Reading the reading brain: A new meta-analysis of functional imaging data on reading

Journal of Neurolinguistics 26 (2013) 214–238

Contents lists available at SciVerse ScienceDirect

Journal of Neurolinguisticsjournal homepage: www.elsevier .com/locate/

jneurol ing

Reading the reading brain: A new meta-analysisof functional imaging data on reading

Isabella Cattinelli a,b,*, N. Alberto Borghese a, Marcello Gallucci b,Eraldo Paulesu b,c

aDepartment of Computer Science, Università degli Studi di Milano, 20135 Milan, Italyb Psychology Department, Università degli Studi di Milano-Bicocca, 20126 Milan, Italyc IRCCS Galeazzi, Milan, Italy

a r t i c l e i n f o

Article history:Received 20 January 2012Received in revised form 1 August 2012Accepted 6 August 2012

Keywords:Meta-analysisReading neural circuitsfMRIPETClusteringWordsPseudowords

* Corresponding author. Present address: FresenE-mail addresses: [email protected] (I. Catti

0911-6044/$ – see front matter � 2012 Elsevier Lthttp://dx.doi.org/10.1016/j.jneuroling.2012.08.001

a b s t r a c t

Over the last 20 years, reading has been the focus of much researchusing functional imaging. A formal assessment of the implications ofthiswork for amore general understandingof reading processes is stilllacking. We performed a new meta-analysis based on an optimizedhierarchical clustering algorithm which automatically groups activa-tionpeaks into clusters; the functional role of the clusterswas assessedon the basis of statistical criteria. We considered the literature from1992 to 2008, focussing exclusively on experiments based on singlewords or pseudowords from the following four classes of tasks:reading, lexical decision, phonological decision and semantic tasks.Our analysis was restricted to alphabetic orthographies andwas basedon 35 studies.We identified threenetworks: (1) a difficultymodulatednetwork includingBroca’s area andattention-relatedbrain regions; (2)a word-related network, primarily involving regions of the lefttemporal lobeandof theanterior fusiformregion, knowntoparticipateto semantic processes; (3) a pseudoword-related network in the basaloccipito-temporal regions and in the left inferior parietal cortex. Thesesubnetworks constitute the basis upon which a plausible functionalmodel of reading is proposed, where orthographic, phonological, andsemantic processes are recruited to compute the phonology ofawritten stimulus based on cooperative and competitivemechanisms.The results of this meta-analysis held face validity when comparedwith the results of literature published until mid 2010, the time ofcompletion of data collection.

� 2012 Elsevier Ltd. All rights reserved.

ius Medical Care, Via Crema 8, 26020 Palazzo Pignano (CR), Italy.nelli), [email protected] (N.A. Borghese).

d. All rights reserved.

mailto:[email protected]

mailto:[email protected]

www.sciencedirect.com/science/journal/09116044

http://www.elsevier.com/locate/jneuroling

http://www.elsevier.com/locate/jneuroling

http://dx.doi.org/10.1016/j.jneuroling.2012.08.001


I. Cattinelli et al. / Journal of Neurolinguistics 26 (2013) 214–238 215

1. Introduction

The processes involved in single word reading, i.e., in the mapping of an orthographic represen-tation to its phonological form, have been studied extensively, both with behavioural, modelling, andneuroimaging experiments.

In particular, the introduction of neuroimaging techniques and their extensive application instudying the reading processes have allowed a general picture of the main neuroanatomic componentsto be drawn. These are the components of the “reading network”, that is a set of regions in the lefthemisphere encompassing the temporal, parietal, and frontal lobe that are found to be consistentlyactivated during reading tasks. However, there is still considerable disagreement on the precise role ofthese areas, and how they interact to produce skilful reading. For instance, in recent years specialattention has been given to the posterior part of the left midfusiform gyrus as specifically responsiblefor word recognition, hence the name of Visual Word Form Area (Cohen et al., 2000, 2002). However,subsequent studies have challenged this interpretation by showing that this area is not exclusivelyactivated when reading words, but also during non-orthographic tasks such as picture naming orauditory word repetition (for a review, see Price & Devlin, 2003). Similarly, the angular gyrus has beenlinked to semantic access, but has also been associated with word form recognition (Price, 2000).Although virtually everyone can easily agree on the fact that skilful reading requires the interplay ofsemantic, orthographic, and phonological processes, an undisputed consensus onwhich regions are tobe recognized as the neural substrate of any of these subnetworks is still far from being achieved.

To try and clarify these points of disagreement, one can reconsider the published data to assesswhich sets of results were strong enough to be replicated and to represent genuine new knowledge inthe area. To this end, a number of excellent review articles have been published (Fiez, 1997; Fiez &Petersen, 1998; Petersen & Fiez, 1993; Price, 2000; Price et al., 2003; Price & Mechelli, 2005; Pughet al., 2000). However, these usually lack a formal assessment on the replicability of the previouslypublished data, leaving a good deal of uncertainty when one faces contradictory findings for verysimilar experimental protocols and techniques.

One possible alternative to qualitative reviews is provided by formal meta-analyses, as theirquantitative-oriented approach makes them more rigorous and less prone to subjective bias. Withimaging datasets of the size of that on reading, meta-analysis has the potential of “separating thewheatfrom the chaff” in distinguishing solid observations from more ephemeral ones, and may be to permitnew insights into the processes subserved by a given brain region. In brain imaging, meta-analyses aregenerally used to identify groups of activation peaks that fall sufficiently close in stereotactic space tobe interpreted as reflecting a common functional-anatomical entity. The functional significance of anyof these entities then needs to be scrutinized, on the basis of the background information about theexperiments that generated the activation peaks constituting them.

The practice of performing meta-analysis in the field of neuroimaging is becoming more and morecommon. Several studies, differing in the specific meta-analytic technique employed and the inves-tigated cognitive domain, have appeared in the literature in recent years; just for the reading domainonly, the meta-analytic approach has already been taken in various published works. In Turkeltaub(2002) 11 PET studies on aloud single word reading were taken as the testbed for demonstrating theActivation Likelihood Estimate (ALE) method for meta-analysis. Each activation peak of the dataset wasmodelled as a 3D Gaussian distribution, resulting in a statistical map (an ALE map) where each voxelwas assigned the probability of containing at least one point in the activation dataset. The statisticallysignificant voxels in the map therefore corresponded to the most consistent activation sites across theconsidered neuroimaging studies: that is, the bilateral motor cortex, pre-SMA, superior temporal sulci,bilateral cerebellum, and left fusiform gyrus. The ALE approach, and a similar method named Aggre-gated Gaussian-Estimated Sources (AGES – Chein, Fissell, Jacobs, and Fiez (2002)), were used in twometa-analyses on reading, comparing alphabetic systems to logographic ones. In Bolger, Ca, andSchneider (2005), consistent foci of activation for alphabetic systems (eastern orthographic systemsare outside the scope of the present work, and therefore we will not discuss those results here) werefound in the left ventral occipito-temporal region, left superior posterior temporal area, left inferiorfrontal gyrus, and left insula/premotor cortex. Tan, Laird, Li, and Fox (2005) focused on phonological

I. Cattinelli et al. / Journal of Neurolinguistics 26 (2013) 214–238216

tasks, and found areas consistently involved in the phonological processing of alphabetic words in theleft inferior frontal gyrus (both lateral and medial), left supramarginal gyrus, left mid-superiortemporal area, left fusiform gyrus, right superior temporal gyrus and mid-inferior occipital gyrus, andin the cerebellum.

A different approach was taken in the meta-analysis by Jobard, Crivello, and Tzourio-Mazoyer(2003), who analysed 35 fMRI and PET studies, for a total of 622 activation coordinates, by means ofa hierarchical clustering procedure. Clusters of activation peaks were automatically obtained; then,based on the characteristics of the experiments that generated the peaks inside one given cluster,a plausible functional role for that cluster was advanced. Based on such procedure, the authorsconcluded that the available data on reading support the existence of two distinct networks in the lefthemisphere: one subserving a sub-lexical grapheme-to-phoneme routine (posterior and mid superiortemporal gyrus, mid part of the middle temporal gyrus, supramarginal gyrus, and pars opercularis ofthe inferior frontal gyrus); the other being related to semantic access (inferior temporal gyrus,posterior middle temporal gyrus, and pars triangularis of the inferior frontal gyrus). Differently fromthe other mentioned works, the meta-analysis by Jobard et al. focused on discriminating among brainregions involved in different reading sub-processes, rather than just identifying areas that areconsistently activated when reading regardless of finer distinctions. Such approach potentially allowsfor a clearer understanding of the actual neural implementation of the reading ability, by suggestingwhich sub-networks can be discerned within the general reading network.

However, Jobard et al. (2003) paper has a number of limitations that invited us to reconsider theissue of a meta-analysis on reading. First of all, the dataset they collected was rather heterogeneous(for instance, data from both alphabetic and non-alphabetic orthographies were considered), whichmight have confounded the results. Moreover, no statistical testing was used to support the advancedqualitative interpretation of the obtained activation clusters, and many of these clusters were notanalysed at all. Finally, a number of technical issues, including an inherent limitation of hierarchicalcluster (Cattinelli, Valentini, Paulesu, & Borghese, Submitted for publication) (i.e., the non-uniquenessof the solution when the input data order is changed), precluded an exact replication of Jobard’sresults.

Even if we deemed these limitations to be minor, several years have passed since the publication ofJobard et al. (2003): new imaging studies on the reading domain have appeared ever since and, still, theorganization of the reading network in the human brain is matter of lively debate. For this reason, wehave decided to perform a new meta-analysis of the neuroimaging data available on single wordreading, with the aim of clarifying the role of the neuro-anatomical regions involved in this process. Inparticular, by reassessing patterns of neural activation associated with word and pseudoword reading,we intended to investigate whether a given neural region is most likely involved in orthographic,semantic, or phonological processing, or rather in integrating all these different aspects, possibly ina way that is modulated by the difficulty, across different psycholinguistic variables, of the material tobe read.

Our meta-analysis was able to isolate both a pseudoword- and aword-specific network: our resultssuggest that the former may be implied in integrating orthographic and phonological information,whereas the latter may implement semantic processes. A set of difficulty-sensitive regions was alsoidentified, for which we advance a role in determining the final pronunciation of the written stimulus,based on possibly conflicting information provided by previous processing stages.

