Neuroscience and Biobehavioral Reviews 80 (2017) 646–656

Contents lists available at ScienceDirect

Neuroscience and Biobehavioral Reviews

journal homepage: www.elsevier.com/locate/neubiorev

Meta-analysis

Reviewing the functional basis of the syntactic Merge mechanism for T

language: A coordinate-based activation likelihood estimation meta-analysis
⁎
Emiliano Zaccarella , Marianne Schell, Angela D. Friederici
Max Planck Institute for Human Cognitive Brain Sciences, Department of Neuropsychology, Stephanstraße 1a, 04103 Leipzig, Germany

A R T I C L E I N F O A B S T R A C T

Keywords: The ability to create structures out of single words is a key aspect of human language. This combinatorial
Merge capacity relies on a low-level syntactic mechanism—Merge—assembling words into hierarchies. Neuroscience
Language has explored Merge by comparing syntax to word-lists. Here, we first review potential issues with the word-lists
Meta-analysis materials. We then perform an activation likelihood estimation (ALE) on the reported foci, to reveal functional
ALE
convergence for Merge at whole-brain level. Finally, we run probabilistic tractography on an independent po-
Pars opercularis
pulation to observe how these convergent activations anatomically connect. Functionally, we found that when
confounding activity was removed, consistency for Merge was only observable in the left pars opercularis (BA44)
and in the inferior part of the posterior superior temporal sulcus/gyrus (pSTS/STG; BA22). Structurally, we could
confirm that the two regions are connected through dorsal fiber bundles. We therefore suggest that the cortical
implementation of linguistic Merge consists of a left fronto-temporal interaction between BA44 (syntactic pro-
cessor) and pSTS/STG (integrative processor), which communicate to each other along dorsal white matter
fascicles.

1. Introduction hierarchical assemblies of words forming more composite phrasal and

sentential constituents (represented in square brackets below) which
In our everyday life we are constantly faced to linguistic expressions may in turn be part of bigger ones: [the man], [[the man] [eats [an
we have never encountered before. Although such expressions can be apple]]], etc. At the most atomic level, words—the basic building blocks
new to us and might vary in terms of complexity, we are always im- of linguistic structures—can be distinguished into two different cate-
mediately able to identify what their structures and related meanings gories: content class categories and functional class categories. Content
might be. This adaptive linguistic capacity appears to be grounded on a words are those items primarily carrying lexical-semantic information,
very basic computation—known as Merge—which creates complex which serve descriptive content and referential weight, and are context-
structures out of single words according to the grammar of the language independent. They refer to events (either states or actions) and entities
in use (Chomsky, 1995). Because Merge constitutes the cognitive basis participating in them (Baker, 2003). Content words typically are nouns,
upon which our linguistic competence is founded, the understanding of adjectives, verbs and adverbs, and are often called members of the open
its cortical implementation is strongly advocated by both linguistic class since the class can increase its members. Function words, con-
theory (Rizzi, 2012), evolutionary linguistics (Bolhuis et al., 2014) and versely, primarily carry syntactic information, since they have reduced
neurobiology of language (Zaccarella and Friederici, 2016). Merge is semantic content, and work for structural assignment by linking other
taken to be the universal fundamental structure-building computation items. They are therefore context-dependent. Function words typically
of natural language syntax (Berwick et al., 2013; Chomsky, 1995). We include prepositions, pronouns, determiners, conjunctions and auxiliary
can formally represent this computation as: α β → {α β}, which means verbs, and are also called members of the closed class since they do not
“take two elements α and β, and string them together to form a new set increase in number. The way content and function words combine to-
containing both”. The two elements in the input, α and β can be two gether under Merge is essential to the construction of linguistic se-
lexical items (for example the and man)—which together form a bigger quences, in which the hierarchical relationships are strictly established
object {the man}. Depending on the relationship between the categories according the syntactic nature of the words entering the computation.
involved, sets are labeled and the hierarchy is established, e.g. the man In this sense, a determiner cannot combine with another determiner
is a determiner phrase (DP). As such, linguistic sequences are (the the) since it would not be possible to establish a hierarchical

⁎
Corresponding author.
E-mail address: [email protected] (E. Zaccarella).

http://dx.doi.org/10.1016/j.neubiorev.2017.06.011
Received 24 February 2017
Available online 23 July 2017
0149-7634/ © 2017 Elsevier Ltd. All rights reserved.
E. Zaccarella et al. Neuroscience and Biobehavioral Reviews 80 (2017) 646–656

