D., 2012; Leech, Kamourieh, Beckmann, & Sharp,2011). This distinction will not be apparent in a standard subtrac-tive analysis. A further motivation for the present study arose fromthe publication by Smith and colleagues (2009). These authorsdemonstrated that correlated activity within separable distributedfunctional networks was present in ‘resting state’ fMRI data ac-quired from healthy subjects (Smith et al., 2009). When relatingthe anatomical distribution of these networks to a central databaseof regional ‘activations’ obtained from many functional imagingstudies, a largely left-lateralized network distributed between infe-rior parietal, inferolateral temporal and posterior frontal corticescorresponded most closely with a cohort of functional imagingstudies that had investigated some aspects of language processing.Our hypothesis was that this left-lateralized network (includingthe IPC) would be revealed by ICA to be engaged in the spoken lan-guage production condition (Speech) but not when the subjectsperformed a baseline task of non-communicative repetitive move-ments of the tongue (Tongue). In particular, we predicted thatanalysis with ICA would prove more sensitive than a conventionalsubtraction contrast between conditions, using univariate statis-tics, in demonstrating the contribution of IPC to the productionof discourse level spoken language production.

2. Materials and methods

2.1. Participants and fMRI procedure

Twenty-three right-handed, native English speakers partici-pated after giving informed written consent. The data from twoparticipants were excluded because of excessive movement(>4 mm in any plane), one performed very poorly on the speechproduction task and a fourth showed unexpected brain pathology.Therefore, 19 subjects were included in the final analysis (13 fe-males; mean age, 30 years; range, 22–62 years). Approval for thestudy was provided by the ethics committee for the North WestThames region.

MRI data were obtained on a Philips Intera 3.0 Tesla scannerusing dual gradients, a phased array head coil, and sensitivityencoding (SENSE) with an undersampling factor of 2. A ‘‘sparse’’fMRI design (Hall et al., 1999) was used to minimize movement-and respiratory-related artifact associated with speech (Graccoet al., 2005; Mehta, Grabowski, Razavi, Eaton, & Bolinger, 2006).Tasks were performed in response to specific visual stimuli duringan epoch of 7 s (s). Following this, a fixation cross was displayedwhich was the cue for the subjects to discontinue the task. 0.75 slater whole brain functional imaging data was acquired over 2 s.The cycle was then repeated. This technique, to acquire data aftertask-related head movement and immediate speech-related respi-ration has temporarily ceased, is made possible due to the delayedonset of the hemodynamic response function (HRF) and its persis-tence for >10 s.

Functional MR images were obtained using a T2�-weighted, gra-dient-echo, echoplanar imaging (EPI) sequence with whole-brain

F. Geranmayeh et al. / Brain & Language 121 (2012) 47–57 49

coverage (repetition time, 10 s; acquisition time, 2 s; echo time,30 ms; flip angle, 90�). Thirty-two axial slices with a slice thicknessof 3.25 mm and an interslice gap of 0.75 mm were acquired inascending order (resolution: 2.19, 2.19, 4.0 mm; field of view:280, 224, 128 mm). Quadratic shim gradients were used to correctfor magnetic field inhomogeneities within the brain. In addition, ahigh resolution 1 mm3 T1-weighted whole-brain structural imagewas obtained for each subject.

2.2. fMRI paradigm design

Stimuli were designed using E-Prime software (Psychology Soft-ware Tools) and presented through an IFIS-SA system (In Vivo Cor-poration). The experimental paradigm consisted of threeconditions that were randomly presented: spoken language produc-tion (Speech), repetitive tongue movements (Tongue), and a silentrest condition (Rest). During Speech trials, subjects were requiredto define nouns, selected from the Medical Research Council psycho-linguistic database (Wilson, 1988). They were displayed on the cen-ter of a screen and presented on a computer screen to the subjectinside the bore of the scanner. The concrete nouns were all monosyl-labic and were selected, based on the following measures derivedfrom the Medical Research Council psycholinguistic database:familiarity (mean 554, standard deviation – s.d. 37, range, 498–645); imageability (mean 588, s.d. 24, range 535–647); concreteness(mean 591, s.d. 28, range: 492–646); age of acquisition (mean 258,s.d. 43, range 161–344); and Kucera–Francis frequency (mean 40,s.d. 49, range 1–220). The word was preceded by a fixation crosshairfor 0.25 s, followed by the presentation of the word in lower casefont in the center of the screen. The word was replaced after 7 s bythe crosshair for 0.75 s, which was the signal for the subject to ceasespeaking prior to image data acquisition over 2 s. A different wordwas used for each trial. Subjects received training immediately priorto the experiment to familiarize them with the task. They were askedto speak for the entire 7 s to generate as much verbal informationpertaining to the given noun as possible. An example of a subject’sresponse to the word ‘cake’ was ‘‘cake is a sweet item, it’s usually madeof flour, eggs, margarine, and butter’’. The cycle was the same for ton-gue movements, except that the subjects saw the instruction ‘movetongue’ instead of a noun for 7 s, during which time they were re-quired to repeatedly place the tip of the tongue against the upperalveolar ridge at a rate of about 1/s. During the rest trials the subjectssaw the crosshair for the entire 8 s before data acquisition. The de-sign was necessarily event-related due to the use of sparse-sampling.

A second run of the three conditions was performed after eachsubject had topical local anesthetic solution (1% lidocaine) appliedto their tongue. This was to fulfill a supplementary aim of thestudy, which was to investigate the motor-sensory neural conse-quences of altering somatosensory feedback during speech produc-tion. This procedure did not alter the rate of speech production,and after analysis (using the various methods outlined below,including ICA) it was apparent that the mild alteration in somato-sensory feedback as the consequence of topical application of lido-caine had been insufficient to produce observable altered activityin motor-sensory systems controlling speech production. There-fore, the first and second runs were amalgamated in the final anal-yses to demonstrate the role of IPC during spoken languageproduction. After combining the two runs, each of the three condi-tions (Tongue, Rest, Speech) was presented a total of 80 times, di-vided equally over the two runs.

2.3. Behavioral analysis

The speech output was recorded using an MR-compatiblemicrophone attached to ear-defending headphones (MR-Confon).