2. Materials and methods

2.1. Data collection

This meta-analysis considers the neuroimaging literature on reading starting from the earliestpioneering studies (late ’80s) up to 2008. Neuroimaging experiments present a large amount of vari-ability as far as the imaging methodology, experimental design, nature of the stimuli, stimulationprocedures, task demands, which may make the resulting activation data hardly comparable. Wetherefore determined a list of inclusion criteria which had to be met for a study, or a specific resultderived from a given statistical comparison, to be included in this meta-analysis:


- Subjects: we considered only studies including at least six normal adult subjects; no data fromneuropsychological patients were used.

- Scanning technique: fMRI or PET.- Stimuli: single words only (e.g. no word pairs, or sentences).- Alphabetic orthographies only were considered.- Data analysis: only whole-brain analyses were considered (i.e., no ROI approaches).1 p-values werealso checked: for simple comparisons (i.e., task > baseline), activation data were included in ourdatabase only for p-values at least <0.001 (uncorrected for multiple comparisons); for directcomparisons (e.g. words > pseudowords), we accepted a p < 0.01 (uncorrected) threshold. Inaddition, we considered only univariate analyses.

- Tasks: only four classes of tasks were considered: reading (aloud or silently, explicitly or implicitly),lexical decision, phonological decision, and semantic tasks. Other tasks such as priming, and othertasks involving the processing of pairs or triplets of stimuli, were not considered, as we focused onsingle word processing only.

- Contrasts: To maximize the relevance and specificity of the data for the identification of specificprocesses involved in reading, only the results of certain kinds of statistical comparisons wereconsidered: for example, we excluded conjunctions (e.g. Reading and Lexical Decision > Baseline),comparisons among tasks rather than classes of stimuli (e.g. Reading > Lexical Decision), and maineffects (e.g. Readingwords and pseudowords> Baseline). As a guiding principle, we gave preferenceto themost specific comparisons: for instance, when a paper reported activation data from both thecontrast main effect of Words (regular & irregular) > Baseline, and the contrast IrregularWords> Baseline or RegularWords> Baseline, only data from the simple effect contrastswere used.

By applying such criteria, we collected a total of 1176 activation peaks, coming from 35 neuro-imaging studies. This is the same number of studies as in Jobard et al. (2003) as a few studies wereexcluded following the application of the above criteria but, on the other hand, additional and morerecent ones were included in our meta-analysis. As a result our final working dataset was made ofalmost twice the number of activation peaks as in Jobard et al. (2003), at the same time showinga higher degree of homogeneity. Table 1 lists the 35 studies on which this meta-analysis was per-formed, along with all relevant information.

For each activation peak, we recorded all relevant information about the statistical comparison thatgenerated it. In particular, each peak was classified according to the following variables of interest: lexi-cality (wordornonword); regularity; frequency (also includingorthographicneighbourhood size); syllabiclength; type of orthography (deep or shallow).We focused on these specific variables as they are known tomodulate naming latencies and thus arguably the difficulty indecoding an orthographic stimulus (Ferrand&New,2003; Forster&Chambers,1973;Glushko,1979;McCann&Besner,1987;Paap&Noel,1991;Paulesuet al., 2000; Seidenberg, Waters, Barnes, & Tanenhaus, 1984; Taraban & McClelland, 1987).

Of course, it would have been impossible to take full account of these variables and to evaluate theirimpact on the data in a formal statistical manner: however, this information proved useful fora qualitative assessment of the possible functional meaning of a cluster.

2.2. Template normalization

Most of the latest papers reporting activation data from neuroimaging experiments adopta stereotactic normalization to the MNI (Montreal Neurological Institute) template. However, some

1 The whole-brainess of the datasets included for the meta-analysis was assumed on the grounds of the description given inthe papers considered. Incomplete sampling of the brain can cause interpretation problems for clusters at the border of the fieldof view if, for example, one class of task was the driving force while coming from studies with a more complete sampling of thebrain. This matter is made more complex by the fact that the usable field of view of the data (the anatomy effectively shared byall subjects as assessed after stereotactic normalization) is generally not carefully described in the methods sections. However,direct inspection of certain clusters, like for example the ones in the supplementary motor region, including the clusterassociated with silent reading (see the results section), shows that there was a good mix of PET and fMRI studies, the PETstudies of the mid 90s being by definition whole brain studies and the fMRI studies being from the last decade.

Table 1Studies included in this meta-analysis.

Reference Technique Subjects Design

(Beauregard et al., 1997) PET 10 men Blocked(Binder et al., 2003) fMRI 15 women, 9 men Event-related(Binder et al., 2005) fMRI 12 women, 12 men Event-related(Bookheimer, Zeffiro, Blaxton,

Gaillard, & Theodore, 1995)PET 8 men, 8 women Blocked

(Buchel, Price, & Friston, 1998) PET 6 Blocked(Carreiras, Mechelli, & Price, 2006) fMRI 10 women, 6 men Blocked(Chee, O’Craven, Bergida, Rosen,

& Savoy, 1999)fMRI 5 men, 3 women

(but one subject excluded)Blocked

(Cohen et al., 2002) fMRI 6 women, 1 man (first exp.) Blocked (first exp.)(Fiebach, Friederici, Muller,

& von Cramon, 2002)fMRI 5 women, 7 men Event-related

(Fiebach, Ricker, Friederici,& Jacobs, 2007)

fMRI 8 women, 8 men Event-related

(Fiez et al., 1999) PET 5 women, 6 men Blocked(Gates & Yoon, 2005) fMRI 9 Blocked(Hagoort et al., 1999) PET 3 women, 8 men Blocked(Herbster, Mintun, Nebes,

& Becker, 1997)PET 5 men, 5 women Blocked

(Howard et al., 1992) PET 7 men, 5 women Blocked(Jessen et al., 1999) fMRI 5 women, 7 men Event-related(Joubert et al., 2004) fMRI 10 men Blocked(Kiehl et al., 1999) fMRI 6 men Blocked(Kuchinke et al., 2005) fMRI 12 women, 8 men Event-related(Mechelli, Friston, & Price, 2000) fMRI 1 woman, 5 men Blocked(Mechelli, Gorno-Tempini,

& Price, 2003)fMRI 7, 13 (2 exp. varying

stimulus duration or rate)Blocked

(Mechelli et al., 2005) fMRI 10 women, 12 men Blocked(Menard, Kosslyn, Thompson,

Alpert, & Rauch, 1996)PET 8 men Blocked

(Meschyan & Hernandez, 2006) fMRI 7 women, 5 men Blocked(Moore & Price, 1999) PET 8 men Blocked(Paulesu et al., 2000) PET 6 Italian, 6 English Blocked(Perani et al., 1999) PET 14 men Blocked(Price et al., 1994) PET 6 men Blocked(Price, Moore, & Frackowiak, 1996) PET 6 men Blocked(Price, Wise, & Frackowiak, 1996) PET From 6 to 10 Blocked(Price et al., 2006) PET 18 men Blocked(Rumsey et al., 1997) PET 14 men Blocked(Valdois et al., 2006) fMRI 12 (reading), 8

(lexical decision)Event-related

(Vigneau, Jobard, Mazoyer,& Tzourio-Mazoyer, 2005)

fMRI 23 (simple comparisons),13 (direct comp.)

Blocked

(Vinckier et al., 2007) fMRI 8 women, 4 men Blocked

Studies whose activation data have been included in the present meta-analysis. For each of them, a bibliographic reference forthe paper describing the study, the employed neuroimaging technique, details about recruited subjects, and the type ofexperimental design are reported.


studies, especially the oldest ones, adopted a stereotactic normalization based on the Talairach andTournoux (1988) space. Therefore, a transformation was needed in order to make activation datafrom all considered studies comparable.

After determining which template was used in each paper (as explicitly indicated in the text, orinferred on the basis of the software used for the statistical analysis2), we converted coordinatesoriginally reported in Talairach space to MNI space, by using the tal2mni MATLAB script (http://imaging.mrc-cbu.cam.ac.uk/imaging/MniTalairach). Our final working dataset, therefore, involvedexclusively activation coordinates in MNI space.

2 This can be easily inferred for SPM, as the MNI space has been adopted since the SPM96 release.

http://imaging.mrc-cbu.cam.ac.uk/imaging/MniTalairach

http://imaging.mrc-cbu.cam.ac.uk/imaging/MniTalairach


2.3. Clustering procedure

The 1176 stereotactic coordinates of our database were submitted to a clustering algorithm, in orderto automatically group peaks close in space into plausible summary regions of activation. As in Jobardet al. (2003), we adopted the hierarchical clustering approach (Xu & Wunsch, 2005) with Ward’scriterion (Ward, 1963).

In hierarchical clustering, each input data point is initially assigned to a singleton cluster. At eachprocessing step, the two closest (according to the chosen dissimilarity measure) clusters are mergedinto one. Thus, at each step the number of clusters is decreased by one, until the last step is reachedand, eventually, only one cluster, including all input data, is obtained. For each step, the existingclusters represent an admissible grouping of input data: what changes among different steps, is just thenumber of clusters. This hierarchical process is graphically represented as a dendrogram, or tree (Xu &Wunsch, 2005).

Many criteria may be used to determine which two clusters are to be merged at each step: here,Ward’s criterionwas adopted, whichminimizes the increase in the total intra-cluster variance resultingfrom merging a pair of clusters. The final clusters are obtained cutting the dendrogram at some level,through an adequate threshold. Here, similarly to Jobard et al. (2003), we adopted a spatial criterionbased on an estimate of the average spatial resolution of imaging experiments: accordingly, the set ofclusters is characterized by an average standard deviation on the three directions not greater than7.5 mm. Therefore, starting from the leaves of the dendrogram, our algorithm climbed up the tree andstopped just before the threshold was exceeded.

Even though hierarchical clustering is a standard and widely-used paradigm, it nonetheless suffersfrom the problem of non-uniqueness of the solution: it may produce different solutions depending onthe order of the input data (Morgan & Ray, 1995). We therefore developed a variant of the originalalgorithm that guarantees the uniqueness of the clustering solution, independent of the data inputorder. Thanks to the definition of an equivalence relation on the dendrograms, only a small subset of allthe possible dendrograms that can be obtained starting from a given dataset is explored (Cattinelliet al., Submitted for publication). The explored dendrograms are then cut at the same level (byusing the above criterion based on average standard deviation), and the associated clustering solutionsare compared to identify the best one according to a quality criterion. For this aim, we adopteda measure of between-cluster error sum of squares (B-ESS) defined as

B-ESS ¼XjCj

k¼1

nkðmk � mXÞ2

where jCj is the number of cluster in the considered solution, nk is the number of elements in cluster k,mk is the mean of cluster k, and mx is the mean of the entire dataset. The best clustering solution is heredefined to be the one having maximum B-ESS (that is to say, clusters are well-separated in the outputspace). Our clustering procedure identified four different clustering solutions (having B-ESS values of2.4023 � 106, 2.3977 � 106, 2.4017 � 106, and 2.3980 � 106). The final solution, corresponding to B-ESS¼ 2.4023�106, consisted of a set of 57 clusters, having average standard deviation of 6.98, 7.34, and7.11 mm, respectively on the x, y, and z direction.