relationship in which one of the two elements would dominate the based statistical parametric images. On these images, statistically sig-
other. Conversely, a determiner can successfully combine with a noun nificant activation for a specific voxel then can be calculated following
(the man) since an asymmetrical relationship between the two elements classical threshold methodologies. The main advantages to use such
exists, in which the determiner dominates the noun. At the neuroana- qualitative, objective coordinate-based algorithm are essentially three:
tomical level, the functional studies interested in the cortical im- first, concordance is examined at voxel-level using standardized local
plementation of Merge have taken direct advantage of the possibility to information, compared to label-based meta-analyses which use instead
generate lists of words that do not form structures, to compare the load anatomical labels assigned to specific locations. As such, the ALE ana-
generated by simple word-string processing (the the) against the load lysis is not susceptible to errors due to the use of generalized location
generated by syntactic-building processing (the man). This type of ex- information taken from different atlases or the adoption of anatomical
perimental manipulation allows the direct investigation of Merge while labels, which are assigned by the individual investigators to the re-
leaving both stimulus length and lexical access balanced across condi- ported foci. Second, ALE methods make use of subject size information
tions. The possibility to use this type of contrast has stimulated the for each foci group to calculate how blurred the Gaussian function for a
production of a growing number of functional studies, which have been specific focus has to be (the Full-Width Half-Maximum; FWHM) to more
published in the last twenty-five years in the neurolinguistic literature, realistically implement the variable uncertainty. In this respect, while
across languages and modalities, using both Positron Emission Tomo- label-based meta-analyses give the same activation weight to two stu-
graphy (PET) and functional Magnetic Resonance Imaging (fMRI). See dies differing in terms of experimental sample size and power effect
Appendices A and B for a detailed list of the studies. Across the ex- (e.g. 3 subjects vs. 18 subjects), the ALE algorithm will assign a higher
periments, all the perisylvian language network and neighboring re- weight (a tighter and taller Gaussian) to the foci of the study with a
gions have been basically found to participate in sentence processing bigger sample size, compared to the ones belonging the study with a
compared to word-list processing, including Broca’s area (BA44/45) in smaller sample size. Third, ALE random effects methods allows testing
the left inferior frontal cortex; the frontal operculum/anterior insula for qualitative differences across groups of studies, by pooling the
(FOP/aINS); the anterior portion of the left temporal lobe (BA38); the studies in subgroups and measuring the corresponding source of var-
posterior portion of the left superior temporal sulcus/gyrus (pSTS/STG; iation based on such qualitative differences.
BA22). As a matter of fact, a review on the neurological distribution of With respect to word-list studies taken into account in the present
formal syntactic operations could not report any functional character- study, one qualitative difference, which can be measured by ALE
ization for Merge, as no consistency across the data was available at the methods concerns with the nature of the word-list control condition
time (Grodzinsky and Friederici, 2006). used in the specific studies. A recent review on the brain basis of lan-
In the attempt to overcome such inconsistency across the data, a guage processing, Friederici (2011) noticed that the word-list condi-
very recent meta-analysis used the frequency of localization of func- tions often employ both content words (e.g. nouns, verbs, adjectives)
tional activity of different linguistic operations across the studies (label- and function words (e.g. determiners, prepositions, conjunctions) to-
based distribution meta-analysis), to offer a quantitative measurement gether, which may erroneously enhance rather than decrease the con-
of the neuroanatomical localization of the linguistic processes beyond struction of minimal structures in the non-syntactic conditions, for ex-
single word level (Hagoort and Indefrey, 2014). The reliability for a ample: “Money the [the client washed]” (Kuperberg et al., 2000); “[Her
particular region to be considered as active was assessed by obtaining a eyes during close] the she ceremony” (Vandenberghe et al., 2002). Cru-
relative binomial distribution based on the average number of activated cially, these studies report high involvement of temporal regions,
regions reported per experiment divided by the total number of brain compared to the infero-frontal regions, which are conversely not found
regions taken in the analysis (n = 112). Interestingly, the authors in- as active during subtraction analysis in the same subjects. One hy-
cluded in the report a subset of studies contrasting sentence-level pro- pothesis would then be that if control conditions consist of remaining
cessing with lower-level control conditions, ranging from resting con- syntactic chunks enhancing low-level syntactic processing—because of
ditions, or fixation of a hair cross, or word-lists—a subgroup (n = 15) the use of both content and function words together—then the com-
of which is included in the present study. Compared to control condi- parison of licit structures against word-lists might have removed the
tions below sentence level, the meta-analysis found that the temporal Merge effect, given the high automaticity of this type of process (Hahne
lobes and the posterior IFG bilaterally were significantly more active et al., 2002). Conversely, studies using content-only or function-only
than the resting condition, with clear left hemisphere dominance. word-list control conditions, in which no syntactic process is required,
Moreover, large parts of the right parietal and right inferior temporal might most truly reveal neuroanatomical correlates of Merge during the
cortices were not found to be active, which led the authors to conclude comparison between the syntactic and the non-syntactic condition. In-
that these areas are thus not reliably involved during sentence com- terestingly, this subclass of studies rather reports a selective involve-
prehension. Within the word-lists control condition set, they reported ment of the inferior frontal regions, therefore suggesting syntax-sensi-
activity in the temporal cortex—the anterior/superior temporal gyrus tivity in the area (Bornkessel and Schlesewsky, 2006; Friederici, 2011;
(BA38/22), the middle temporal gyrus (BA21), the posterior/superior Hagoort, 2005). This may in turn reflect functional dissociation be-
temporal gyrus (BA21/22)—in the inferior frontal cortex—pars trian- tween content words and function words during sentence comprehen-
gularis (BA45), pars orbitalis (BA47)—and in the frontal cortex—BA8/ sion (Bastiaansen et al., 2005; Bradley and Garrett, 1983; Brown et al.,
9. Importantly, they noted that the most dorsal part of the IFG (pars 1999; Friederici, 1985; Mohr et al., 1994; Neville et al., 1992; Osterhout
opercularis; BA44), was not reliably activated during passive listening et al., 1997; Pulvermuller, 1995; Shapiro and Jensen, 1986; Small et al.,
to simple sentences, which lead them to hypothesize that this area is 1998; Wang et al., 2008). A major role for BA44 in syntactic processing
either not involved in any sentence-level combinatorial process, or that finds strong support in artificial language studies, which show the in-
this process is not necessarily active for passive listening because of volvement of the area during the processing of hierarchical phrase-
good-enough processing strategies not requiring full compositional structure grammars (Bahlmann et al., 2008; Friederici et al., 2006a), as
analysis (Ferreira et al., 2002). well as during the processing of natural language structures, in which
Compared to label-based distribution meta-analysis, the activation the stimulus material is operationalized in terms of syntactic com-
likelihood estimation (ALE) analysis (Laird et al., 2005; Turkeltaub plexity manipulation, either by using embedding or scrambling ma-
et al., 2002) calculates the degree of agreement across studies by nipulations (Friederici et al., 2006b; Makuuchi et al., 2009; Meyer
modeling the activated foci as centers of a Gaussian probability dis- et al., 2012; Santi and Grodzinsky, 2010). At the structural level, Dif-
tribution, rather than as single points. The three-dimensional Gaussian fusion Tensor Imaging (DTI) studies further indicate that BA44 is con-
maps are then summed to one other following several thousand of nected to the posterior temporal region (pSTS/STG) via long-range fiber
random iterations, whose final output would correspond to p-value- bundles—anatomically identified with the arcuate fascicle/superior

647
E. Zaccarella et al. Neuroscience and Biobehavioral Reviews 80 (2017) 646–656

Table 1
Summary of studies included in the meta–analysis.