The recordings of Speech trials were transcribed verbatim andsuprasegmental aspects were not included in the transcription.The transcription was then analyzed to calculate both the numberof content words (nouns, verbs and adjectives, with function wordsexcluded) and the total number of syllables (function words in-cluded) produced per trial. Inevitably, in some trials there wereoccasional morphological, syntactic and/or semantic errors in theutterance, with intermittent repetition of a word. These errors ofproduction were included in the breakdown of content wordsand syllables, but they never amounted to more than 5% of the to-tal count for either content words or syllables produced. Fillers(‘um’, ‘er’, etc.) were not included in the count of produced sylla-bles. A Pearson correlation analysis was used to identify the rela-tionship of content words to syllables in both runs, and a pairedt-test was used to check for differences in Speech production be-tween the two runs.

2.4. Univariate whole-brain analysis

Univariate analysis was carried out within the framework of thegeneral linear model using FEAT (FMRI Expert Analysis Tool) Ver-sion 5.98, part of FSL (FMRIB’s Software Library, http://www.fmri-b.ox.ac.uk/fsl). The following image pre-processing steps wereapplied: realignment of EPI images for motion correction usingMCFLIRT (Jenkinson, Bannister, Brady, & Smith, 2002); non-brainremoval using BET (Brain Extraction Tool) (Smith, 2002); spatialsmoothing using a 6 mm full-width half-maximum Gaussian ker-nel; grand-mean intensity normalization of the entire four dimen-sional dataset by a single multiplicative factor; and high passtemporal filtering (Gaussian-weighted least-squares straight linefitting, with sigma = 50 s) to correct for baseline drifts in the signal.Time-series statistical analysis was carried out using FILM (FMRIB’sImproved Linear Modeling) with local autocorrelation correction.Registration to high resolution structural and Montreal Neurologi-cal Institute (MNI) standard space images (MNI 152) were carriedout using FMRIB’s Linear Image Registration Tool (FLIRT). Z(Gaussianized T/F) statistic images were threshold using clustersdetermined by Z > 2.3 and a corrected cluster significance thresh-old of P = 0.05.

The combination of the different runs at the individual subjectlevel was carried out using a fixed effects model. Individual designmatrices were created, modeling the different behavioral condi-tions. Contrast images of interest were produced from these indi-vidual analyses and used in the second-level higher analysis.Higher-level between-subject analysis was carried out using amixed effects analysis with the FLAME (FMRIB’s Local Analysis ofMixed Effects) tool, part of FSL. Final statistical images were cor-rected for multiple comparisons using Gaussian Random Field-based cluster inference with a height threshold of Z > 2.3 and acluster significance threshold of P < 0.05.

Continuous physiological monitoring of the respiratory and car-diac cycles and the end tidal carbon dioxide levels, can be used tocorrect retrospectively for some of the artifacts inherent in fMRIdata (Glover, Li, & Ress, 2000; Oldfield et al. 2011; Wise, Ide, Poulin,& Tracey, 2004), but this is more practical when there is continuousimage acquisition and not when the data are acquired by sparse-sampling. Therefore, this data were not acquired during this study.An inability to correct for these sources of noise will have reducedthe signal-to-noise ratio, reducing the chance of demonstratingsignal coupled to neural activity in regions with subtle changesin BOLD signal (Birn, Diamond, Smith, & Bandettini, 2006).

2.5. Independent component analysis (ICA)

This was carried out using group temporal concatenation prob-abilistic ICA implemented in MELODIC (Multivariate Exploratory

Fig. 1. Standard sagittal T1-weighted anatomical slices through the left and right cerebral hemispheres, overlaid with activity from the contrast of spoken languageproduction (speech) with non-communicative repetitive tongue movements (tongue). The statistical threshold was set at Z > 2.3, cluster-corrected. Anterior is to the left. TheMNI co-ordinates are along the x-axis, negative being to the left. The greater the number, the more lateral the sagittal slice. Regions of activity were located in: (1) pre-SMAand anterior cingulate cortex; (2) bilateral anterior insula; (3) bilateral superior temporal cortex including left and right medial planum temporale; (4) left posterior inferiorgyrus (incorporating Broca’s area) and extending dorsally into posterior middle frontal gyrus.

50 F. Geranmayeh et al. / Brain & Language 121 (2012) 47–57

Linear Decomposition into Independent Components) Version3.10, part of the FSL software (Beckmann & Smith, 2004). This ap-proach to the ICA was used rather than tensor-ICA (Beckmann &Smith, 2005), as the temporal presentation of the stimuli was dif-ferent between subjects. Such multivariate analysis can extractimportant information from the data that is not always apparentfrom a subtractive univariate analysis (e.g., Leech et al., 2012).ICA takes advantage of fluctuations in the fMRI data to separatethe signal into spatially distinct components. With a TR of 10 s,the measured fluctuations in this data will be restricted to low fre-quency fluctuations. A particular advantage of ICA, which increasessensitivity when detecting net regional neural responses, is con-trolling for time-series unrelated to brain function. These will beidentified as separate components; for example, movement-re-lated artifact not removed by the initial image pre-processing.

Data pre-processing for the ICA included masking of non-brainvoxels, voxel-wise de-meaning of the data and normalization of

Fig. 2. Standard T1-weighted anatomical slices overlaid with activity from the contrast oproduction (speech). The statistical threshold was set at Z > 3, cluster-corrected. The fourhemispheres, with anterior to the left. The MNI co-ordinates are along the x-axis, negativeThe final image, on the right, is a coronal slice. Activity was predominantly distributedright mid-insular cortex; (2) posterior cingulate cortex; (3) left and right motor-sensorextending back into lateral occipital cortex.

the voxel-wise variance of the noise. The ICA was set up to decom-pose the data into 30 independent components containing distrib-uted neural networks, movement artifact and physiological noise.The choice of the number of component maps reflects a trade-offbetween granularity and noise. It is motivated by the attempt tomaximize the homogeneity of function within each network whilemaximizing the heterogeneity between them. Previous applica-tions of ICA to fMRI data have used 20–30 component maps (Beck-mann, DeLuca, Devlin, & Smith, 2005; Leech et al., 2011; Smithet al., 2009), and this study adopted the same approach.