2.4. Anatomical labelling

Each cluster was then automatically assigned to an anatomical label based on the AAL (AutomatedAnatomical Labelling – Tzourio-Mazoyer et al., 2002) and Brodmann templates available under MRICro(Rorden & Brett, 2000). The mean coordinate of each cluster was mapped on these templates to extractthe anatomical region to which the coordinate belongs: this was achieved in a completely automaticway through a MATLAB script that was developed to this aim. Only in those cases when the automaticpolling of the templates returned a void label3 (i.e., when the given coordinate falls into a non-mapped

3 The automatic polling procedure on the AAL template returned void labels for 4 out of 57 clusters.


brain area) the templates were manually inspected in order to find the labelled anatomical regionclosest to the peak in hand.

2.5. Anatomical consistency of clusters

Before proceeding to any further assessment of the composition of each cluster, clusters thatincluded stereotactic peaks of both the occipito-temporal region and the cerebellum were manuallysplit, because a cluster including both kinds of coordinates was deemed to be anatomicallyimplausible.

2.6. Functional interpretation of clusters

The assessment of the functional relevance of a given cluster was performed with a preliminaryqualitative assessment followed by a quantitative statistical evaluation using a binomial test.

The qualitative exploration of the datawasmade in order to identify any trend along twomain axes:lexicality and difficulty4 of the materials. For example, a given cluster may have shown a clear trend forbeing composed of stereotactic coordinates emerging from comparisons of reading words as opposedto reading pseudowords. One such cluster would then be tested for statistical significance along thelexicality axis. On the other hand, from another cluster, a different pattern might have emerged witha substantial contribution of stereotactic coordinates derived from the comparison of pseudowordsversus real word reading and irregular words versus regular words reading: such cluster would bequalitatively labelled as a brain region sensitive to the difficulty of the materials and then subjected toa binomial test along the difficulty axis in order to check for statistical significance.

Statistical testing was performed to assess the significance of the disproportion between coordi-nates belonging to two different categories (e.g. word-related peaks versus pseudoword-related peaks)within each cluster. If such disproportion were due to chance only, we would expect that the distri-bution of peaks from each category within the considered cluster would mirror the overall distributionin the dataset. Therefore, for testing the significance of a cluster along the lexicality axis, we firstselected those contrasts that are word-specific (e.g. word > baseline, word > pseudoword), and thosethat are pseudoword-specific (e.g. pseudoword > baseline, pseudoword > word). We then computedthe total number of activation peaks belonging to the two categories (words ¼ 724 (w68%),pseudowords¼ 346 (w32%)). Further, if a cluster of k peaks, containing nword-related peaks was beingtested for being word-specific, we applied the binomial test to assess the probability to observe at leastn successes out of k trials under the null hypothesis (p ¼ 0.68). If the cluster was instead tested for itspseudoword-specificity, the null hypothesis would have been modelled by a binomial distributionhaving p¼ 0.32. In both cases, when the p-value returned by the procedure was less or equal than 0.05,the null hypothesis was rejected.

We proceeded similarly when testing for difficulty; in this case, contrasts were divided betweenthose contrasting a more difficult item to an easier one, and the vice versa (for the difficulty class wehad a p ¼ 0.40; we did not test in the reversed direction).

A binomial test was used because in each comparison two classes only were considered, andbecause, being an exact test, the possible scarce cardinality of a cluster would not hamper thesignificance of the result (as it would be the case with chi-squared test, that imposes specificrequirements over the cardinality of the groups being tested). For all tests, a p-value less or equal than0.05 was considered to be the threshold for significance.

4 A “difficulty-modulated” label was attached during the qualitative exploration of the data to a given cluster when noobvious psycholinguisitic label could be attributed to the cluster and yet, taken together, the overall direction of the effects waspointing towards a difficulty factor. For example, cluster #45, the left opercular inferior frontal region, on top of being identifiedby some peaks derived from trivial reading > baseline contrast, was identified by a mix of comparison such as pseudo-words > words; low frequency words > high-frequency words but also irregular words > regular words. These are allcomparisons in which a longer RTs may be recorded for the former condition of each pair to testify a greater difficulty while noobvious shared psycholinguistic tag can be attached according to the mere lexical status of the stimuli or the experimentalconditions. The qualitative exploration of the data was then followed by the binomial tests.


In what follows, for illustration purposes, we also report clusters that passed only a qualitativeassignment (as described earlier in this section) to one of the three classes anticipated (word-specific,pseudoword-specific, difficulty-sensitive). All other clusters were assigned to a class called non-differentiated. Among these clusters, we attempted to highlight those having higher probability ofactually being completely aspecific, by performing binomial tests along either the lexicality or thedifficulty axis, yielding three p-values (one for word-sensitivity, one for pseudoword-sensitivity, andone for difficulty-sensitivity) for each cluster. We propose that clusters whose minimum p-value wasstill greater than 0.4 are of high chance of being genuinely non-differentiated.

2.7. Recent papers

The process of performing a meta-analysis in a lively area of research such as the one of reading hasthe inevitable drawback that by the time that the analysis is completed and the ensuing paper iswritten, new data have appeared in the literature.

We turn this potential issue into an opportunity by reasoning that a valid meta-analysis should holdpredictive power when faced with subsequent studies. Indeed, a comparison with newer findingsshould offer a good testbed for themeta-analysis itself. Therefore, in discussing the results of our meta-analysis wewill also mention related findings that have been reported in the literature after the secondhalf of 2008, time of completion of our data collection procedure. These additional studies wereselected on the basis of the same inclusion principles that guided our meta-analysis and with a pref-erence for those studies investigating specific sub-processes of reading. Results on these new data haveconfirmed the predictions made by the meta-analysis.

3. Results

The procedures described left us with a final number of 64 clusters. Tables S-I and S-II in theSupplementary Materials describe these 64 clusters in detail.

As explained in the Materials and Methods section, we first performed a qualitative analysis as inJobard et al. (2003). This lead us to label each cluster as (a) sensitive to item difficulty, (b) preferentiallyinvolved in word processing, or (c) preferentially involved in pseudoword processing. When none ofthese categories applied, the cluster was considered to be (d) non-differentiated. In what follows, wewill describe these four categories of clusters; for each category, we will also indicate which clusterssurvived the binomial testing for significance.

A list of clusters for each category can be found in Tables 2–5; statistically significant clusters aremarked by an asterisk. Figs. 1–4 graphically show the location of the clusters within each category:a cluster is represented as an ellipsoidal blob, centred on the mean coordinate of the cluster; thestandard deviation of the cluster determines the length of the semiaxes of the ellipsoid. Darker blobsindicate those clusters for which the binomial test was significant. A spreadsheet reporting detailedinformation about the composition of each cluster is available as part of the Supplementary Materials.

3.1. Difficulty-modulated network

Fig. 1 illustrates a network modulated by task difficulty (see also Table 2). Within this cortical–subcortical network composed of 16 clusters, four clusters survived statistical assessment by thebinomial test: the left inferior frontal gyrus (pars opercularis), the left superior parietal lobule, the rightmid cingulum and the pons/cerebellum.

3.2. Word-related network

Clusters that showed a preferential involvement in word reading are listed in Table 3 and shown inFig. 2. These were mostly left-lateralized in a largely distributed network, including frontal, parietal,temporal, and occipital regions.

Over a total of 13 clusters, 8 survived statistical testing; all of them are located in the left hemi-sphere. Two clusters fall in the anterior cingulum; two clusters are located in the parietal cortex

Table 2Difficulty-modulated clusters.

Anatomical label Mean X Mean Y Mean Z AvgStd Num Cluster ID Anatomical label Mean X Mean Y Mean Z AvgStd Num Cluster ID

Frontal_Inf_Tri_L_45 �46 31 14 5 17 15 Frontal_Inf_Tri_R_45 43 30 28 10 20 53Frontal_Inf_Tri_R_45 49 19 5 7 21 27

Frontal_Inf_Oper_L �35 15 16 6 29 16Frontal_Inf_Oper_L_6* �51 9 14 6 43 45Supp_Motor_Area_L_6 0 �1 60 5 30 46Precentral_L_6 �49 �2 31 6 40 26Postcentral_L_6 �47 �9 45 6 29 25Parietal_Sup_L_7* �25 �66 51 7 24 49Parietal_Inf_L_40 �49 �40 47 7 21 32

Cingulum_Mid_R_24* 2 8 41 8 38 47Temporal_Sup_L_22 �56 �16 3 7 29 20Thalamus_L �16 �13 6 10 31 57 Thalamus_R 18 �21 6 10 18 44

Putamen_R 29 14 4 7 20 28Pons/Cerebellum_R* 11 �28 �25 8 8 14

Clusters exhibiting a difficulty effect, as resulted by a qualitative assessment of their composition. Clusters for which the binomial test was significant (p � 0.05) are reported in bold andmarked with a *. For each cluster, its anatomical label is reported, together with its mean coordinate, average standard deviation (averaged on the three directions), and number of includedactivation peaks.

I.Cattinelliet

al./Journal

ofNeurolinguistics

26(2013)

214–238

222

Table 3Word-related clusters.


Frontal_Sup_L_9 �15 45 36 6 10 33Frontal_Mid_R_9 35 12 53 6 8 40

Cingulum_Ant_L_10* �9 48 0 10 13 12Cingulum_Ant_L_11* �6 27 �7 9 19 11Angular_L_39* �44 �75 31 6 16 4

Cingulum_Mid_R_23 3 �38 38 7 14 2Precuneus_L_23* �3 �57 30 6 13 1Temporal_Mid_L_21* �51 �51 17 8 28 24Temporal_Mid_L_21* �57 �45 �1 6 19 23Temporal_Mid_L_20* �61 �18 �21 5 9 19Fusiform_L_37* �28 �39 �12 6 12 37a

Occipital_Mid_R_39 38 �69 25 8 15 36Calcarine_R_17 6 �96 6 9 18 30

Clusters showing specificity for word processing, based on our first qualitative analysis.Wemarkedwith a * and reported in bold are those clusters for which the effect was confirmed by theresult of a binomial test.