Study Subjects (N) Word-list type Peaks (N)

Bottini et al. (1994) 6 content words and function words 16

Stowe et al. (1998) 12 content words and function words 2
Stowe et al. (1999) 12 content words and function words 1
Wong et al. (1999) 5 content words 1
Kuperberg et al. (2000) 4 content words and function words 11
Homae et al. (2002) 9 content words and function words 2
Hashimoto and Sakai (2002) 16 content words and function words 5
Vanderberghe et al. (2002) 12 content words and function words 3
Indefrey et al. (2004) 18 content words 1
Maguire et al. (2004) 12 function words 2
Humphries et al. (2005) 12 content words 4
Xu et al. (2005) 22 content words and function words 9
Rogalsky et al. (2009) 14 content words 2
Snijders et al. (2009) 18 content words and function words 24
Pallier et al. (2011) 40 content words and function words 11
Goucha et al. (2015a) 22 content words and function words 8
Goucha et al. (2015b) 27 content words and function words 3
Matchin et al. 2017 16 content words and function words 7
Zaccarella et al. (2017) 18 content words 2
Total 295 114

Studies have been organized according to their sample size (number of subjects), type of word-list employed, and number of peaks. See Appendix A for exact peak locations and
anatomical regions, according to the authors of the corresponding study. A separate detailed summery of studies not included in the meta-analysis is listed in Appendix B.

longitudinal fascicle (AF/SLF)—which run dorsally along the left 1993; Uchiyama et al., 2008). Two additional studies could not be in-
hemisphere (Friederici, 2011; Wilson et al., 2011). This dorsal con- cluded in the final list because they restricted their analysis to a specific
nection has been suggested to play a central role during the evaluation region of interest only, or masked with other contrasts (Humphries
of linguistic complexity, in which BA44 acts as a higher-order syntactic et al., 2006; Jobard et al., 2007). It is indeed important to note that ALE
processor (Friederici, 2012), while the pSTS/STG integrates multiple seeks spatial convergence across whole-brain, therefore the inclusion of
sources of information, like thematic-role assignment (Den Ouden et al., the studies restricting their functional search within smaller volumes
2012) or semantic predictability (Obleser and Kotz, 2010). will increase the chance to get very large values within that region. The
Based on these above considerations, the aim of the present study pool so far, which only included studies published between 1993 and
was: (1) to measure the neuroanatomical convergence for Merge by 2014 (N = 15), was finally updated with a refined search in the new
means of the ALE algorithm, to verify which areas show consistent version of Web of Science with the search term “(fMRI OR functional
activity for the process across the studies reported in the neurolinguistic magnetic resonance imaging OR PET OR positron emission tomo-
literature; (2) to test the hypothesis that functional activity correlating graphy) AND (sentence OR phrase) AND (lists OR random)” for the
with syntactic Merge is more strongly localized in the left inferior period 2014–2017. This search yielded 19 additional studies, nine of
frontal regions (BA44)—if residual syntactic information carried by mix which were fMRI experiments with healthy subjects. Among the nine
word-lists studies is removed from the analysis—as well as in the most studies, three studies reporting contrasts between sentences/phrases
posterior temporal region; (3) if the predicted activation clusters are and word-lists could be included in the final pool (Goucha and
successfully retrieved, to evaluate the connectivity profile linking the Friederici, 2015; Matchin et al., 2017; Zaccarella et al., 2017). One
inferior frontal gyrus (IFG) to the posterior temporal region, using an additional study was excluded because it restricted the analysis to a
independent dataset as testing population. specific region of interest only (Zaccarella and Friederici, 2015). One
peak from the second of two fMRI studies reported in Goucha and
Friederici (2015) and evaluating the impact of derivational morphology
2. Material and methods (first study) and inflectional morphology (second study) on sentential
and word-list processing was removed because it also restricted the
2.1. Literature search and selection analysis to a single anatomical region. The remaining five studies from
the period 2014–2017 had been removed early on based on different
The preliminary pool of studies was based on a recent label-based exclusion criteria, which we deemed problematic for the ALE analysis
meta-analysis (Hagoort and Indefrey, 2014), which analyzed 151 he- (see Appendix B). The final pool consisted of 19 studies with a total
modynamic experiments on sentence processing. A subset of the 85 number of 114 foci and a total number of 295 subjects (see Table 1; see
studies contrasting sentence-level processing with lower-level controls, also Appendices A and B for a detailed description).
or syntactically more complex sentences with syntactically less de-
manding sentences had been already included in previous meta-ana-
lyses (Indefrey 2011, 2012). A second subset of 66 studies was gener- 2.2. Activation likelihood estimation (ALE)
ated by conducting a search in Thomson Reuters Web of Knowledge
with the search term “fMRI OR functional magnetic resonance imaging The coordinates of those studies reporting activation in Talairach
OR PET OR positron emission tomography AND semantic AND sen- space (Talairach and Tournoux, 1988) were transformed to match the
tence”. Within this pool, the authors isolated 15 studies, which directly MNI space using the convert foci function included in the GingerALE
contrasted sentential stimuli against word-list strings. Five additional toolbox (Eickhoff et al., 2012; Eickhoff et al., 2009; Turkeltaub et al.,
studies contrasting sentential stimuli against word-list strings, which 2012), which can be found at brainmap.org/ale. The GingerALE version
were not included in this group of studies, but which the authors re- 2.3.6 was employed to run the ALE algorithm (Eickhoff et al., 2017). To
ported within their meta-analysis for other contrasts were also added to perform the ALE calculation, the more conservative Turkeltaub ALE
the list. Within the list, three studies had to be excluded because no Method was selected to correct for within-experiment effects derived
coordinate peaks were provided (Friederici et al., 2000; Mazoyer et al., from foci proximity (Turkeltaub et al., 2012). Probability distributions