The data were projected into a 30-dimensional subspace usingprincipal component analysis. The whitened observations weredecomposed into a set of 30 component maps and associated vec-tors describing the temporal variations across all runs and subjectsby optimizing for non-Gaussian spatial source distributions using afixed-point iteration technique (Hyvärinen, 1999). Estimated com-ponent maps were divided by the standard deviation of the

f non-communicative repetitive tongue movements (tongue) with spoken languageimages on the left of the figure are sagittal slices through the left and right cerebralbeing to the left. Next there are two axial slices, the co-ordinates being in the z-axis.

across sensory cortices and posterior half of the default mode network: (1) left andy cortex; (4) precuneus; (5) right posterior inferior parietal cortex (angular gyrus)

Fig. 3. Standard T1-weighted anatomical slices showing the three components from the ICA analysis: 1, 24 and 30. Components 1 (shown in blue) and 24 (shown in red/yellow) demonstrated correlated activity for speech that was significantly greater than for tongue. Component 30 (shown in green) demonstrated correlated activity fortongue that was significantly greater than for speech. The statistical threshold was set at Z > 4. The images are sagittal views (MNI co-ordinates in the x-axis, negative being tothe left), with, in addition, one coronal slice (MNI co-ordinate in the y-axis) and one axial slice (MNI co-ordinate in the z-axis). The results from the ICA analysis demonstrateda much more widely distributed network for speech compared to the univariate analysis displayed in Fig. 1. Component 30 revealed a network, predominantly posterior, witha distribution attributable to the default mode network. The numbered regions showing correlated activity for speech greater than for tongue are: (1) posterior cingulatecortex; (2) dorsal anterior cingulate cortex; (3) left and right inferior parietal cortex; (4) left dorsolateral prefrontal cortex, including Broca’s area; (5) left inferolateraltemporal cortex; (6) pre-supplementary area and dorsal anterior cingulate cortex; (7) lateral premotor cortex; (8) anterior insula; (9) left and right superior temporal gyri.The numbered regions showing correlated activity for tongue greater than for speech are: (10) bilateral inferior parietal cortex; (11) extensive posterior medial activity, inprecuneus and posterior cingulate cortex. Areas of overlap between Components 24 and 30 in the left inferior parietal cortex and posterior midline regions are demonstratedwith I. An area of overlap between Components 1 and 24 in Broca’s area is demonstrated with §.

F. Geranmayeh et al. / Brain & Language 121 (2012) 47–57 51

residual noise and thresholded by fitting a Gaussian/Gamma mix-ture model to the histogram of intensity values (Beckmann &Smith, 2004). Of the 30 components, 19 were clearly related toresidual movement artifact (characterized by the majority of thesignal distributed around the edge of the brain or within the lateralventricles). Eleven had correlated signal distributed within brainparenchyma. Of these, we chose the two, illustrated in Fig. 3, asbeing most likely to be involved in language processing, based onextensive meta-analyses of the cortical components of languageand speech (Smith et al., 2009; Vigneau et al., 2006). In addition,one further component was included, one that incorporated a largepart of the default mode network. This was included because it hasbeen argued (Binder et al., 1999) that language-related processes,particularly the retrieval of semantic knowledge, will occur bothduring speech comprehension and production, and an awake brainwithout external stimulation. The associated subject-specific timecourses for these three components were regressed against thegeneral linear model design matrices and tested for significance(P < 0.05) in order to identify within these three componentswhether activity was greater during Speech than Tongue, and viceversa. As the analysis was highly hypothesis driven (given therestriction of the analysis to these three components), correctionfor multiple comparisons was not made.

2.6. Region of interest analysis (ROI)

To determine the activation of brain regions within one of theICA components, region of interest (ROI) analysis were carriedout. ROIs were defined based on the results of the group ICA anal-ysis. Component 24 (C24), showed regions in which coherentactivity was greater for Speech than Tongue, including the leftparietal (IPC), left inferolateral temporal (ILT) and left posteriorfrontal regions. Since we were interested in the role of left IPC

in spoken language production, we used this component as thebasis of the ROI analysis. This component was split using athreshold of Z = 5, a level chosen as it isolated relatively large re-gions within each of the three lobes, to create separate anatomi-cal masks of left IPC (6407 voxels, each voxel 8 mm3), frontal(5440 voxels) and ILT (1552 voxels) regions (Fig. 4). Using FSLFEATQuery, each ROI mask was re-registered to the same spaceas individual pre-processed functional data from the univariateanalysis. Within each ROI, mean BOLD activity effect size duringSpeech and Tongue contrasted with Rest, across the two runs,were calculated for individual subjects. Since the regions wereidentified based on a component that was itself selected becauseit showed a fit to the general linear model, inferential group sta-tistics were not considered to be valid. However the mean per-centage signal change, with 95% confidence intervals, within theROIs provided useful descriptive information.

3. Results

3.1. Rate of speech production

During Speech, the mean rate of syllable production across thetwo runs was 16.3 syllables per 7 s epoch (range of number of syl-lables produced in a trial, across all subjects and all trials: 7.9–22.1syllables). The mean rate of content word production was 10 wordsper 7 s epoch (range of number of content word produced in a trial,across all subjects and all trials: 5.2–14.9 words). There was nooverall difference in mean performance between the two runs(P = 0.9 for syllables; P = 0.8 for content words). The rates for bothmean number of syllable and mean number of content word pro-duction were highly correlated between the two runs for each sub-ject (correlation coefficient, r = 0.87 for syllables; r = 0.9 for contentwords). Unsurprisingly, the overall rates of production of syllables

Fig. 4. (A) This illustrates the left inferior parietal (blue), left dorsal frontal (yellow), and left inferolateral temporal regions-of-interest, as determined from the ICA analysis(regions 3, 4 and 5 in Fig. 3). (B) The bars, with 95% confidence intervals, show mean percentage signal change averaged across the region of interest during speech andtongue, relative to Rest. There was a deactivation in the left parietal region during speech compared to tongue. In the frontal region, there was more activity during speechcompared to tongue and a smaller but similar trend was seen in the inferolateral temporal region.

52 F. Geranmayeh et al. / Brain & Language 121 (2012) 47–57

and content words were highly correlated across each subject (cor-relation coefficient, r = 0.9). The rate of performing repetitive ton-gue movements were not recorded, but subjects had rehearsed toperform the task at a rate of �1/s.

3.2. Univariate subtraction analysis

The contrast of Speech with Tongue demonstrated significantlygreater activity in the following cerebral cortical regions: the pre-supplementary motor area (pre-SMA), merging with activity in thedorsal anterior cingulate cortex (ACC); left dorsolateral prefrontalcortex (dlPFC), with activity in the middle frontal gyrus extendinginto the left inferior frontal gyrus (IFG), incorporating the pars tri-angularis (pTr), pars opercularis (pOp) and lateral orbito-frontalcortex (lOFC); left and right anterior insula; and left and right supe-rior temporal gyri (incorporating primary auditory cortex, planumtemporale and lateral temporal neocortex) as far ventral as themore dorsal middle temporal gyrus) (Table 1A, Fig. 1).