I.Cattinelliet

al./Journal

ofNeurolinguistics

26(2013)

214–238

223

Table 4Pseudoword-related clusters.


Angular_R_7* 30 �62 49 7 28 6Angular_L_40* �35 �49 34 6 21 31Temporal_Inf_L_20 �48 �46 �19 6 17 22aFusiform_L_19 �36 �80 �12 4 5 52aOccipital_Inf_L_18 �18 �93 �11 7 19 17Occipital_Inf_L_37 �41 �64 �11 4 26 21

Occipital_Mid_R_19 36 �82 2 5 13 35Occipital_Inf_R_19 46 �69 �12 5 10 8aCerebelum_Crus1_R 39 �70 �27 5 7 8b

Clusters specifically involved in pseudoword reading, according to our preliminary qualitative inspection. Clusters that survived statistical testing are shown in bold and marked with a *.

I.Cattinelliet

al./Journal

ofNeurolinguistics

26(2013)

214–238

224

Table 5Non-differentiated clusters.


Frontal_Mid_L_6 �29 7 51 9 26 54Frontal_Mid_Orb_R_47 34 48 �3 8 6 42

Frontal_Inf_Orb_L_47* �38 26 �6 8 37 56Supp_Motor_Area_L_8 �3 26 51 6 19 34

Postcentral_R_4 42 �18 45 8 23 43Postcentral_R_4 50 �3 32 8 26 41Parietal_Inf_R_40* 45 �50 43 6 12 5Heschl_R 56 �8 8 6 15 50Temporal_Mid_R_22 55 �37 4 8 29 51Fusiform_R_37 39 �53 �16 6 12 7a

Fusiform_L_20 �33 �25 �25 10 21 38Hippocampus_R* 35 �7 �16 9 7 13

Occipital_Mid_L_19 �23 �73 27 5 14 3Occipital_Mid_L_18* �29 �89 8 7 18 18 Occipital_Mid_R_19 36 �82 2 6 13 35

Cuneus_R_19 12 �85 30 7 10 29Lingual_R_17 3 �66 10 8 19 48

Lingual_L_18 �17 �73 �7 5 4 9a Lingual_R_18 25 �91 �12 3 11 55aCerebelum_6_L �12 �64 �19 5 8 9b Cerebelum_6_R 8 �73 �17 6 17 10Cerebelum_6_L* �21 �49 �27 6 10 37b Cerebelum_6_R 28 �52 �22 6 12 7bCerebelum_Crus1_L �31 �80 �23 7 22 52b Cerebelum_Crus1_R* 21 �85 �25 6 11 55bCerebelum_Crus1_L* �42 �57 �28 4 8 22b

Vermis_8* 5 �64 �35 10 21 39

Clusters that were not included in any of the previous categories (i.e., difficulty, word, or pseudoword) were considered to be non-differentiated. Among these, we report in bold, andmarkedby a *, those clusters for which this aspecificity is more likely to be statistically significant.

I.Cattinelliet

al./Journal

ofNeurolinguistics

26(2013)

214–238

225

Fig. 1. Clusters labelled as showing a difficulty effect. In darker blue, clusters surviving also statistical testing. (For interpretation ofthe references to colour in this figure legend, the reader is referred to the web version of this article.)


(respectively, in the posterior part of the angular gyrus and in the precuneus); three clusters can befound in the middle temporal gyrus; and one final cluster is in the anterior part of the fusiform gyrus.

3.3. Pseudoword-related network

Eight clusters were classified as being preferentially involved in pseudoword reading (see Table 4and Fig. 3). The pseudoword network encompasses bilateral parietal areas, mostly left-lateralizedtemporo-occipital regions, and the right cerebellum.

Two individual clusters survived a direct statistical assessment. These are located in the rightangular/superior parietal gyrus and in the left inferior parietal lobule, in the supramarginal gyrus.Because of its well-known involvement in orthographic processing, the set of the four left ventraloccipito-temporal clusters was also analysed collectively, yielding a significant pseudoword effect; thisjoint analysis was equivalent to an ad-hoc change in spatial resolution.

Fig. 2. Clusters specifically involved in word reading. Clusters for which the binomial test reported a significant p-value are shown inorange. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 3. Clusters showing a specificity for pseudoword processing. Among these clusters, those having a statistically significant effectare depicted in purple. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version ofthis article.)


3.4. Non-differentiated clusters

The remaining 27 clusters, which could not be assigned to any of the previous categories, wereconsidered to be non-differentiated clusters (see Table 5 and Fig. 4). This generic network includesbilateral frontal regions, right parieto-temporal areas, bilateral fusiform gyri, the right hippocampus,bilateral occipital cortex and cerebellum. In particular, eight clusters were identified as reliably non-differentiated. Four of them are located in the left hemisphere: these are an inferior frontal gyrus(pars orbitalis) cluster, a middle occipital gyrus, and two clusters in the cerebellum. As for the righthemisphere, we have an inferior parietal cluster, a hippocampus cluster, and two cerebellum clusters.

3.5. Task-specific clusters

Wealso assessed the possibility that some of the aforementioned clusters contained also task-specificeffects: this information is given in Table 6 (Task specific clusters) where we provide the indication of

Fig. 4. Clusters in the non-differentiated category. Those clusters that are more likely to be completely non-differentiated in theirprocessing are shown in blue. (For interpretation of the references to colour in this figure legend, the reader is referred to the webversion of this article.)

Table 6Task-specific clusters.


Reading aloud > silent readingPostcentral_R_4 (c) 42 �18 45 8 23 43Heschl_R (c) 56 �8 8 6 15 50

Thalamus_L_0 (a) �16 �13 6 10 31 57 Thalamus_R_0 (a) 18 �21 6 10 18 44

Silent reading > reading aloudFrontal_Sup_L_9 (b) �15 45 36 6 10 33Front_Inf_Orb_L_47 (c) �38 26 �6 8 37 56Supp_Motor_Area_L_8 (d) �3 26 51 6 19 34Temp_Mid_L_20 (b) �61 �18 �21 5 9 19Occipital_Mid_L_18 (c) �29 �89 8 7 18 18

Reading > lexical decisionHeschl_R (c) 56 �8 8 6 15 50

Occipital_Mid_L_19 (c) �23 �73 27 5 14 3Cerebelum_Crus1_L (c) �31 �80 �23 7 22 52b

Lexical decision > readingFrontal_Mid_L_6 (c) �29 7 51 9 26 54

Front_Inf_Tri_R_45 (a) 43 30 28 10 20 53Cingulum_Mid_R_23 (b) 3 �38 38 7 14 2

Parietal_Inf_L_40 (a) �49 �40 47 7 21 32Angular_L_39 (b) �44 �75 31 6 16 4Temp_Mid_L_21 (b) �57 �45 �1 6 19 23Occipital_Mid_L_18 (c) �29 �89 8 7 18 18

Clusters showing a reliable effect of task as revealed by binomial tests (p � 0.05). Areas marked with (a) are also associated with the difficulty effect, those marked with (b) with the wordeffect, those marked with (c) were non-differentiated.

I.Cattinelliet

al./Journal

ofNeurolinguistics

26(2013)

214–238

228


which cluster also belonged to oneof theeffects connectedwith the lexical status of the stimulus or to thedifficulty factor. Most of the task-related effects were in “non-differentiated” clusters, while someinteresting joint effects were within the word-related clusters (for example, cluster #4 and cluster #23,the left angular and left middle temporal gyri) for the comparison lexical-decision tasks versus reading.

4. Discussion

Our meta-analysis identified three reading systems: a network sensitive to the difficulty of thematerials, a network specifically involved in word processing and a network preferentially involved inpseudoword reading. In what follows, we will discuss each of these subnetworks separately.

In addition, we also found a set of clusters that somehow participate to reading processes, but towhich it was impossible to attach a specific functional commitment, even in amerely qualitativeway. Itis worth noting that these non-differentiated clusters do not necessarily identify regions of conjunc-tions of activations across stimuli of different psycholinguistic nature; rather, they most likely repre-sent brain regions brought into the dataset because of relatively aspecific control conditions used asa baseline for the task of reading during the imaging experiments considered.

4.1. Difficulty-modulated network

The clusters assigned to this network are characterized by a predominant number of activationpeaks coming from statistical comparisons where a more difficult item type is contrasted to an easierone; for instance, pseudoword > word (difficulty in a lexical sense), exception word > regular word(difficulty in a spelling-to-sound-consistency sense), low-frequency word > high-frequency word(difficulty in a frequency sense). Therefore, we propose that these brain regions are sensitive to thecomputational load required by the reading task in hand, rather than to any of the consideredpsycholinguistic variables per se (lexicality, or consistency, or frequency); a plausible interpretation forthis effect is, in fact, that in these regions difficult tasks induce longer processing times that, asa consequence of time integration, result in stronger BOLD signal.

Therewere four such clusters (See Fig.1): one in the left inferior frontal gyrus (pars opercularis), onein the (right) mid cingulum, one in the left superior parietal gyrus, and one in the pons.

A crucial role in reading is traditionally attributed to the left inferior frontal gyrus (pars opercularis),long considered to be part of the output system since Broca’s study (1861). In their review, Fiez andPetersen (1998) discuss the role of the left frontal operculum in transforming orthographic repre-sentations into phonological ones, and point out its sensitivity to consistency and frequency of writtenstimuli: this area appears to bemore activated by low-frequency exceptionwords and by pseudowords,rather than by regular words. This pattern is consistent with our hypothesis of a difficulty effect takingplace in the left inferior frontal gyrus. Our interpretation is also shared by others (Fiez, Balota, Raichle,& Petersen, 1999; Mechelli et al., 2005). Mechelli et al. (2005) suggest that “this region may contribute tothe effortful selection, retrieval, and/or manipulation of phonological representations critical for readingand other phonological tasks”.

Neuropsychological studies provide additional converging evidence on this point. Price et al. (2003)described a case of semantic dementia associated with surface dyslexia, investigated through a fMRIexperiment: in this patient, reading elicited a greater activation of the left opercular frontal region, ascomparedwith normal subjects; thismay suggest that, being the semantically-mediated route disruptedby neural atrophy, a surviving phonological routewas stressed in an attempt of compensation. The groupof patients describedby Fiez, Tranel, Seager-frerichs, andDamasio (2006), having a left frontal operculumlesion, showed a significantly greater impairment in reading pseudowords than regularly spelledwords;however, the same patients had also troubles in reading low-frequency irregular words. It was advancedthat a lesion centred on the left inferior frontal gyrus determines a general deficit in phonological outputprocesses, and not only in the transformation from sub-lexical orthographic to phonological represen-tations. The effects of such a deficit would be more evident as task difficulty increases.