648
E. Zaccarella et al. Neuroscience and Biobehavioral Reviews 80 (2017) 646–656

were modeled as Gaussian function with an automatically calculated 2.5. DTI diffusion-weighted MRI data analysis
median smoothing filter of about 9 mm (FWHM). A standard threshold
of P < 0.001 (cluster-forming threshold) and 0.05 for cluster-level The diffusion MRI data were analyzed using LIPSIA (Max Planck
inference was applied to the output images, with 10000 thresholding Institute for Human Cognitive and Brain Sciences, Leipzig, Germany)
permutations. Labels for the resulting maxima were automatically as- and FSL (FMRIB, University of Oxford, UK). The T1-weighted structural
signed according to the Talairach Daemon Toolbox included in the scans were skull-stripped and the brain images were aligned with the
software (Lancaster et al., 1997; Lancaster et al., 2000). Three different mid-sagittal plane and the line connecting the anterior and posterior
ALE analyses were performed: (1) in Analysis 1, all foci were given as commissure (AC-PC line) following the Talairach Atlas (Talairach and
dataset input, to observe the overall pattern of activation for normal Tournoux, 1988). The 7 images without diffusion weighting were used
licit structures vs. word-lists irrespective of the type of word-list em- to estimate motion correction parameters using rigid-body transfor-
ployed; (2) In Analysis 2, two sub-datasets of foci were created ac- mations (Jenkinson et al., 2002), implemented in FSL. Motion correc-
cording to the type of the word-list that each corresponding study tion parameters were interpolated for all 67 vol and combined with a
employed: Content AND Function word-lists consisted of 13 experi- global registration to the T1 anatomy computed with the same method.
ments (number of foci: 102; number of subjects: 216); Content OR The gradient direction for each volume was corrected using the rotation
Function word-lists consisted of 6 experiments (number of foci: 12; parameters. The registered images were interpolated to the new re-
number of subjects: 79); (3) in Analysis 3, a conjunction analysis was ference frame with an isotropic voxel resolution of 1 mm and ad-
performed to show voxel overlapping between the two sub-datasets. ditionally to the native 1.72 mm resolution for tractography. Finally,
for each voxel, a diffusion tensor was fitted to the data and the Frac-
tional Anysotropy (FA) was computed (Basser and Pierpaoli, 1996). For
2.3. Mass overlap analysis in the inferior frontal gyrus
each subject, the FA image was then normalized to a standard FA
template provided with FSL (FMRIB58_FA) with a resolution of
On the basis of the two functional convergence maps—Content AND
1 × 1 × 1 mm using a non-linear registration. Anatomical connectivity
Function word-lists; Content OR Function word-lists—that were ob-
between the functionally activated areas for the Content OR Function
tained from Analysis 2, we performed a mass overlap analysis in the
word-lists analysis was investigated with probabilistic tractography
IFG, to explore sub-regional spatial distribution for the two functional
using FSL to generate multiple streamlines passing through the pair of
clusters in BA44 and BA45. Areas 44 and 45 from the Anatomy Toolbox
activated regions. The two seeds of interest for the probabilistic trac-
served as independent anatomical mask for the overlap analysis
tography were mapped as single points in the MNI template and then
(Amunts et al., 1999; Eickhoff et al., 2005). The two anatomical regions
transformed into 3 mm spherical regions-of-interest (ROIs). To register
were first resliced to a 2 × 2 × 2 voxel size resolution, to match the
the ROIs to each subject’s native space, the inverse of the non-linear
convergence maps’ voxel size from the GingerALE Toolbox, and then
mapping between the FA template in MNI space and FA maps was
added together to form a combined Broca’s binary image. Each func-
calculated and the ROIs were warped into the native space of each
tional cluster was in turn first multiplied with Broca’s mask to calculate
subject. Each subject’s specific probabilistic tractography map was
the relative proportion of mass falling within the mask, and then
scaled with a logarithmic function to normalize the values, back-
multiplied with each anatomical region separately, to calculate the
transformed into the standard anatomical space and merged across
percent of mass falling either in BA44 or in BA45.
subjects. A mean tractography map was extracted and displayed as iso-
surface of the pathway (arbitrary iso-value of 0.30) together with the
2.4. DTI diffusion-weighted MRI acquisition parameters MNI brain template using brainGL (code.google.com/p/braingl/).

Diffusion-weighted data from twenty-two right-handed subjects 3. Results

(Oldfield, 1971) were included in the analysis (11 female; mean age
28.5 years, standard deviation (SD) 3.62 years; all native German 3.1. ALE results: functional convergence for Merge
speakers). They belong to the same population set included in a func-
tional experiment on syntactic processing as reported in Zaccarella and The ALE meta-analysis of all studies pooled together revealed four
Friederici (2015). Diffusion-weighted images were acquired on a significant clusters within the left hemispherical cortex (Table 2 and
whole-body 3-T Siemens TIM Trio Scanner (Siemens Healthcare, Er- Fig. 1). The first cluster comprised the pars opercularis (BA44) and the
langen, Germany) using a 32-channel phased-array head coil. A twice- pars trinagularis (BA45) extending dorso-medially to BA46 and in-
refocused spin echo EPI sequence was used ((Reese et al., 2003); feriorly to BA47. The second cluster encompassed the inferior part of
TE = 100 ms, TR = 12 s, 128 × 128 image matrix, the posterior superior temporal sulcus/gyrus (pSTS/STG; BA22) and the
FOV = 220 × 220 mm2, 88 axial slices (no gap), resolution: angular gyrus (AG; BA39). Two smaller clusters were reported in the
1.72 × 1.72 × 1.7 mm). Additionally, fat saturation was employed middle/anterior temporal gyrus (MTG; BA21/BA22) and in the tem-
together with 6/8 partial Fourier imaging and generalized auto-cali- poral pole (TP; BA38). When the two subsets were analyzed as in-
brating partially parallel acquisitions (GRAPPA, acceleration dependent samples, Content AND Function word-lists revealed pattern
factor = 2; (Griswold et al., 2002)). Diffusion weighting was iso- of activity which looked similar to the ones reported for the overall
tropically distributed along 60 diffusion-encoding gradient directions analysis, although the activity in the frontal region was now confined to
with a b-value of 1000 s/mm2. Additionally, seven datasets with no BA45/BA46/BA47, with almost no involvement of BA44. Moreover, no
diffusion weighting (b0) were acquired initially and interleaved after convergence cluster was found in the temporal pole (Table 3 and
each block of 10 diffusion-weighted images as anatomical reference for Fig. 2). Content OR Function word-lists conversely resulted in two very
off-line motion correction. The diffusion MRI sequence lasted ∼16 min. specific clusters of activity (Table 4 and Fig. 2). The first cluster was
T1-weighted 3D MP-RAGE (magnetization-prepared rapid gradient strongly limited to BA44, while the second cluster was found in the
echo) images ((Mugler and Brookeman, 1990); TI = 650 ms, most inferior part of the posterior superior temporal sulcus, extending
TR = 1300 ms, alpha = 10°, FOV = 256 × 240 mm, resolution anteriorly to the posterior superior temporal gyrus (BA22). Finally, the
1 × 1 × 1 mm) were previously acquired with a non-slice-selective conjunction analysis revealed a total amount of 39 overlapping voxels
inversion pulse. The 3D anatomical images were used for alignment and between the two subsets: the largest amount of overlapping voxels (31
preprocessing of the diffusion data. Written informed consent was ob- voxels, 79.48%) was located in the posterior superior temporal gyrus
tained from all subjects, according to the ethical approval of the Uni- (BA22), the remaining 8 voxels (8 voxels, 20.51%) were located in the
versity of Leipzig. inferior frontal gyrus, in BA44 (Table 5).

649
E. Zaccarella et al. Neuroscience and Biobehavioral Reviews 80 (2017) 646–656

Table 2 Table 3
Analysis 1: Licit structures vs. word-list controls. Analysis 2: Content AND Function word-list studies.