The reverse contrast of Tongue with Speech revealed activitydistributed across the default mode network predominantly in itsposterior extent (midline and lateral parietal cortices) (Table 1A,Fig. 2). The lateral parietal activity was more evident on the rightthan on the left, although it is usually described as symmetricalfor the default mode network (Buckner et al., 2008; Gusnard & Rai-chle, 2001; Raichle et al., 2001); this provided an indirect clue thatleft parietal cortex was, in part, engaged by the Speech condition.In addition, there was bilateral anterior parietal and mid-insular

activity, presumably related to the greater somatosensory feed-back during the Tongue condition (Dhanjal et al., 2008).

3.3. Independent component analysis

Of the thirty components generated, 19 were related to move-ment and other sources of artifact, as evidenced by activity as arim around the edge of the brain, within the lateral ventricles, orthroughout the venous sinuses. These were excluded from furtherconsideration. Eleven distributed brain networks were identifiedwith correlated signal confined to the brain parenchyma. Anothereight were excluded from further consideration, as the correlatedsignal was distributed throughout regions that have not been re-lated to language and semantic processing per se (Binder et al.,2009; Vigneau et al., 2006). Examples of excluded components(not illustrated), are the eighth component with activity confinedto the cerebellum, the 18th component with activity in the visualcortex and the second component which was distributed over vi-sual, bilateral primary motor-somatosensory cortices and the sup-plementary motor area. This last component was equally activeduring both Speech and Tongue conditions, and is best interpretedas a distributed system for perception of the visual stimuli and theexecution of a motor response, irrespective of the task demand.

The two components related specifically to Speech are shown inFig. 3 (see also Table 1B); Component one (C1), and Component 24(C24) accounted for 17.9% and 0.7% of the variance respectively. C1,in which coherent activity during Speech was significantly greater

Table 1(A) Univariate subtractive analysis. Activation peaks within significant clusters (P < 0.05 cluster corrected) for univariate subtractive contrasts of speech against tongue (Fig. 1),and tongue against speech (Fig. 2). (B) Independent component analysis. Sub-peaks within the distributed areas of coherent activity during speech and tongue (Fig. 3). Component1 (C1) includes the blue areas of correlated activity, Component 24 (C24) includes the red/yellow areas of correlated activity, and Component 30 (C30) refers to green areas ofcorrelated activity shown in Fig. 3. MNI coordinates refer to the voxel with the highest Z value situated within the center of a particular region. Cluster sizes are included for areasof activity determined by the univariate analysis.

F. Geranmayeh et al. / Brain & Language 121 (2012) 47–57 53

than during Tongue (t (17) = 1.73, P = 0.05), encompassed the fol-lowing regions: pre-SMA and dorsal ACC; left and right lateral pre-motor and primary motor cortex; the left dlPFC including, left IFG,pOp, pTr and lOFC; left and right anterior insula; and the length ofboth superior temporal gyri, including planum temporale. Addi-tional subcortical correlated activity, not seen in the thresholded

univariate analysis, was observed in the left and right putamen,thalamus and paravermal cerebellum. Therefore, this componentproved sensitive to revealing most of the regions previously de-scribed as being involved in the motor control of speech produc-tion (Bohland & Guenther, 2006; Guenther et al., 2005; Rieckeret al., 2005).

54 F. Geranmayeh et al. / Brain & Language 121 (2012) 47–57

The second component, C24, also showed correlated activity sig-nificantly greater for Speech than for Tongue (t (17) = 2.11,P = 0.025). As illustrated in Fig. 3 (see also Table 1B), C24 was a pre-dominantly left-lateralized frontal–temporal–parietal network.The constituent parts were: medial frontal cortex, centered on par-acingulate cortex with a separate region in posterior cingulate cor-tex; an extensive left dlPFC including left IFG, pOp, pTr and lOFC;left anterior insula; left inferolateral temporal cortex (ILT); the leftIPC, including both supramarginal and angular gyri, extending asfar dorsal as the intraparietal sulcus and as far posterior as the lat-eral occipital cortex; and the right IPC, confined to the dorsal half ofthe angular gyrus. The only subcortical regions were the left puta-men and right lateral cerebellum. There were areas of partial over-lap with C1, shown in Fig. 3.

Component 30 (C30) has also been included (Fig. 3 andTable 1B). This component, in which coherent activity during Ton-gue was significantly greater than that during Speech (t (17) = 3.15,P = 0.003), accounted for 4.23% of the variance. It comprised a pre-dominantly posterior network, with a distribution conforming tothe posterior half of the default mode network, a network in whichcorrelated activity is most evident in states of ‘rest’ or automaticrepetitive passive tasks when subjects are not engaged in exter-nally oriented decisions.

3.4. ROI analysis

The ROI analyses are displayed in Fig. 4. The left inferior frontalROI predictably showed more activity during Speech relative toTongue as evident in the whole-brain contrast, and there was asmaller but similar trend in the same direction in the left ILT cor-tex. In contrast, activity in the large left IPC decreased in Speech.This is in contrast to the ICA analysis that demonstrated a compo-nent where the activity in this region was associated with Speech,even though the ROI results showed that the overall activity in theleft IPC was less during Speech than Tongue.

4. Discussion

The present study clearly demonstrates the participation of theleft IPC in spoken language production using discourse rather thansingle word or sentences. This was evident as a left lateralizedfrontal–temporal–parietal component, with coherent activity inthe left IPC and other language regions in left ILT region and dLPFC,including ‘classic’ Broca’s area, during the Speech compared to theTongue task. The apparent paradox was that overall activity in theIPC appeared to be weaker during Speech than Tongue as sug-gested by the univariate ROI analysis. This can be attributed, atleast in part, to greater activity within the default mode networkduring the Tongue task. It required the analysis with ICA to sepa-rate out the distinct but overlapping components of task relatedactivity in the IPC.

4.1. Univariate subtraction analysis

The starting point was the univariate analysis of the contrast ofSpeech with Tongue. As a first pass, it might be expected that theunivariate contrast would reveal summed cerebral activity acrossall the processing levels involved in producing spoken discourse,for example, from the formulation of the message through to theconstruction and completion of an articulatory plan (Indefrey & Le-velt, 2004; Levelt, 2001). Using a conventional statistical threshold,the univariate analysis ‘subtracted’ activity related to somatosen-sory feedback and motor feedforward activity in primary motor-sensory cortex that is expected in the Tongue task (Bookheimer,Zeffiro, Blaxton, Gaillard, & Theodore, 2005). It also subtracted sub-

cortical structures, including the paravermal cerebellum, to finallyreveal activity confined to the dorsal frontal (midline and lateral),the anterior insular cortex and the superior temporal regions.