Additional support to the existence of a difficulty effect localized in the left frontal operculum can befound also in studies that have been published after our meta-analysis was completed. In the fMRIstudy by Graves, Desai, Humphries, Seidenberg, and Binder (2010), reaction times in reading aloud


were positively correlated with activity in the left inferior frontal gyrus and anterior insula, and thesame areas also showed a negative correlationwith word frequency and consistency. These areas wereinterpreted to be involved in generic (i.e., not specific for reading) executive processes, sensitive toincrease in task load. This is compatiblewith our proposal that the left frontal operculum contributes toreading in a difficulty-modulated manner, although we also propose that its involvement might bespecific to phonological output processes, rather than just due to more general attentive, workingmemory, or executive processes. Hauk, Davis, and Pulvermüller (2008) found a negative correlationwith frequency for silent word reading in the left inferior frontal gyrus. This finding was confirmed inCarreiras, Riba, Vergara, Heldmann, andMünte (2009), where activation in frontal regions (specifically,left pre/SMA and inferior frontal gyrus/insula bilaterally) was negatively correlated with wordfrequency for a lexical decision task. In Levy et al. (2008) a precentral gyrus focus close to our leftfrontal operculum cluster was activated for the comparison pseudowords>words, which contributes toa difficulty effect (as acknowledged in this particular study, too, as pseudowords are the type ofstimulus involving the greatest processing load). Also Kronbichler et al. (2009) reported foci in theinferior frontal gyri and left precentral gyrus for pseudoword (letter-deviant versions of real words)reading as opposed to word reading. Peeva et al. (2010) explicitly interpreted their left frontal oper-culum focus of activation as a difficulty effect. Lastly, Nosarti, Mechelli, Green, and Price (2010) againidentified the left frontal operculum as part of a network involved in processing load (“when nonlexicaland lexical/semantic processing are inconsistent”).

The difficulty-sensitive network we identified also include the parietal lobe and the anteriorcingulate cortex. These have been implied in attentional processes with and without explicit eye-movements (review in Bottini & Paulesu, 2003). Neuroimaging studies consistently found attention-related activation in the superior parietal lobule, frontal eye field, and supplemental eye field(Corbetta, 1998; Kastner & Ungerleider, 2000). These findings therefore provide support to the inter-pretation of the superior parietal cluster andmid-cingulum cluster as involved in attentional processes,whose contribution is presumably modulated by task difficulty.

Lastly, the cluster whose centre is located in the dorsal part of the pons could tentatively be related totheascendingmodulatorypathwaysof the reticular formation involved inattention. Yet, the small numberof activationpeaks (eight) grouped in this cluster suggests that thisfinding should be treatedwith caution.

To summarize the results on the difficulty network, previous evidence, and our own, both supportthe hypothesis that the left inferior frontal gyrus, pars opercularis, participates in translating ortho-graphic representations into phonological ones, a process that is sensitive to difficulty. The mid-cingulum cluster and the left superior parietal cluster (and possibly the pons) might reflect atten-tional processes that are more vastly recruited for more effortful stimuli, such as exception words andpseudowords, which require more cognitive resources. It may be argued, especially for the superiorparietal region, that the difficulty of the stimulus triggers a more intense eye movement activity forextensive exploration of the orthographic string. Remarkably similar conclusions have been drawn byBinder, Medler, Desai, Conant, and Liebenthal (2005) who offer a collective interpretation of a putative“difficulty network”, as the activity within this network correlated with the reaction times for theirreading task. In their words “... in both cases” – i.e., reading pseudowords and low-frequency irregularwords – “the relative unfamiliarity of the particular correspondence makes the mapping process lessefficient, resulting in an increased load on attentional (FEF, IPS, anterior cingulate), working memory (IFG,precentral gyrus, IPS), decision (IFG, anterior cingulate gyrus, anterior insula), and response monitoring(anterior cingulate gyrus) mechanisms. Activation of these regions is therefore consistent with the expectedand observed differences in task difficulty between conditions, and not indicative of a specialized route forrule-based phonological assembly.”

4.2. Word-related network

Our meta-analysis also identified a set of left brain regions showing a significant preference for realwords as opposed to pseudowords. These areas form a left temporoparietal network including theposterior part of the angular gyrus, the precuneus, the middle temporal gyrus, and the anterior part ofthe fusiform gyrus; in addition, two clusters in the left anterior cingulum were found.


Words differ from pseudowords both because they are familiar items, and because they possesssemantic value. The orthographic (word-form) and the semantic representation of a word might becoded simultaneously, or in a separate fashion; in the latter case, we should be able to distinguish brainregions that are devoted to a pure orthographic processing, and others that perform only semanticprocessing. However, our results support the alternative hypothesis of an extended reading subnet-work involved in both semantic and lexical processing, in a non-segregate way.

In fact, the evidence from previous imaging literature collectively suggests that the word-sensitiveareas that we have identified in this work, are in fact involved in semantic processing as well. Indeed,Price and Mechelli (2005) report that the left angular gyrus, middle temporal cortex, anterior fusiformgyrus, as well as the left inferior frontal gyrus in the pars orbitalis and triangularis are consistentlyactivated in semantic tasks as opposed to phonological tasks. The panoramic description of thesemantic network given by the meta-analysis of Binder, Desai, Graves, and Conant (2009) is alsoconsistent with our findings.

Additional support may be found in the literature on reading disorders. In Brambati, Ogar, Neuhaus,Miller, and Gorno-Tempini (2009) a VBM (voxel-based morphometry) analysis was performed ona group of 54 subjects, including both normal subjects and patients suffering from different forms ofneurodegenerative disease. Correlations between grey matter volume and reading accuracy, withrespect to exception words and pseudowords respectively, were computed. This analysis revealed thata positive correlation exists between accuracy for exception word and grey matter volume in the leftanterior middle and superior temporal gyri, in the left temporal pole, and in the anterior part of thefusiform gyrus. Moreover, semantic deficits co-occur in most cases of surface dyslexia (Patterson &Lambon Ralph, 1999), and the foci of lesion observed in most of the surface dyslexic patientsdescribed by Vanier and Caplan (1985) is largely overlapping with our word-related network.

Moreover, our results on what we identified as a semantic network were replicated in recentneuroimaging studies. A word-semantic role for the anterior occipito-temporal region is supported bythe study by Seghier, Lee, Schofield, Ellis, and Price (2008), who found that activation in this regioncorrelatedwith off-line performance in irregularly spelledword reading; the correlation had a negativesign: the more was the disadvantage for irregular word reading, the more was the activation in theanterior occipito-temporal region. The interpretation of this finding offered by the authors is that moreneural labour is required in this region by those who have more limited skills for irregular wordreading, which might correspond to an increased dependence on semantic processing to compensate.Similar considerations apply in the study by Graves et al. (2010) where the activation of the left middletemporal gyrus and inferior temporal sulcus showed a negative correlationwith word consistency. Theauthors also reported on positive correlations with word frequency and imageability in the angulargyrus and precuneus/posterior cingulated cortex bilaterally, suggesting a semantic role for these areasthat fits well with the results of our meta-analysis. The involvement in semantic processing of theseregions has also been advanced in Levy et al. (2008), where the words > pseudowords comparisonidentified a focus of activation in the bilateral inferior parietal lobule, close to our word-related leftangular cluster, and in Carreiras et al. (2009), where greater activation for high-frequency words versuslow frequency words in a lexical decision task was found in the precuneus.

On the other hand, in disagreement with part of the literature on reading and language in general,our study does not identify any inferior frontal cluster in the semantic network. In fact, the orbitalpart of the inferior frontal gyrus (BA 47) was definitely non-differentiated as far as lexicality, regu-larity or frequency of the items in hand. Although several studies have associated the more anteriorand ventral aspect of the inferior frontal gyrus (BA 47/45) with semantic processing, the results of ourmeta-analysis do not support such association: this suggests that the nature of neural processing inthis brain region might require additional, carefully controlled studies in order to be fullydisambiguated.

4.3. Pseudoword-related network

Finally, there were two parietal clusters having a reliable pseudoword effect: one located in the leftinferior parietal gyrus, and one in the right parietal cortex, between the angular and the superiorparietal gyrus.


The left inferior parietal cluster falls in a region frequently associated with sub-lexical phonologicalprocesses for tasks not necessarily involving reading (Démonet, Price, Wise, & Frackowiak, 1994;Paulesu, Frith, & Frackowiak, 1993; Paulesu et al., 1996), at the border between the supramarginal andthe angular gyrus. This region falls anteriorly to aword-related angular gyrus cluster. The identificationof two functionally different and separate clusters within the same larger inferior parietal regions helpsto clarify the role of the inferior parietal cortex, particularly with reference to the angular gyrus. Theassociation of the angular gyrus to reading is as old as neuropsychology, starting from the observationsof Dejerine on alexia with agraphia (Dejerine, 1892). The fact that our meta-analysis found two clustersin this region, with a different functional attribution (word-semantic processing versus pseudowordreading) might suggest that the historical controversy about the role of the left angular gyrus in readingis, at least to some extent, due to a specialization of this brain area into separate neural subsystems.

The position of the right parietal cluster mirrors the left parietal cluster associated with the diffi-culty effect. A similar functional interpretation can be given for this right-sided region: the complexityof the input stimulus may trigger more attentional resources, with particular reference to theexploratory eye movements necessary to analyse the structure of the stimulus.

While these two parietal clusters were the only ones to survive statistical testing individually, fourleft ventral occipito-temporal clusters, once analysed collectively, showed a significant pseudowordpreference. This suggests the existence of a more distributed network, subsuming all these clusters, towhich the pseudoword effect can be attributed.

The ventral occipito-temporal network encompassed by the four clusters that were analysedcollectively includes the midfusiform region known as Visual Word Form Area (VWFA; Cohen et al.,2000, 2002). This region, centred on Talairach coordinates x ¼ �43, y ¼ �54, z ¼ �12, has beeninterpreted as being dedicated to the analysis and recognition of orthographic forms for visuallypresented letter strings. Currently, there is some consensus on the view that this region showsa functional specialization, although not exclusive, to the orthographic analysis of units of writtenmaterial (from single letters up to, possibly, entire words). However, Price and Devlin (2003) reviewneuropsychological and neuroimaging evidence against this interpretation. From a neuropsychologicalperspective, they report that no one-to-one correspondence has been found in literature between purealexia (specific reading deficit, not accompanied by agraphia or by oral comprehension disruption) andfocal lesions to the VWFA. As for the neuroimaging evidence, Price and Devlin (2003) show that thismidfusiform region is actually activated also in tasks that do not involve the presentation of ortho-graphic stimuli, such as picture naming (and to a greater extent than word reading itself), and tasksusing auditory material. It has been proposed (Price & Mechelli, 2005) that a more plausible inter-pretation of the role of the VWFA is that of an interface for the retrieval of phonology from a visualinput. Both positions (in favour or against the existence of a visual word form area) agree in that theputative computation taking place in the VWFA is not exclusively committed to either words orpseudowords, but is an early, shared step in reading associated with orthographic input processing.