Cluster BA MNI Coordinates (mm) ALE (× 10−2) Cluster Size (mm3) Cluster BA MNI Coordinates (mm) ALE (x 10−2) Cluster Size (mm3)

x y z x y z

1 47 −44 34 −8 2.80 7080 1 47 −46 34 −6 2.65 5384

45 −52 28 10 2.40 45 −52 28 10 2.40
44 −50 12 16 1.73 46 −42 22 20 1.24
46 −46 22 22 1.41 44 −50 12 16 0.95
46 −42 22 20 1.40
2 21 −54 −4 −22 1.69 1792
2 22 −58 −42 0 2.32 3992 22 −54 −10 −12 1.47
22 −44 −56 18 2.04
3 22 −44 −56 18 2.04 1752
39 −58 −56 10 1.27
39 −58 −56 10 1.20
3 21 −54 −4 −22 1.69 1704
4 21 −60 −42 −2 1.73 1512
22 −54 −10 −12 1.47
21 −48 −44 2 1.02
4 38 −48 16 −26 2.04 1032
Functional clusters obtained from Analysis 2, showing significant neuroanatomical con-
Functional clusters obtained from Analysis 1, showing significant neuroanatomical con- vergence when Content AND Function word-lists studies alone were given as input to the
vergence when all studies were given as input to the ALE algorithm. Mask ALE algorithm. Mask Dimensions = 77 × 96 × 79; Number of within-brain
Dimensions = 77 × 96 × 79; Number of within-brain voxels = 229,781; Number of voxels = 229,781; Number of foci = 102; Number of experiments = 13; Total number of
foci = 114; Number of experiments = 19; Total number of subjects = 295; Maximum subjects = 216; Maximum ALE score (× 10−2) = 2.65; Threshold Method = Cluster-
ALE score (× 10−2) = 2.80; Threshold Method = Cluster-level Inference; Thresholding level Inference; Thresholding Value = 0.05; Thresholding Permutations = 10,000;
Value = 0.05; Thresholding Permutations = 10,000; Cluster-Forming Value = 0.001. Cluster-Forming Value = 0.001.

3.2. Mass overlap analysis in the inferior frontal gyrus (Table 3). The proportion of mass falling in Broca’s mask for the Con-
tent OR Function convergence map was 88.23% (Table 4). To note, the
The proportion of mass falling in Broca’s mask for the Content AND very few remaining voxels outside the area were located just along the
Function convergence map was 41.90% of the whole activation mass border of the mask towards the insular complex. In the following step,
found in the inferior frontal cortex, which extended to BA46 and BA47 when the BA44 and BA45 were used as separate anatomical masks, the

Fig. 1. Licit structures vs. word-list controls.

Functional clusters from Analysis 1 showing significant neuroanatomical convergence when all studies were given as input into the ALE algorithm. Sagittal slices taken form a template
brain (Colin27_T1_seg_MNI) ranging from x = −58 to x = −44. Brodmann Area (BA) 44/45/47; posterior superior temporal sulcus/gyrus (pSTS/STG; BA22); middle temporal gyrus
(MTG; BA21); temporal pole (TP; BA38).

650
E. Zaccarella et al. Neuroscience and Biobehavioral Reviews 80 (2017) 646–656

Fig. 2. Word-list subsets functional profiles.

Functional clusters from Analysis 2 with subset split based on the type of word-list control condition used in the corresponding experiment (Content AND Function, green-to-yellow;
Content OR Function, blue-to-turquoise). (A) Sagittal plane at x = –52. (B) Transverse plane at z = 12. (C) Sagittal plane at x = −58 showing functional overlap in the pSTS/STG. In the
orange rectangle, mass overlap of the two convergence maps together with BA45 (red) and BA44 (yellow) anatomical masks from the Anatomy Toolbox (depicted in D). (E) Bar graph
representing mass overlap proportions (%) of the two convergence maps over BA45 (red) and BA44 (yellow). (F) Transverse plane at z = 12 with mass overlap of the two convergence
maps together with BA45 (red) and BA44 (yellow) in the orange rectangle. Inferior Frontal Gyrus (IFG); Brodmann Area (BA) 44/45; posterior superior temporal sulcus/gyrus (pSTS/STG;
BA22). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Table 4 Table 5
Analysis 2: Content OR Function word-list studies. Analysis 3: Conjunction Analysis.

Cluster BA MNI Coordinates (mm) ALE (× 10−2) Cluster Size (mm3) Cluster BA MNI Coordinates (mm) Cluster Size N Voxels
(mm3) (2 × 2 × 2)
x y z x y z

1 44 −60 14 12 0.84 680 1 22 −58 −44 2 248 31

44 −52 12 16 0.82 2 44 −52 12 16 64 8

2 22 −58 −44 2 0.82 616

Functional clusters obtained from Analysis 3, showing overlapping voxels between the
22 −62 −48 6 0.73
functional clusters obtained from Analysis 2 for Content AND Function and Content OR
Function word-lists studies. Mask Dimensions = 77 × 96 × 79; Number of within-brain
Functional clusters obtained from Analysis 2, showing significant neuroanatomical con-
voxels = 229,781.
vergence when Content OR Function word-lists studies alone were given as input to the
ALE algorithm. Mask Dimensions = 77 × 96 × 79; Number of within-brain
voxels = 229,781; Number of foci = 12; Number of experiments = 6; Total number of 3.3. DTI results: structural connections for Merge
subjects = 79; Maximum ALE score (× 10−2) = 0.84; Threshold Method = Cluster-level
Inference; Thresholding Value = 0.05; Thresholding Permutations = 10,000; Cluster- The analysis of the diffusion tensor imaging data for the twenty-two
Forming Value = 0.001.
subjects revealed a dorsal connection along the AF/SLF between the
two local maxima in BA44 and in the pSTS/STG, which were retrieved
amount of mass from the Content AND Function map almost entirely during the subset (Content OR Function) analysis (Fig. 3).
fell in BA45 (92.55%). On the contrary, the total amount of mass from
the Content OR Function map exclusively fell in BA44 (100% of the
amount of mass falling in Broca’s mask). See Fig. 2C–F. 4. Discussion

Our meta-analysis set out to identify across the neurolinguistic lit-

erature the neuroanatomical regions showing consistent activation for
Merge, the most fundamental syntactic computation of human

651
E. Zaccarella et al. Neuroscience and Biobehavioral Reviews 80 (2017) 646–656

Fig. 3. Probabilistic tractography between BA44 and pSTS/STG.