Much of the left inferior frontal activity, extending ventrally intoclassic Broca’s area (Brodmann areas 44 and 45) and lOFC, can beattributed to language-specific processes, including lexical selectionand phonological and articulatory planning. This conclusion is basedon many decades of clinical studies (Caplan, 1987) and meta-analy-ses of functional neuroimaging language studies that implicate thisregion in the processing of phonological, semantic and syntacticinformation (Binder et al., 2009; Indefrey & Levelt, 2004; Vigneauet al., 2006). However, this region is also shown to be responsiblefor domain-general conflict resolution, and part of its relationshipto language may be that conflicts arise when selecting among arange of possible semantic and lexical representations during spo-ken language (Novick, Kan, Trueswell, & Thompson-Schill, 2009;Schnur et al., 2009; Thompson-Schill, D’Esposito, & Kan, 1999).

Midline dorsal frontal activity, in the ACC (Picard & Strick, 2001)and more dorsal midline cortex of the superior frontal gyrus, inpre-SMA (Klein et al. 2007), have also been strongly associatedwith speech production (Bohland & Guenther, 2006; Rieckeret al., 2005; Tremblay & Gracco, 2009). Again, however, activityin these regions is not necessarily language-specific within thecontext of the Speech task, which required subjects to choose fromamong multiple items of stored semantic knowledge and overtlyexpress a few under time constraint. These same midline dorsalfrontal regions, often operating through anatomical and functionalconnections with the lateral prefrontal cortex, have also beenimplicated in domain-general cognitive control across many differ-ent kinds of task (Kerns et al., 2004; Miller & Cohen, 2001; Ridder-inkhof, Ullsperger, Crone, & Nieuwenhuis, 2004). Activity inrelation to domain-general cognitive control may also apply tothe bilateral activity in the insular cortices. The superior precentralgyrus of the left anterior insula has been associated with a speech-specific role, based on clinical (Baldo, Wilkins, Ogar, Willock, &Dronkers, 2011; Dronkers, 1996) and on functional imaging studies(Ackermann & Riecker, 2004; Riecker et al., 2005). However, thepresent study demonstrated increased activity in more ventraland bilateral anterior insular cortices during Speech compared toTongue movement. In keeping with previous study by Rieckerand colleagues (2000), there were no insular cortical activity dur-ing the Tongue condition. The discrepancy between bilateral insu-lar activity associated with Speech in the present study and the leftlateralized speech related insular activity in previous studies, maybe the result of involvement of this region in cognitive control thatis not specific to language (Ullsperger, Harsay, Wessel, & Ridder-inkhof, 2010), but is engaged by the nature of task used. Alterna-tively this discrepancy could be due to the processing of internalmental states elicited by external events, expressed through auto-nomic outflow (Seeley et al., 2007). This processing may be an ex-pected consequence of arousal during the time-restricted cognitivetask used in the present study.

The posterior cortical activity observed in the univariate con-trast of Speech against Tongue was limited to the left and rightsuperior temporal gyri. Much of this activity is likely attributableto post-articulatory self-monitoring (Golfinopoulos, Tourville, &Guenther, 2010) and processing of self-generated auditory feed-back related to speech production. The study design does not allowus to distinguish this processing from the planning or executionprocesses of speech production given the absence of an auditorybaseline task. No activity that survived the statistical thresholdwas evident in more ILT region and there was a depression of activ-ity in the left IPC during Speech as evident in the ROI analysis. Thisis in contrast to meta-analyses of functional neuroimaging studiesthat have implicated these regions in linguistic and semantic pro-cesses (Binder et al., 2009; Vigneau et al., 2006).

F. Geranmayeh et al. / Brain & Language 121 (2012) 47–57 55

4.2. ICA analysis

It required the ICA analysis to demonstrate the posterior corti-cal regions involved in the production of spoken discourse. Thefirst component of the ICA, C1, demonstrated coherent activityacross a distributed system that largely mirrored the result fromthe univariate analysis: a network comprising the medial frontal,dlPFC (including Broca’s area in the posterior IFG), left and rightanterior insula, and the left and right superior temporal gyri. Incontrast, the 24th component, C24, demonstrated a largely left-lat-eralized frontal–temporal–parietal network, of which only thefrontal cortex was evident in the univariate analysis. In a studyby Smith and colleagues (2009) relating resting state networks ina population of subjects, to the distribution of activation focigleaned from a large database of functional imaging studies, theauthors demonstrated a left frontal–temporal–parietal restingstate network that was related to a range of language activationstudies. This component had an almost identical distribution tothat observed in our C24, and we showed that C24 was overall moreactive for Speech than Tongue. This, is therefore more direct evi-dence for the role of this left frontal–temporal–parietal networkin speech, over and above comparing the distribution of the net-work, at rest, to previously published foci of language activationas performed by Smith and colleagues (2009).

Fig. 3 demonstrates a few regions where there is partial overlapbetween the networks identified by components C1 and C24. Theseparation of the networks centered on the left inferior frontalgyrus is of particular interest, as it mirrors a study on resting statenetworks in the human brain (Kelly et al., 2010). In that study,functional connectivity with ventral premotor cortex was distrib-uted between primary motor-sensory cortex and the length ofthe superior temporal gyrus, whereas functional connectivity withthe adjacent Broca’s area was with ILT and IPC.

4.3. Role of parietal lobe in spoken language production

Although the whole-brain univariate analysis demonstrated noactivity in IPC and the ROI showed a depression of activity in thisregion during Speech, C24 of the ICA revealed more coherent activ-ity during Speech production than Tongue movement in the IPC,left frontal and posterior ILT regions. This implicates the IPC asone component of a very distributed left hemisphere cortical sys-tem involved in formulating and/or production of spoken sentenceproduction. In a study by Van de Ven, Esposito, and Christoffels(2009) that also used ICA to identify coherent language networks,two networks involving the left IPC, were identified during picturenaming compared to other tasks. Clinical lesion studies (Borovskyet al., 2007; DeLeon et al., 2007) and meta-analyses of languagestudies have demonstrated that the IPC is associated with languageprocessing, particularly the processing of semantic content involv-ing the angular gyrus (Binder et al., 2009; Vigneau et al., 2006;). Inaddition to being associated with semantic processing, the leftangular gyrus has been identified as part of the default mode net-work (Buckner et al., 2008; Gusnard & Raichle, 2001; Raichle et al.,2001) that is deactivated during goal-directed tasks compared withbaseline. Seghier and colleagues (2010), using univariate contrastsbetween a large range of different language conditions, have dem-onstrated that there is a partial overlap between the semantic sys-tem and the default mode network within the ventral-medial-dorsal extent of the left angular gyrus. Similarly, in the presentstudy there was some overlap of correlated activity in left IPC forthe Speech-related frontal–temporal–parietal network (C24) andthe Tongue-related default mode network (C30) as shown in Fig. 3.