However, a pseudoword effect here remains plausible if one considers that pseudowords may bemore resource demanding if representations are held here of orthographic strings of a grain size up tothat of a whole word. Incomplete matching of a pseudoword with pre-existing templates may lead toextensive search through the neuronal representation, hence, larger activations in the comparison ofpseudowords versus words (cf. Paulesu et al., 2000).

It should also be noted that the different subdivisions of the left fusiform gyrus have shown somedegree of specificity to either word or pseudoword processing. In particular the most posterior part ofthe fusiform gyrus has already been associated with pseudoword processing in a number of studies(see, for instance, Mechelli et al., 2005), as opposed to the more anterior aspect that is reported to bemore consistently activated by semantic tasks. The VWFA lies in between these subdivisions. Due to thenecessarily limited resolution of our meta-analysis, it cannot be guaranteed that such a fine distinctionabout fusiform subdivisions is preserved, and therefore pseudoword-related activations that might beascribed to the more posterior part, might confound the picture regarding the VWFA. However, ourdata did capture the anterior-posterior functional distinction for the left fusiform gyrus, with theposterior fusiform cluster being classified among pseudoword-related areas (once the significance wasassessed collectively across four clusters), and the anterior fusiform cluster entering the word-related(semantic) network.


In addition, it is worth noting that among the ventral inferior temporal cortices associated withpseudoword reading lies the so-called lateral-inferotemporal multimodal area (LIMA), an area ofoverlay between activations for reading or for phonological processing (Cohen, Jobert, Le Bihan, &Dehaene, 2004), and therefore a region where an initial integration between orthography andphonology may take place.

To sum up, our pseudoword-related network comprises a right parietal cluster possibly involved inattentional processes; a left inferior parietal cluster for which a role in phonological processing can beadvanced; and a set of ventral occipito-temporal regions where we hypothesize that an earlyorthographic-phonological integration takes place. This view is compatible with a similar account ofthe VWFA (see for instance Graves et al., 2010; Hillis et al., 2005; Price & Devlin, 2003), which isencompassed by our occipito-temporal clusters. Even though the ventral occipito-temporal regionstend to respond more strongly to pseudowords than to words, this stage of orthographic processingappears to be common to both categories of stimuli.

Additional support to our results on the pseudoword network comes from recent studies, such asthe one by Brambati et al. (2009), where a correlational analysis between pseudoword reading accu-racy and grey matter volume was performed and found to be significant for the left inferior parietallobule (angular gyrus), the posterior aspect of the middle and superior temporal gyri, and the posteriorfusiform gyrus. In Levy et al. (2008), pseudoword reading yielded greater activation in a left parietalregion close to our pseudoword-specific cluster in the same region, and in the left occipito-temporalarea, in agreement with our findings. Seghier et al. (2008) reported a negative correlation betweenpseudoword reading performance and activation in a network involving the left posterior occipito-temporal region and bilateral intraparietal cortex, suggesting that an ineffective pseudowordreading strategy would increase the processing load on such areas. Activation in the left occipito-temporal region for pseudoword reading was found also in Kronbichler et al. (2009) and Nosartiet al. (2010).

4.4. Towards a functional model of reading

In this final sectionwe sketch a functional model of reading that takes into account both the resultsof our meta-analysis and data from patients. The model is illustrated in Fig. 5.

We propose that the ventral occipital and occipito-temporal regions (pink blobs in Fig. 5), whichinclude the so-called VWFA, may be a first neuronal station where visual information is routed to afterearly visual processing. Here, orthographic processing takes place, and possibly there is a first stage oforthographic to phonological integration, particularly in the more anterior and lateral part of this set ofareas, in a region roughly corresponding to the area called LIMA (Cohen et al., 2004). Further evidenceof some degree of overlap between a phonological task and a reading task has been recently found inthis same region (Danelli and Paulesu, personal communication).

Interactions of these occipito-temporal regions with the word-semantic network (orange blobs inFig. 5) may contribute to complete the extraction of word phonology for real words, particularly forwords with irregular spelling. On the other hand, during pseudoword reading the left supramarginalgyrus (purple blob in Fig. 5) may contribute either to further integration between phonologicalknowledge and orthographic representations, or to phonological implementation in speech output.

Semantically-mediated andmore phonologically-mediated mappings are then routed to the frontallobe (blue blob in Fig. 5) for further processing: the final pronunciation may be decided here, by meansof integrative and competitive mechanisms.

Some predictions can be made on the kind of neuropsychological deficits that would be observedwhen damaging any of the components of the proposedmodel. A left occipito-temporal damagewouldimpair basic orthographic processing, thereby resulting in letter-by-letter reading, a suggestionconsistent with classical anatomical findings in alexiawithout agraphia (see review in Cappa & Vignolo,1999) and recent subdural electrical inhibition data (Mani et al., 2008). This observation supports theidea that both real words and pseudowords are necessarily processed here in a normal fluent reader.

According to our proposedmodel, a left inferior parietal lesion, or a damage causing a disconnectionbetween the ventral occipito-temporal cortex and this site, would hamper aspects of the grapheme-to-phoneme mappings or phonological implementation; this would be especially detrimental to

Fig. 5. A hypothetical model of the reading network based on the results of this meta-analysis. No explicit connections are rep-resented as these are not evaluated in this study. In pink, orthographic input nodes collectively more active for pseudoword reading,but also involved in word reading; in orange, word-specific nodes also involved in semantic processes; in purple, a region associatedwith pseudoword reading; in blue, areas sensitive to items difficulty, with the left inferior opercular region associated withphonological output processes. For a full description, see the final section of the Discussion.


pseudoword reading, since these cannot rely on semantic mediation to recover phonological codes.Such a lesion may therefore produce a pattern of phonological dyslexia, not deprived, though, fromother phonological symptoms. However, data from patients are not completely consistent with themodel we are hypothesizing here: in fact, lesions of the left supramarginal gyrus are not associatedwith a pure phonological dyslexia; rather, patients with such lesion typically show either conductionaphasia (Damasio & Damasio, 1980) or phonological short-term memory deficits (Shallice & Vallar,1990; Vallar, Di Betta, & Silveri, 1997). The former syndrome is associated with errors in pseudowordreading (Bisiacchi, Cipolotti, & Denes, 1989); however, errors with pseudoword processing are notlimited to reading tasks. To make matters more complicated, there are patients with a lesion involvingthe left supramarginal gyrus and phonological short-term memory deficits who could read pseudo-words (T. Shallice, personal communication about patient JB; G. Vallar personal communication aboutpatient PV). More data may be needed to clarify the question.

On the other hand, if the semantic network is damaged, then surface dyslexia would be expected:the semantic contribution to reading would be compromised, and this could hamper especially thereading of irregular words, that rely more than regular words on semantic mediation. The dominantpattern of brain damage seen in patients with surface dyslexia (Vanier & Caplan, 1985) fits well withthis prediction.

Finally, a lesion in the left frontal operculumwould result in a reading deficit especially for the mostdifficult stimuli, such as pseudowords and irregular words (see the patients described in Fiez et al.,2006), rather than in a reading deficit sensitive only to the lexical status of the stimuli.

We can conclude that the functional model emerging from our meta-analysis captures reasonablywell the spectrum of neuropsychological deficits in reading, representing a valuable interpretation ofthe neural dynamics underlying the reading ability. We can now turn to examine how this model canbe mapped to some of the classical cognitive models of reading: in particular, we are considering heretwo of the most popular theories, namely the dual-route theory (Coltheart, Curtis, Atkins, & Haller,


1993; Coltheart, Rastle, Perry, Langdon, & Ziegler, 2001), and the connectionist single-mechanismaccounts (e.g. Plaut, McClelland, Seidenberg, & Patterson, 1996; Seidenberg & McClelland, 1989).

Perhaps one of the main differences between the two classes of models lies in the importance theyattribute to semantics in mapping orthography to phonology. In the dual-route theory, semantics playsa marginal role, as it is not considered crucial for computing the correct pronunciation of the word. Infact, surface dyslexia is explained, within this theory, as the result of damage to the orthographiclexicon where whole word-forms are stored, separately from the representation of their meaning. Onthe other hand, in connectionist models the semantic system plays a more decisive role: orthographiccodes can be mapped directly to phonological ones (direct, phonological pathway), but they can alsobenefit from semantic influences (indirect, semantic pathway). The two pathways cooperate to achieveskillful reading according to a division of labour scheme (Harm & Seidenberg, 2004; Plaut et al., 1996):since the semantic pathway intervenes to support naming for words only, the phonological pathwaylearns to count on such contribution and experiences less pressure to perfectly master all words. Asa result, the contribution of the semantic pathway tends to get greater for “hard”words (irregular and/or low frequency items), whereas the purely phonological pathway gets to be more finely tuned topseudoword reading (which, on the other hand, cannot take advantage of any semantic information).

The results of our meta-analysis appear to be more in line with the class of connectionist models: infact, we were not able to isolate any region involved in purely (i.e., non-semantic) orthographic pro-cessing that might be considered the neural implementation of an orthographic lexicon, whereas wedid identify a semantic network whose distribution is fully consistent with lesion patterns (typicallyinvolving the middle temporal gyrus/superior temporal sulcus or the angular gyrus – Vanier & Caplan,1985) observed in surface dyslexia. However, our results alone cannot be used to adjudicate betweenthese competing theories of reading, especially because their founding assumptions, althoughapparently very different, turn out to be very hard to test (and discriminate) with functional neuro-imaging experiments.

5. Conclusion

In this paper a meta-analysis on neuroimaging studies investigating reading processes has beenpresented. Word-specific, pseudoword-specific, and difficulty-sensitive brain regions were identified.Evidence from the literature supports the interpretation that the word-related network may coincidewith the neural system involved in semantic processing while the pseudoword-related one may beinvolved in the direct mapping of orthographic representations into phonological ones. The difficultyeffect, especially as far as the left frontal operculum is concerned, may be interpreted as the result ofcompetition among discrepant, or unclear, phonological responses that arise when the mapping fromorthography to phonology is a difficult one (either for reasons of consistency, frequency, or lexicality).