Dorsal connections (orange) between two local maxima from the
Content OR Function subset analysis, used as spherical seeding points
in BA44 (−52, 12, 16) and in the posterior temporal cortex (−62,
−48, 6). Brodmann Area (BA) 44/45; posterior superior temporal
sulcus/gyrus (pSTS/STG; BA22). Arcuate fascile (AF)/superior long-
itudinal fascile (SLF). Anatomical template: MNI152_T1_2mm_brain.

language. The ALE method revealed that within the left hemisphere, function of structural hierarchy regardless of the structural complexity
BA44 and the pSTS/STG appear to be the two regions showing reliable itself (Zaccarella and Friederici, 2015). In this study the authors found
functional activity during the comparison between normal linguistic activity in the area already when participants were asked to process
sequences and word-list control sequences. Because of the very basic two-word basic syntactic phrases in which complexity was broken
nature of the process, regional selectivity for Merge in the IFG appears down to one single Merge application. It therefore appears that in the
to be strongly sensitive to the presence of unwanted syntactic proces- human brain BA44 corresponds to the neuroanatomical implementation
sing in the word-list sequences, since Merge-related activity in BA44 is of the Merge mechanism—by which structures are built together on the
only strongly visible when chunking possibilities in the word-list con- basis of abstract syntactic rules. This finding—we claim—represents a
trols are completely removed. This suggests that BA44 might be the crucial step forward, towards the understanding of the neuro-cognitive
computational core of the syntactic Merge mechanism in the human basis upon which our linguistic competence might be cortically re-
brain. Probabilistic tractography further confirmed that BA44 interacts presented (Bolhuis et al., 2014; Ding et al., 2016).
with the posterior temporal cortex (pSTS/STG) via dorsal fiber fascicles Compared to BA44, its neighboring region BA45 is found specifi-
within the left hemisphere. cally active among the studies using mixed word-list controls, whose
Specifically concerning the role of BA44 in syntactic processing, the chunking possibilities most probably cancelled out syntactic activity
subset analysis performed in this study shows that whenever the stimuli from the resulting subtraction analysis. Under this account, one plau-
are built in such a way that word-list control conditions consist of both sible explanation is that BA45 might rather be involved during the
content words (e.g. nouns, verbs, adjectives) and function words (e.g. processing of semantic information, as often proposed in the literature
determiners, prepositions, conjunctions), then no activation is found in (Bookheimer, 2002; Friederici, 2011; Hagoort, 2005; Hagoort and
the area (Homae et al., 2002; Humphries et al., 2005; Vandenberghe Indefrey, 2014; Hoen et al., 2006). A wide range of semantic processes
et al., 2002). Rather, the more anterior part of the IFG (BA45) seems to associated with the region includes for example propositional evalua-
be particularly engaged, with just an elusive cluster of bordering voxels tion (Dien et al., 2008; Hagoort, 2005; Zhu et al., 2013), plausibility
extending into BA44. Conversely, BA44 is strongly active, when either judgment (Caplan et al., 2008), semantic relatedness (Newman et al.,
content-only or function-only class lists are used and no Merge com- 2010) or semantic acceptability (Zhu et al., 2009). In contrast to BA45,
putation is possible (Maguire and Frith, 2004; Zaccarella et al., 2017). the posterior temporal region—which has been found here as active in
Hence, the use of both word classes in the word-lists appears to enhance both subsets of studies—is generally suggested to be an integration area
rather than inhibit the construction of minimal structures in the so- in which syntactic and thematic-based information are integrated
called “non-syntactic” conditions. From this, it follows that in the (Friederici, 2011; Grodzinsky and Amunts, 2006). Interestingly, pre-
syntax minus mixed word-list subtraction analysis, consistent syntactic vious investigations report the posterior superior temporal cortex to be
activity in BA44 is removed, because of the presence of syntactic pro- active when lexical-semantic information is available in the sentence
cessing in the word-list condition containing members of the two word (Bornkessel et al., 2005; Friederici et al., 2009; Stowe et al., 1998), or
classes as well. Crucially, a strong involvement of BA44 for the con- when thematic assignment at the sentential level is harder to evaluate
struction of syntactic structures comes from several neuroimaging stu- (Ben-Shachar et al., 2003; Bornkessel et al., 2005; Constable et al.,
dies linking the region to the cognitive concept of structural complexity 2004; Friederici et al., 2009; Roder et al., 2002). More recently, the
(Friederici, 2002; Grodzinsky and Santi, 2008; Hagoort, 2005, 2008; notion of thematic assignment in the region has been extended below
Perani et al., 2011), in which the build up of complex syntactic struc- sentential level, to include prepositional constructions in which the
tures is operationalized in different ways either as being deviant from functional role expressed by the preposition has to be integrated with
default sentence ordering (Friederici et al., 2006b) or by being multiply the noun fulfilling that role (Zaccarella et al., 2017). Because of the
hierarchically embedded, in both natural language (Makuuchi et al., integrative nature of the area, it is plausible to propose that the pSTS/
2009), and in artificial grammar studies (Bahlmann et al., 2008). This STG in language works in tide interplay with BA44 to map lexical/
observation reinforces the conclusion put forward in a word-list thematic information to syntactic argument hierarchies, such that the
study—removed from the final ALE list because it only restricted the BA44 works as pure syntactic merging processor, while the pSTS/STG
analysis to the posterior IFG—which argued that BA44 activates as a acts as integrative device (Bornkessel et al., 2005; den Ouden et al.,