The data-driven ICA analysis cannot determine the specific pro-cesses during spoken language that are supported by IPC. C24 high-lights the entire IPC region, which is now known to consist of at

least seven cytoarchitectonic zones (Caspers et al., 2006) with sep-arable anatomical white matter connections (Mars et al. 2011).This anatomical division implies that each zone may have a differ-ent processing function. Individual language studies have impli-cated parietal cortex in processes as diverse as selection ofarticulatory gestures (which may be domain-general, and includenon-communicative movements of the articulators) (Tremblay &Gracco, 2010) to the integration of verbal information over long(sentence and paragraph length) time scales during speech com-prehension (Lerner, Honey, Silbert, & Hasson, 2011), and therefore,quite possibly, during the construction of connected speech. Interms of more distributed network function, activity distributedbetween left frontal and IPC cortex may indicate the operation ofverbal working memory (Jacquemot & Scott, 2006), and betweenleft frontal and temporal cortex the operation of syntactic pro-cesses (Tyler et al., 2010), networks that will operate in parallelwhen defining a noun cue. Whereas investigating low-order corti-ces, such as the early visual system, has worked well with func-tional imaging, high-order cortices, such as in the parietal lobe,are perhaps best understood in terms of overlapping componentssupporting separate distributed functional networks (Henson,2005). It has been proposed that these networks work in parallelto achieve a single but computationally complex goal. It is not pos-sible from this study alone, when the two tasks were so widelyseparated in terms of component cognitive processes, to assert thatcoordinated activity across the left frontal–temporal–parietal net-work is language-specific, and the activity in any one of these indi-vidual broad cortical regions may also contribute to othernetworks that are not specific to speech production. However,the data from this study adds further evidence in favor of this fron-tal–temporal–parietal network being associated with language,when related to the meta-analyses of others, such as Vigneau(2006) and Smith (2009) and their colleagues.

Therefore, the left dlPFC, IPC and posterior ILT contributions tospoken language production evident in the ICA analysis, the conse-quence of exploring low frequency fluctuations in fMRI BOLD sig-nal, unmasked the functional role of the IPC during the Speechtask. This was despite a net decrease in regional activity as otherco-localized neuronal populations, involved in other cognitive pro-cesses, were simultaneously suppressed during Speech.

5. Conclusion

This study confirms that a large extent of the left IPC is involvedin spoken discourse, a crucial aspect of language production. It isprobable that within this extent there are subsystems performingmore than one processing role to complete such a complex cogni-tive act. This notion is captured by the meta-analysis of Vigneauand colleagues (2006), although the great majority of studies in-cluded in that review had investigated language tasks that werenot directly related to the production of spoken discourse. Whatis apparent from this study is that the response of the IPC wasnot captured by a relative increase in BOLD signal. Combining aunivariate subtraction analysis with a multivariate analysis has en-hanced the interpretation of this study. The reliable presence of theleft frontal–temporal–parietal system as a resting state network(Smith et al., 2009) suggests that this is a robust and stable distrib-uted system in the healthy adult human brain.

References

Ackermann, H., & Riecker, A. (2004). The contribution of the insula to motor aspectsof speech production: A review and a hypothesis. Brain and Language, 89(2),320–328.

Awad, M., Warren, J. E., Scott, S. K., Turkheimer, F. E., & Wise, R. J. S. (2007). Acommon system for the comprehension and production of speech. Journal ofNeuroscience, 27, 11455–11464.

56 F. Geranmayeh et al. / Brain & Language 121 (2012) 47–57

Baldo, J. V., Wilkins, D. P., Ogar, J., Willock, S., & Dronkers, N. F. (2011). Role of theprecentral gyrus of the insula in complex articulation. Cortex, 47(7), 800–807.

Beckmann, C. F., DeLuca, M., Devlin, J. T., & Smith, S. M. (2005). Investigations intoresting-state connectivity using independent component analysis. PhilosophicalTransactions of the Royal Society Biological Science, 360, 1001–1013.

Beckmann, C. F., & Smith, S. M. (2004). Probabilistic independent componentanalysis for functional magnetic resonance imaging. IEEE Transactions onMedical Imaging, 23, 137–152.

Beckmann, C. F., & Smith, S. M. (2005). Tensorial extensions of independentcomponent analysis for multi subject FMRI analysis. Neuroimage, 25(1), 294–311.

Binder, J. R., Desai, R. H., Graves, W. W., & Conant, L. L. (2009). Where is the semanticsystem? A critical review and meta-analysis of 120 functional neuroimagingstudies. Cerebral Cortex, 19, 2767–2796.

Binder, J. R., Frost, J. A., Hammeke, T. A., Bellgowan, P. S. F., Rao, S. M., & Cox, R. W.(1999). Conceptual processing during the conscious resting state: A functionalMRI study. Journal of Cognitive Neuroscience, 11, 80–93.

Birn, R. M., Diamond, J. B., Smith, M. A., & Bandettini, P. A. (2006). Separatingrespiratory-variation-related fluctuations from neuronal-activity-relatedfluctuations in fMRI. Neuroimage, 31, 1536–1548.

Blank, S. C., Scott, S. K., Murphy, K., Warburton, E., & Wise, R. J. S. (2002). Speechproduction: Wernicke, Broca and beyond. Brain, 125, 1829–1838.

Bohland, J. W., & Guenther, F. H. (2006). An fMRI investigation of syllable sequenceproduction. Neuroimage, 32, 821–841.

Bookheimer, S. Y., Zeffiro, T. A., Blaxton, T. A., Gaillard, W., & Theodore, W. H. (2005).Activation of language cortex with automatic speech tasks. Neurology, 2000(55),1151–1157.

Borovsky, A., Saygin, A. P., Bates, E., & Dronkers, N. (2007). Lesion correlates ofconversational speech production deficits. Neuropsychologia, 45, 2525–2533.