Our results, which have been validated by more recent neuroimaging studies on reading processesand by patients’ data, contribute to delineate a clearer picture of the reading systems in the humanbrain. However, some open questions remain. For instance, the nature of neural processing in theidentified pseudoword network still needs to be clarified. Moreover, although our results providedsome support for the class of connectionist models of reading, the available data appear to be stillinsufficient for definitely adjudicating between competing theories. Therefore, it is advisable that new,carefully constructed neuroimaging work is performed, so that new, informative data can increase thepower of future meta-analyses to delineatemore in details the circuits involved in reading, as well as inother cognitive functions.

Appendix A. Supplementary materialSupplementarymaterial related to this article can be found at http://dx.doi.org/10.1016/j.jneuroling.

2012.08.001.

References

Beauregard, M., Chertkow, H., Bub, D., Murtha, S., Dixon, R., & Evans, A. (1997). The neural substrate for concrete, abstract, andemotional word lexica: a Positron Emission Tomography study. Journal of Cognitive Neuroscience, 9(4), 441–461.




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

Binder, J. R., McKiernan, K. A., Parsons, M. E., Westbury, C. F., Possing, E. T., Kaufman, J. N., et al. (2003). Neural correlates oflexical access during visual word recognition. Journal of Cognitive Neuroscience, 15(3), 372–393.

Binder, J. R., Medler, D. A., Desai, R., Conant, L. L., & Liebenthal, E. (2005). Some neurophysiological constraints on models ofword naming. NeuroImage, 27(3), 677–693.

Bisiacchi, P. S., Cipolotti, L., & Denes, G. (1989). Impairment in processing meaningless verbal material in several modalities: therelationship between short term memory and phonological skills. The Quarterly Journal of Experimental Psychology, 41 A(2),293–319.

Bolger, D. J., Perfetti, C. A., & Schneider, W. (2005). Cross-cultural effect on the brain revisited: universal structures plus writingsystem variation. Human Brain Mapping, 25, 92–104.

Bookheimer, S. Y., Zeffiro, T. A., Blaxton, T., Gaillard, W., & Theodore, W. (1995). Regional cerebral blood flow during objectnaming and word reading. Human Brain Mapping, 3, 93–106.

Bottini, G., & Paulesu, E. (2003). Functional neuroanatomy of spatial perception, spatial processes, and attention. In P. W.Halligan, U. Kischka, & J. C. Marshall (Eds.), Handbook of clinical neuropsychology (pp. 697–723). New York: Oxford UniversityPress.

Brambati, S. M., Ogar, J., Neuhaus, J., Miller, B. L., & Gorno-Tempini, M. L. (2009). Reading disorders in primary progressiveaphasia: a behavioral and neuroimaging study. Neuropsychologia, 47, 1893–1900.

Broca, P. (1861). Loss of speech, chronic softening and partial destruction of the anterior left lobe of the brain. Bulletin de laSociété Anthropologique de Paris, 2, 235–238.

Buchel, C., Price, C., & Friston, K. (1998). A multimodal language region in the ventral visual pathway. Nature, 394, 274–277.Cappa, S. F., & Vignolo, L. A. (1999). The neurological foundations of language. In G. Denes, & L. Pizzamiglio (Eds.), Handbook of

clinical and experimental neuropsychology (pp. 155–179). Psychology Press.Carreiras, M., Mechelli, A., & Price, C. J. (2006). Effect of word and syllable frequency on activation during lexical decision and

reading aloud. Human Brain Mapping, 27(12), 963–972.Carreiras, M., Riba, J., Vergara, M., Heldmann, M., & Münte, T. F. (2009). Syllable congruency and word frequency effects on brain

activation. Human Brain Mapping, 30, 3079–3088.Cattinelli, I., Valentini, G., Paulesu, E., & Borghese, N. A. A novel approach to the problem of non-uniqueness of the solution in

hierarchical clustering, Under Review.Chee, M. W., O’Craven, K. M., Bergida, R., Rosen, B. R., & Savoy, R. L. (1999). Auditory and visual word processing studied with

fMRI. Human Brain Mapping, 7(1), 15–28.Chein, J. M., Fissell, K., Jacobs, S., & Fiez, J. A. (2002). Functional heterogeneity within Broca’s area during verbal working

memory. Physiology & Behavior, 77, 635–639.Cohen, L., Dehaene, S., Naccache, L., Lehericy, S., Dehaene-Lambertz, G., Henaff, M. A., et al. (2000). The visual word form area:

spatial and temporal characterization of an initial stage of reading in normal subjects and posterior split-brain patients.Brain, 123(Pt 2), 291–307.

Cohen, L., Jobert, A., Le Bihan, D., & Dehaene, S. (2004). Distinct unimodal and multimodal regions for word processing in theleft temporal cortex. NeuroImage, 23, 1256–1270.

Cohen, L., Lehericy, S., Chochon, F., Lemer, C., Rivaud, S., & Dehaene, S. (2002). Language-specific tuning of visual cortex?Functional properties of the visual word form area. Brain, 125(Pt 5), 1054–1069.

Coltheart, M., Curtis, B., Atkins, P., & Haller, M. (1993). Models of reading aloud: dual-route and parallel-distributed-processingapproaches. Psychological Review, 100, 589–608.

Coltheart, M., Rastle, K., Perry, C., Langdon, R., & Ziegler, J. (2001). DRC: a dual route cascaded model of visual word recognitionand reading aloud. Psychological Review, 108(1), 204–256.

Corbetta, M. (1998). Frontoparietal cortical networks for directing attention and the eye to visual locations: identical, inde-pendent, or overlapping neural systems? Proceedings of the National Academy of Sciences of the United States of America, 95,831–838.

Damasio, H., & Damasio, A. R. (1980). The anatomical basis of conduction aphasia. Brain, 103, 337–350.Dejerine, J. (1892). Contribution à l’étude anatomo-pathologique et clinique des differentes varietiés de cecité verbale.Mémoires

de la Société de Biologie, 4, 61–90.Démonet, J.-F., Price, C., Wise, R., & Frackowiak, R. (1994). Differential activation of right and left posterior sylvian regions by

semantic and phonological tasks: a positron-emission tomography study in normal human subjects. Neuroscience Letters,182, 25–28.

Ferrand, L., & New, B. (2003). Syllabic length effects in visual word recognition and naming. Acta Psychologica, 113(2), 167–183.Fiebach, C. J., Friederici, A. D., Muller, K., & von Cramon, D. Y. (2002). fMRI evidence for dual routes to the mental lexicon in

visual word recognition. Journal of Cognitive Neuroscience, 14(1), 11–23.Fiebach, C. J., Ricker, B., Friederici, A. D., & Jacobs, A. M. (2007). Inhibition and facilitation in visual word recognition: prefrontal

contribution to the orthographic neighborhood size effect. NeuroImage, 36(3), 901–911.Fiez, J. A. (1997). Phonology, semantics, and the role of the left inferior prefrontal cortex. Human Brain Mapping, 5(2), 79–83.Fiez, J. A., Balota, D. A., Raichle, M. E., & Petersen, S. E. (1999). Effects of lexicality, frequency, and spelling-to-sound consistency

on the functional anatomy of reading. Neuron, 24(1), 205–218.Fiez, J. A., & Petersen, S. E. (1998). Neuroimaging studies of word reading. Proceedings of the National Academy of Sciences of the

United States of America, 95(3), 914–921.Fiez, J. A., Tranel, D., Seager-frerichs, D., & Damasio, H. (2006). Specific reading and phonological processing deficits are asso-

ciated with damage to the left frontal operculum. Cortex, 42, 624–643.Forster, K. I., & Chambers, S. (1973). Lexical access and naming time. Journal of Verbal Learning and Verbal Behaviour, 12, 627–635.Gates, L., & Yoon, M. G. (2005). Distinct and shared cortical regions of the human brain activated by pictorial depictions versus

verbal descriptions: an fMRI study. NeuroImage, 24(2), 473–486.Glushko, R. J. (1979). The organization and activation of orthographic knowledge in reading aloud. Journal of Experimental

Psychology: Human Perception and Performance, 5, 674–691.


Graves, W. W., Desai, R., Humphries, C., Seidenberg, M. S., & Binder, J. R. (2010). Neural systems for reading aloud: a multi-parametric approach. Cerebral Cortex, 20, 1799–1815.

Hagoort, P., Indefrey, P., Brown, C., Herzog, H., Steinmetz, H., & Seitz, R. J. (1999). The neural circuitry involved in the reading ofGerman words and pseudowords: a PET study. Journal of Cognitive Neuroscience, 11(4), 383–398.

Harm, M. W., & Seidenberg, M. S. (2004). Computing the meanings of words in reading: cooperative division of labor betweenvisual and phonological processes. Psychological Review, 111(3), 662–720.

Hauk, O., Davis, M. H., & Pulvermüller, F. (2008). Modulation of brain activity by multiple lexical and word form variables invisual word recognition: a parametric fMRI study. NeuroImage, 42, 1185–1195.

Herbster, A. N., Mintun, M. A., Nebes, R. D., & Becker, J. T. (1997). Regional cerebral blood flow during word and nonwordreading. Human Brain Mapping, 5(2), 84–92.

Hillis, A. E., Newhart, M., Heidler, J., Barker, P., Herskovits, E., & Degaonkar, M. (2005). The roles of the “visual word form area” inreading. NeuroImage, 24, 548–559.

Howard, D., Patterson, K., Wise, R., Brown, W. D., Friston, K., Weiller, C., et al. (1992). The cortical localization of the lexicons.Positron emission tomography evidence. Brain, 115(6), 1769–1782.

Jessen, F., Erb, M., Klose, U., Lotze, M., Grodd, W., & Heun, R. (1999). Activation of human language processing brain regions afterthe presentation of random letter strings demonstrated with event-related functional magnetic resonance imaging.Neuroscience Letters, 270(1), 13–16.

Jobard, G., Crivello, F., & Tzourio-Mazoyer, N. (2003). Evaluation of the dual route theory of reading: a metaanalysis of 35neuroimaging studies. NeuroImage, 20(2), 693–712.

Joubert, S., Beauregard, M., Walter, N., Bourgouin, P., Beaudoin, G., Leroux, J. M., et al. (2004). Neural correlates of lexical andsublexical processes in reading. Brain and Language, 89(1), 9–20.