652
E. Zaccarella et al. Neuroscience and Biobehavioral Reviews 80 (2017) 646–656

2012). This functional interaction is strongly confirmed by the struc- Considering these past findings, a crucial suggestion of our meta-ana-
tural data reported here, which reveal dorsal connections between the lysis is that given the functional differences between the two classes,
two areas along the AF/SLF when local activation maxima are taken as within-class word-lists (Content OR Function) rather than mixed word-
anatomical seeds for probabilistic tractography along the hemisphere lists (Content AND Function) seem to be an ideal solution to keep
(Friederici and Gierhan, 2013; Saur et al., 2008). syntactic processing in the control condition as low as possible. As such,
Finally, this meta-analysis also questions the role of the anterior content-only word-lists have very reduced syntax, because functional
temporal lobe in syntactic processing, and thereby neurolinguistic lexicon has been removed (Friederici et al., 2000; Indefrey et al., 2004).
models which suggested the area to be specifically responsible for Similarly, function word-only lists appear to have very reduced syntax,
structure building during natural language comprehension (Humphries because they loose their syntactic triggering when a proper surrounding
et al., 2005; Rogalsky and Hickok, 2009). According to our subset context is missing (Maguire and Frith, 2004).
analysis, however we could not reveal any convergent activity in the Concluding, the present ALE meta-analysis found that BA44 to-
TP. If syntactic information were indeed effectively processed in the gether with the pSTS/STG are the two regions to be reliably involved in
area one would have expected TP activity to show convergence across Merge processing once confounding syntactic effects in the word-list
the studies. A semantic compositional role for the region rather appears controls are removed. Probabilistic tractography further confirms
to be likely since it is consistent with recent findings within the neu- dorsal structural fiber bundles connecting the two regions. Within this
rolinguistic literature, at multiple levels. A series of recent magne- network, it appears that BA44 works as pure syntactic merger where
toencephalography (MEG) experiments looking at conceptual compo- words are combined together to form hierarchies based on their cor-
sitional effects in phrasal constructions, consistently reported the TP to responding syntactic information (Zaccarella and Friederici, 2015),
be active during the construction of more complex semantic re- while pSTS/STG maps the lexical/thematic information to syntactic
presentations, when concepts had to be combined together (Bemis and hierarchies built by BA44 (den Ouden et al., 2012). On this account, the
Pylkkanen, 2011, 2013; Del Prato and Pylkkanen, 2014). Moreover, an discovery that highly convergent locations of activation for structure-
influential study investigating the relation between the length of con- building processing in BA44 and in the pSTS/STG exist across the data
stituent size and the increase of correlating neural activity found that recapitulates previous single studies investigating Merge at different
the TP activated only when lexico-semantic information was available. levels of syntactic complexity and spanning from two-word phrases to
(Pallier et al., 2011). In two other fMRI studies, Baron and Osherson longer sentential stimuli (Goucha and Friederici, 2015; Zaccarella and
(2011) and Baron et al. (2010) showed that the neural representation of Friederici, 2015; Zaccarella et al., 2017). In this sense, the possibility to
complex concepts in the left anterior temporal lobe is additive, such collapse together and statistically evaluate the convergence of neuronal
that the superimposition of the voxel-wise neural representation of activity from separate findings in meta-analytic fashion enables the
simple concepts (e.g., young + man) correlates with the neural re- identification of a shared cortical localization for Merge across in-
presentation of complex concepts in the TP formed by adding the two to dependent population sets and stimulus complexity, and consequently,
each other (e.g., young man). Clinically, the syndrome of semantic it strengthens the neurobiological hypothesis of a universal rule-based
variant primary progressive aphasia (semantic PPA; also known as se- combinatorial system processing linguistic strings in a hierarchical
mantic dementia) also poses a serious challenge to the view of TP being fashion through infero-fronto-temporal dorsal connections (Goucha
involved during the evaluation of syntactic information (Hodges et al., et al., 2017). As such, this meta-analysis is positioned at the interface
1992; Leyton et al., 2011; Snowden et al., 1989; Warrington, 1975). between linguistic theory and the neurobiology of language, to in-
Semantic dementia is indeed characterized by bilateral atrophy of the vestigate how a fundamental component of our linguistic knowledge
TP with profound lexical semantic deficits, but spared syntactic func- maps onto neural structures of the human brain. The neuroanatomical
tionality (Wilson et al., 2012), to include normal sensitivity to syntactic description of Merge reported here strongly points towards its cortical
violations in on-line and off-line tasks (Cotelli et al., 2007; Grossman reality at the neural level, and it therefore constitutes—we believe—a
et al., 2005). At a more general level, strong support to the view that TP critical point of reference for future studies aiming to uncover its
is specifically engaged during semantic processing can be traditionally phylogenic development (Berwick et al., 2013; Dehaene et al., 2015;
found within the semantic memory literature, particularly within the Goucha et al., 2017), its ontogeny (Roy et al., 2015) as well as its formal
distributed-plus-hub models of neuroanatomical distribution of the subroutines (Murphy, 2015; Zaccarella and Friederici, 2016).
cortical semantic network, which sees the TP as a highly integrative
amodal area across different modality-specific systems (Lambon Ralph Conflicts of interest
et al., 2010; Patterson et al., 2007; Visser et al., 2010).
A final remark concerns with the cognitive status of function and None of the authors has conflicts of interest to declare.
content words in linguistic representations, in the light of the present
findings. From a psycholinguistic point of view, support to their dis- Acknowledgments
tinction is long known from both clinical (Berndt and Caramazza, 1980;
Friederici, 1981; Goldstein, 1948; Luria, 1970; Pulvermuller, 1995; We are indebted to Alfred Anwander for his help during the analysis
Small et al., 1998; Swinney et al., 1980; Weisenberg and McBride, of the DTI data and for his comments on earlier versions of the
1935) and imaging studies (Kutas and Hillyard, 1980; Neville et al., manuscript. We apologize to all authors whose eligible articles we
1992; Nobre and McCarthy, 1994; Pulvermuller et al., 1995; ter Keurs might have missed. This work was supported by a grant of the European
et al., 1999; Van Petten and Kutas, 1991; Wang et al., 2008). More Research Council (ERC-2010-AdG20100407 awarded to ADF).
recently, a MEG study investigated the neural modulation generated by
two types of phrases at two-word level and found that only phrases Appendix A
enhancing conceptual combination, which were formed by a content
word (adjective) + noun (red cups), increased activity in the TP, while Summary of studies included in the meta–analysis with peak loca-
phrases formed by a function word (numeral) + noun (two cups) did tion and anatomical region.
not (Del Prato and Pylkkanen, 2014). This is interesting in the light of Studies have been organized according to their sample size (number
one of the studies included in this meta-analysis which reported that of subjects), type of word-list employed, peak location (3D coordinates)
simple prepositional phrases—formed by a function word (preposition, and anatomical regions, according to the authors of the corresponding
on) and a determiner phrase (the ship)—increased activity in BA44, study. Coordinates showing decimal digits have been converted from
whereas simple sentences—formed by the same determiner phrase and the original Talairach space into the MNI space using the convert foci
a content word (verb, sinks)—involved BA45 (Zaccarella et al., 2017). function included in the GingerALE Toolbox.