Braun, A. R., Guillemin, A., Hosey, L., & Varga, M. (2001). The neural organization ofdiscourse: An H2

15O-PET study of narrative production in English and Americansign language. Brain, 124, 2028–2044.

Brownsett, S. L. E., & Wise, R. J. S. (2010). The contribution of the parietal lobes tospeaking and writing. Cerebral Cortex, 20, 517–523.

Buckner, R. L., Andrews-Hanna, J. R., & Schacter, D. L. (2008). The brain’s defaultnetwork: Anatomy, function, and relevance to disease. Annals of the New YorkAcademy of Sciences, 1124, 1–38.

Caplan, D. (1987). Neurolinguistics and linguistic aphasiology: An introduction.Cambridge, UK: Cambridge University Press.

Caspers, S., Geyer, S., Schleicher, A., Mohlberg, H., Amunts, K., & Zilles, K. (2006). Thehuman inferior parietal cortex: Cytoarchitectonic parcellation andinterindividual variability. Neuroimage, 33(2), 430–448.

Catani, M., & ffytche, D. H. (2005). The rises and falls of disconnection syndromes.Brain, 128, 2224–2239.

DeLeon, J., Gottesman, R. F., Kleinman, J. T., Newhart, M., Davis, C., Heidler-Gary, J.,et al. (2007). Neural regions essential for distinct cognitive processes underlyingpicture naming. Brain, 30, 1408–1422.

Dhanjal, N. S., Handunnetthi, L., Patel, M. C., & Wise, R. J. (2008). Perceptual systemscontrolling speech production. Journal of Neuroscience, 28, 9969–9975.

Dronkers, N. F. (1996). A new brain region for coordinating speech articulation.Nature, 384, 159–161.

Esposito, F., Bertolino, A., Scarabino, T., Latorre, V., Blasi, G., Popolizio, T., et al.(2006). Independent component model of the default-mode brain function:Assessing the impact of active thinking. Brain Research Bulletin, 70(4–6),263–269.

Fridriksson, J., Kjartansson, O., Morgan, P. S., Hjaltason, H., Magnusdottir, S., Bonilha,L., et al. (2010). Impaired speech repetition and left parietal lobe damage.Journal of Neuroscience, 30(33), 11057–11061.

Glover, G. H., Li, T. Q., & Ress, D. (2000). Image-based method for retrospectivecorrection of physiological motion effects in fMRI: RETROICOR. MagneticResonance in Medicine, 44, 162–167.

Golfinopoulos, E., Tourville, J. A., & Guenther, F. H. (2010). The integration of large-scale neural network modeling and functional brain imaging in speech motorcontrol. Neuroimage, 52, 862–874.

Gracco, V. L., Tremblay, P., & Pike, B. (2005). Imaging speech production using fMRI.Neuroimage, 26, 294–301.

Guenther, F. H., Ghosh, S. S., & Tourville, J. A. (2005). Neural modelling and imagingof the cortical interactions underlying syllable production. Brain and Language,96, 280–301.

Gusnard, D. A., & Raichle, M. E. (2001). Searching for a baseline: Functional imagingand the resting human brain. Nature Reviews Neuroscience, 2, 685–694.

Hall, D. A., Haggard, M. P., Akeroyd, M. A., Palmer, A. R., Summerfield, A. Q., Elliott,M. R., et al. (1999). ‘‘Sparse’’ temporal sampling in auditory fMRI. Human BrainMapping, 7, 213–223.

Haller, S., Radue, E. W., Erb, M., Grodd, W., & Kircher, T. (2005). Overt sentenceproduction in event-related fMRI. Neuropsychologia, 43, 807–814.

Henson, R. (2005). What can functional neuroimaging tell the experimentalpsychologist? Quarterly Journal of Experimental Psychology A, 58(2), 193–233.

Hyvärinen, A. (1999). Fast and robust fixed-point algorithms for independentcomponent analysis. IEEE Transactions on Neural Networks, 10, 626–634.

Indefrey, P., & Levelt, W. J. M. (2004). The spatial and temporal signatures of wordproduction components. Cognition, 92, 101–144.

Jacquemot, C., & Scott, S. K. (2006). What is the relationship between phonologicalshort-term memory and speech processing? Trends in Cognitive Sciences, 10,480–486.

Jenkinson, M., Bannister, P., Brady, J. M., & Smith, S. M. (2002). Improvedoptimisation for the robust and accurate linear registration and motioncorrection of brain images. Neuroimage, 17, 825–841.

Kelly, C., Uddin, L. Q., Shehzad, Z., Margulies, D. S., Castellanos, F. X., Milham, M. P.,et al. (2010). Broca’s region: Linking human brain functional connectivity dataand non-human primate tracing anatomy studies. European Journal ofNeuroscience, 32, 383–398.

Kemeny, S., Ye, F. Q., Birn, R., & Braun, A. R. (2005). Comparison of continuous overtspeech fMRI using BOLD and arterial spin labeling. Human Brain Mapping, 24(3),173–183.

Kerns, J. G., Cohen, J. D., MacDonald, A. W., 3rd, Cho, R. Y., Stenger, V. A., & Carter, C.S. (2004). Anterior cingulate conflict monitoring and adjustments in control.Science, 303, 1023–1026.

Klein, J. C., Behrens, T. E., Robson, M. D., Mackay, C. E., Higham, D. J., & Johansen-Berg, H. (2007). Connectivity-based parcellation of human cortex usingdiffusion MRI: Establishing reproducibility, validity and observerindependence in BA 44/45 and SMA/pre-SMA. Neuroimage, 34(1), 204–211.

Leech, R., Braga, R., & Sharp, D. (2012). Echoes of the brain within the posteriorcingulate cortex. The Journal of Neuroscience, 32(1), 215–222.

Leech, R., Kamourieh, S., Beckmann, C. F., & Sharp, D. J. (2011). Fractionating thedefault mode network: Distinct contributions of the ventral and dorsalposterior cingulate cortex to cognitive control. Journal of Neuroscience, 31(9),3217–3224.

Lerner, Y., Honey, C. J., Silbert, L. J., & Hasson, U. (2011). Topographic mapping of ahierarchy of temporal receptive windows using a narrated story. Journal ofNeuroscience, 31(8), 2906–2915.

Levelt, W. J. M. (2001). Spoken word production: A theory of lexical access.Proceedings of the National Academy of Sciences United States of America, 98,13464–13471.