Kastner, S., & Ungerleider, L. G. (2000). Mechanisms of visual attention in the human cortex. Annual Review of Neuroscience, 23,315–341.

Kiehl, K. A., Liddle, P. F., Smith, A. M., Mendrek, A., Forster, B. B., & Hare, R. D. (1999). Neural pathways involved in the processingof concrete and abstract words. Human Brain Mapping, 7(4), 225–233.

Kronbichler, M., Klackl, J., Richlan, F., Schurz, M., Staffen, W., Ladurner, G., et al. (2009). On the functional neuroanatomy of visualword processing: effects of case and letter deviance. Journal of Cognitive Neuroscience, 21, 222–229.

Kuchinke, L., Jacobs, A. M., Grubich, C., Vo, M. L., Conrad, M., & Herrmann, M. (2005). Incidental effects of emotional valence insingle word processing: an fMRI study. NeuroImage, 28(4), 1022–1032.

Levy, J., Pernet, C., Treserras, S., Boulanouar, K., Berry, I., Aubry, F., et al. (2008). Piecemeal recruitment of left-lateralized brainareas during reading: a spatio-functional account. NeuroImage, 43, 581–591.

Mani, J., Diehl, B., Piao, Z., Schuele, S. S., Lapresto, E., Liu, P., et al. (2008). Evidence for a basal temporal visual language center:cortical stimulation producing pure alexia. Neurology, 71, 1621–1627.

McCann, R. S., & Besner, D. (1987). Reading pseudohomophones: implications of models of pronunciation and the locus ofthe word-frequency effect in word naming. Journal of Experimental Psychology: Human Perception and Performance, 13,14–24.

Mechelli, A., Crinion, J. T., Long, S., Friston, K. J., Lambon Ralph, M. A., Patterson, K., et al. (2005). Dissociating reading processeson the basis of neuronal interactions. Journal of Cognitive Neuroscience, 17(11), 1753–1765.

Mechelli, A., Friston, K. J., & Price, C. J. (2000). The effects of presentation rate during word and pseudoword reading:a comparison of PET and fMRI. Journal of Cognitive Neuroscience, 12(Suppl. 2), 145–156.

Mechelli, A., Gorno-Tempini, M. L., & Price, C. J. (2003). Neuroimaging studies of word and pseudoword reading: consistencies,inconsistencies, and limitations. Journal of Cognitive Neuroscience, 15(2), 260–271.

Menard, M. T., Kosslyn, S. M., Thompson, W. L., Alpert, N. M., & Rauch, S. L. (1996). Encoding words and pictures: a positronemission tomography study. Neuropsychologia, 34(3), 185–194.

Meschyan, G., & Hernandez, A. E. (2006). Impact of language proficiency and orthographic transparency on bilingual wordreading: an fMRI investigation. NeuroImage, 29(4), 1135–1140.

Moore, C. J., & Price, C. J. (1999). Three distinct ventral occipitotemporal regions for reading and object naming. NeuroImage,10(2), 181–192.

Morgan, B. J. T., & Ray, A. P. G. (1995). Non-uniqueness and inversion in cluster analysis. Applied Statistics, 44(1), 117–134.Nosarti, C., Mechelli, A., Green, D. W., & Price, C. J. (2010). The impact of second language learning on semantic and nonsemantic

first language reading. Cerebral Cortex, 20, 315–327.Paap, K. R., & Noel, R. W. (1991). Dual route models of print to sound: still a good horse race. Psychological Research, 53, 13–24.Patterson, K., & Lambon Ralph, M. A. (1999). Selective disorders of reading? Current Opinion in Neurobiology, 9, 235–239.Paulesu, E., Frith, C., & Frackowiak, R. (1993). The neural correlates of the verbal component of working memory. Nature, 362,

342–345.Paulesu, E., Frith, U., Snowling, M., Gallagher, A., Morton, J., Frackowiak, R. S., et al. (1996). Is developmental dyslexia

a disconnection syndrome? Evidence from PET scanning. Brain, 119, 143–157.Paulesu, E., McCrory, E., Fazio, F., Menoncello, L., Brunswick, N., Cappa, S. F., et al. (2000). A cultural effect on brain function.

Nature Neuroscience, 3(1), 91–96.Peeva, M. G., Guenther, F. H., Tourville, J. A., Nieto-Castanon, A., Anton, J.-L., Nazarian, B., et al. (2010). Distinct representations of

phonemes, syllables, and supra-syllabic sequences in the speech production network. NeuroImage, 50, 626–638.Perani, D., Cappa, S. F., Schnur, T., Tettamanti, M., Collina, S., Rosa, M. M., et al. (1999). The neural correlates of verb and noun

processing. A PET study. Brain, 122(Pt 12), 2337–2344.Petersen, S. E., & Fiez, J. A. (1993). The processing of single words studied with positron emission tomography. Annual Review of

Neuroscience, 16, 509–530.Plaut, D. C., McClelland, J. L., Seidenberg, M. S., & Patterson, K. (1996). Understanding normal and impaired word reading:

computational principles in quasi-regular domains. Psychological Review, 103(1), 56–115.Price, C. J. (2000). The anatomy of language: contributions from functional neuroimaging. Journal of Anatomy, 197(Pt 3),

335–359.Price, C. J., & Devlin, J. T. (2003). The myth of the visual word form area. NeuroImage, 19(3), 473–481.


Price, C. J., Gorno-Tempini, M. L., Graham, K. S., Biggio, N., Mechelli, A., Patterson, K., et al. (2003). Normal and pathologicalreading: converging data from lesion and imaging studies. NeuroImage, 20(Suppl. 1), S30–S41.

Price, C. J., McCrory, E., Noppeney, U., Mechelli, A., Moore, C. J., Biggio, N., et al. (2006). How reading differs from object namingat the neuronal level. NeuroImage, 29(2), 643–648.

Price, C. J., & Mechelli, A. (2005). Reading and reading disturbance. Current Opinion in Neurobiology, 15(2), 231–238.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(1), 40–52.Price, C. J., Wise, R. J., & Frackowiak, R. S. (1996). Demonstrating the implicit processing of visually presented words and

pseudowords. Cerebral Cortex, 6(1), 62–70.Price, C. J., Wise, R. J., Watson, J. D., Patterson, K., Howard, D., & Frackowiak, R. S. (1994). Brain activity during reading. The effects

of exposure duration and task. Brain, 117(6), 1255–1269.Pugh, K. R., Mencl, W. E., Jenner, A. R., Katz, L., Frost, S. J., Lee, J. R., et al. (2000). Functional neuroimaging studies of

reading and reading disability (developmental dyslexia). Mental Retardation and Developmental Disabilities ResearchReviews, 6(3), 207–213.

Rorden, C., & Brett, M. (2000). Stereotaxic display of brain lesions. Behavioural Neurology, 12(4), 191–200.Rumsey, J. M., Horwitz, B., Donohue, B. C., Nace, K., Maisog, J. M., & Andreason, P. (1997). Phonological and orthographic

components of word recognition. A PET-rCBF study. Brain, 120(Pt 5), 739–759.Seghier, M., Lee, H. L., Schofield, T., Ellis, C. L., & Price, C. J. (2008). Inter-subject variability in the use of two different neuronal

networks for reading aloud familiar words. NeuroImage, 42, 1226–1236.Seidenberg, M. S., & McClelland, J. L. (1989). A distributed, developmental model of word recognition and naming. Psychological

Review, 96(4), 523–568.Seidenberg, M. S., Waters, G. S., Barnes, M. A., & Tanenhaus, M. K. (1984). When does irregular spelling or pronunciation

influence word recognition? Journal of Verbal Learning and Verbal Behaviour, 23, 383–404.Shallice, T., & Vallar, G. (1990). The impairment of auditory-verbal short-term storage. In G. Vallar, & T. Shallice (Eds.), Neuro-

psychological impairments of short-term memory (pp. 11–53). Cambridge University Press.Talairach, J., & Tournoux, P. (1988). Co-planar stereotaxic atlas of the human brain: 3-D proportional system: An approach to

cerebral imaging. New York, NY: Thieme Medical Publishers.Tan, L. H., Laird, A. R., Li, K., & Fox, P. T. (2005). Neuroanatomical correlates of phonological processing of Chinese characters and

alphabetic words: a meta-analysis. Human Brain Mapping, 91, 83–91.Taraban, R., & McClelland, J. L. (1987). Conspiracy effects in word pronunciation. Journal of Memory and Language, 26, 608–631.Turkeltaub, P. (2002). Meta-analysis of the functional neuroanatomy of single-word reading: method and validation. Neuro-

Image, 16, 765–780.Tzourio-Mazoyer, N., Landeau, B., Papathanassiou, D., Crivello, F., Etard, O., Delcroix, N., et al. (2002). Automated anatomical

labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. NeuroImage,15, 273–289.

Valdois, S., Carbonnel, S., Juphard, A., Baciu, M., Ans, B., Peyrin, C., et al. (2006). Polysyllabic pseudo-word processing in readingand lexical decision: converging evidence from behavioral data, connectionist simulations and functional MRI. BrainResearch, 1085(1), 149–162.

Vallar, G., Di Betta, A. M., & Silveri, M. C. (1997). The phonological short-term store-rehearsal system: patterns of impairmentand neural correlates. Neuropsychologia, 35, 795–812.

Vanier, M., & Caplan, D. (1985). CT scan correlates of surface dyslexia. In K. E. Patterson, J. C. Marshall, & M. Coltheart (Eds.),Surface dyslexia: Neuropsychological and cognitive studies of phonological reading (pp. 511–522). Hillsdale, NJ: LawrenceErlbaum Associates.

Vigneau, M., Jobard, G., Mazoyer, B., & Tzourio-Mazoyer, N. (2005). Word and non-word reading: what role for the visual wordform area? NeuroImage, 27(3), 694–705.

Vinckier, F., Dehaene, S., Jobert, A., Dubus, J. P., Sigman, M., & Cohen, L. (2007). Hierarchical coding of letter strings in the ventralstream: dissecting the inner organization of the visual word-form system. Neuron, 55(1), 143–156.

Ward, J. H. J. (1963). Hierarchical grouping to optimize an objective function. Journal of the American Statistical Association, 58,236–244.

Xu, R., & Wunsch, D. I. (2005). Survey of clustering algorithms. IEEE Transactions on Neural Networks, 16(3), 645–678.

Reading the reading brain: A new meta-analysis of functional imaging data on reading

Documents

Transcript of Reading the reading brain: A new meta-analysis of functional imaging data on reading