Mars, R. B., Jbabdi, S., Sallet, J., O’Reilly, J. X., Croxson, P. L., Olivier, E., et al. (2011).Diffusion-weighted imaging tractography-based parcellation of the humanparietal cortex and comparison with human and macaque resting-statefunctional connectivity. Neuroscience, 31(11), 4087–4100.

McKiernan, K. A., D’Angelo, B. R., Kaufman, J. N., & Binder, J. R. (2006). Interruptingthe ‘‘stream of consciousness’’: An fMRI investigation. Neuroimage, 29,1185–1191.

Mehta, S., Grabowski, T. J., Razavi, M., Eaton, B., & Bolinger, L. (2006). Analysis ofspeech-related variance in rapid event-related fMRI using a time-awareacquisition system. Neuroimage, 29, 1278–1293.

Miller, E. K., & Cohen, J. D. (2001). An integrative theory of prefrontal cortexfunction. Annual Review of Neuroscience, 24, 167–202.

Novick, J. M., Kan, I. P., Trueswell, J. C., & Thompson-Schill, S. L. (2009). A case forconflict across multiple domains: Memory and language impairments followingdamage to ventrolateral prefrontal cortex. Cognitive Neuropsychology, 26,527–567.

Oldfield, E. H., Brooks, J. C. W., Wise, R. J. S., Padormo, F., Hajnal, J. V., Beckmann, C. F.,et al. (2011). Identification and characterisation of midbrain nuclei usingoptimised functional magnetic resonance imaging. Neuroimage. doi:10.1016/j.neuroimage.2011.08.016.

Papoutsi, M., Stamatakis, E. A., Griffiths, J., Marslen-Wilson, W. D., & Tyler, L. K.(2011). Is left fronto-temporal connectivity essential for syntax? Effectiveconnectivity, tractography and performance in left-hemisphere damagedpatients. Neuroimage, 58(2), 656–664.

Picard, N., & Strick, P. L. (2001). Imaging the premotor areas. Current Opinion inNeurobiology, 11, 663–672.

Raichle, M. E., MacLeod, A. M., Snyder, A. Z., Powers, W. J., Gusnard, D. A., & Shulman,G. L. (2001). A default mode of brain function. Proceedings of the NationalAcademy of Sciences of the USA, 98(2), 676–682.

Ridderinkhof, K. R., Ullsperger, M., Crone, E. A., & Nieuwenhuis, S. (2004). The role ofthe medial frontal cortex in cognitive control. Science, 306, 443–447.

Riecker, A., Ackermann, H., Wildgruber, D., Meyer, J., Dogil, G., Haider, H., et al.(2000). Articulatory/phonetic sequencing at the level of the anterior perisylviancortex: A functional magnetic resonance imaging (fMRI) study. Brain andLanguage, 75, 259–276.

Riecker, A., Mathiak, K., Wildgruber, D., Erb, M., Hertrich, I., Grodd, W., et al. (2005).FMRI reveals two distinct cerebral networks subserving speech motor control.Neurology, 64, 700–706.

Schnur, T. T., Schwartz, M. F., Kimberg, D. Y., Hirshorn, E., Coslett, H. B., & Thompson-Schill, S. L. (2009). Localizing interference during naming: Convergentneuroimaging and neuropsychological evidence for the function of Broca’sarea. Proceedings of the National Academy of Sciences of the United States ofAmerica, 106, 322–327.

Seeley, W. W., Menon, V., Schatzberg, A. F., Keller, J., Glover, G. H., Kenna, H., et al.(2007). Dissociable intrinsic connectivity networks for salience processing andexecutive control. Journal of Neuroscience, 27, 2349–2356.

Seghier, M. L., Fagan, E., & Price, C. J. (2010). Functional subdivisions in the leftangular gyrus where the semantic system meets and diverges from the defaultnetwork. Journal of Neuroscience, 30(50), 16809–16817.

Smith, S. M. (2002). Fast robust automated brain extraction. Human Brain Mapping,17, 143–155.

Smith, S. M., Fox, P. T., Miller, K. L., Glahn, D. C., Fox, P. M., Mackay, C. E., et al. (2009).Correspondence of the brain’s functional architecture during activation andrest. Proceedings of the National Academy of Sciences of the United States ofAmerica, 106, 13040–13045.

F. Geranmayeh et al. / Brain & Language 121 (2012) 47–57 57

Thompson-Schill, S. L., D’Esposito, M., & Kan, I. P. (1999). Effects of repetition andcompetition on activity in left prefrontal cortex during word generation.Neuron, 23, 513–522.

Tremblay, P., & Gracco, V. L. (2009). Contribution of the pre-SMA to the productionof words and non-speech oral motor gestures, as revealed by repetitivetranscranial magnetic stimulation (rTMS). Brain Research, 1268, 112–124.

Tremblay, P., & Gracco, V. L. (2010). On the selection of words and oral motorresponses: Evidence of a response-independent fronto-parietal network. Cortex,46(1), 15–28.

Tremblay, P., & Small, S. L. (2011). Motor response selection in overt sentenceproduction: A functional MRI study. Frontiers in Psychology, 2, 253. doi:10.3389/fpsyg.2011.00253.

Tyler, L. K., Shafto, M. A., Randall, B., Wright, P., Marslen-Wilson, W. D., &Stamatakis, E. A. (2010). Preserving syntactic processing across the adult lifespan: The modulation of the frontotemporal language system in the context ofage-related atrophy. Cerebral Cortex, 20, 352–364.

Ullsperger, M., Harsay, H. A., Wessel, J. R., & Ridderinkhof, K. R. (2010). Consciousperception of errors and its relation to the anterior insula. Brain Structure andFunction, 214, 629–643.

Van de Ven, V., Esposito, F., & Christoffels, I. K. (2009). Neural network of speechmonitoring overlaps with overt speech production and comprehensionnetworks: A sequential spatial and temporal ICA study. Neuroimage, 47,1982–1991.

Vigneau, M., Beaucousin, V., Herve, P. Y., Duffau, H., Crivello, F., Houde, O., et al.(2006). Meta-analyzing left hemisphere language areas: Phonology, semantics,and sentence processing. Neuroimage, 30, 1414–1432.

Wilson, M. (1988). MRC psycholinguistic database: Machine-usable dictionary,version 2.00. Behavioral Research Methods Instruments and Computers, 20, 6–10.

Wise, R. G., Ide, K., Poulin, M. J., & Tracey, I. (2004). Resting fluctuations in arterialcarbon dioxide induce significant low frequency variations in BOLD signal.Neuroimage, 21(4), 1652–1664.