653
E. Zaccarella et al. Neuroscience and Biobehavioral Reviews 80 (2017) 646–656

Appendix B Caplan, D., Chen, E., Waters, G., 2008. Task-dependent and task-independent neurovas-
cular responses to syntactic processing. Cortex 44, 257–275.
Chomsky, N., 1995. The Minimalist Program. MIT Press, Cambridge.
Summary of studies not included in the meta–analysis with peak Constable, R.T., Pugh, K.R., Berroya, E., Mencl, W.E., Westerveld, M., Ni, W.,
location and anatomical region. * = Coordinates not reported (not in- Shankweiler, D., 2004. Sentence complexity and input modality effects in sentence
comprehension: an fMRI study. Neuroimage 22, 11–21.
cluded in the final pool) ** = Small Volume Correction Analysis/ Cotelli, M., Borroni, B., Manenti, R., Ginex, V., Calabria, M., Moro, A., Alberici, A.,
Masked contrast (not included in the final pool). Goucha et al. (2015) Zanetti, M., Zanetti, O., Cappa, S.F., Padovani, A., 2007. Universal grammar in the
consist of two different experiments included within the same published frontotemporal dementia spectrum: evidence of a selective disorder in the cortico-
basal degeneration syndrome. Neuropsychologia 45, 3015–3023.
article, plus a small-volume correction analysis with a single peak, Dehaene, S., Meyniel, F., Wacongne, C., Wang, L., Pallier, C., 2015. The neural re-
which could not be included in the final pool, and therefore listed in the presentation of sequences: from transition probabilities to algebraic patterns and
present table. Five studies from the period 2014–2017 had been re- linguistic trees. Neuron 88, 2–19.
Del Prato, P., Pylkkanen, L., 2014. MEG evidence for conceptual combination but not
moved early on based on different exclusion criteria, which we deemed
numeral quantification in the left anterior temporal lobe during language production.
problematic for the ALE analysis: reduced brain coverage due to in- Front. Psychol. 5, 524.
creased acquisition time at high-resolution 7T scanner (Mellem et al., den Ouden, D.B., Saur, D., Mader, W., Schelter, B., Lukic, S., Wali, E., Timmer, J.,
2016); functional analysis restricted to the prediction phase which oc- Thompson, C.K., 2012. Network modulation during complex syntactic processing.
Neuroimage 59, 815–823.
curred only after sentential or word-list processing was performed Dien, J., Franklin, M.S., Michelson, C.A., Lemen, L.C., Adams, C.L., Kiehl, K.A., 2008.
(Bonhage et al., 2015); concerns with the design manipulation for the fMRI characterization of the language formulation area. Brain Res. 1229, 179–192.
contrast sentences vs. word-lists processing during a picture description Ding, N., Melloni, L., Zhang, H., Tian, X., Poeppel, D., 2016. Cortical tracking of hier-
archical linguistic structures in connected speech. Nat. Neurosci. 19, 158–164.
task, which only revealed activity in regions associated with visual at- Eickhoff, S.B., Stephan, K.E., Mohlberg, H., Grefkes, C., Fink, G.R., Amunts, K., Zilles, K.,
tention, and no involvement of language-related areas (Matchin and 2005. A new SPM toolbox for combining probabilistic cytoarchitectonic maps and
Hickok, 2016); covert sentences vs. word-lists (months of the year) functional imaging data. Neuroimage 25, 1325–1335.
Eickhoff, S.B., Laird, A.R., Grefkes, C., Wang, L.E., Zilles, K., Fox, P.T., 2009. Coordinate-
generation task used to assess individual hemispheric lateralization based activation likelihood estimation meta-analysis of neuroimaging data: a
index for large datasets, for which no group-averaged coordinates were random-effects approach based on empirical estimates of spatial uncertainty. Hum.
reported (Mazoyer et al., 2014; Tzourio-Mazoyer et al., 2016). Brain Mapp. 30, 2907–2926.
Eickhoff, S.B., Bzdok, D., Laird, A.R., Kurth, F., Fox, P.T., 2012. Activation likelihood
estimation meta-analysis revisited. Neuroimage 59, 2349–2361.
References Eickhoff, S.B., Laird, A.R., Fox, P.M., Lancaster, J.L., Fox, P.T., 2017. Implementation
errors in the GingerALE Software: description and recommendations. Hum. Brain
Mapp. 38, 7–11.
Amunts, K., Schleicher, A., Burgel, U., Mohlberg, H., Uylings, H.B., Zilles, K., 1999.
Ferreira, F., Bailey, K.G., Ferraro, V., 2002. Good-enough representations in language
Broca’s region revisited: cytoarchitecture and intersubject variability. J. Comp.
comprehension. Curr. Direct. Psychol. Sci. 11, 11–15.
Neurol. 412, 319–341.
Friederici, A.D., Gierhan, S.M., 2013. The language network. Curr. Opin. Neurobiol. 23,
Bahlmann, J., Schubotz, R.I., Friederici, A.D., 2008. Hierarchical artificial grammar
250–254.
processing engages Broca’s area. Neuroimage 42, 525–534.
Friederici, A.D., Meyer, M., von Cramon, D.Y., 2000. Auditory language comprehension:
Baker, M., 2003. Lexical Categories – Verbs, Nouns and Adjectives. Cambridge University
an event-related fMRI study on the processing of syntactic and lexical information.
Press, Cambridge.
Brain Lang. 75, 289–300.
Baron, S.G., Osherson, D., 2011. Evidence for conceptual combination in the left anterior
Friederici, A.D., Bahlmann, J., Heim, S., Schubotz, R.I., Anwander, A., 2006a. The brain
temporal lobe. Neuroimage 55, 1847–1852.
differentiates human and non-human grammars: functional localization and struc-
Baron, S.G., Thompson-Schill, S.L., Weber, M., Osherson, D., 2010. An early stage of
tural connectivity. Proc. Natl. Acad. Sci. U. S. A. 103, 2458–2463.
conceptual combination: superimposition of constituent concepts in left anterolateral
Friederici, A.D., Fiebach, C.J., Schlesewsky, M., Bornkessel, I.D., von Cramon, D.Y.,
temporal lobe. Cogn. Neurosci. 1, 44–51.
2006b. Processing linguistic complexity and grammaticality in the left frontal cortex.
Basser, P.J., Pierpaoli, C., 1996. Microstructural and physiological features of tissues
Cereb. Cortex 16, 1709–1717.
elucidated by quantitative-diffusion-tensor MRI. J. Magn. Reson. B 111, 209–219.
Friederici, A.D., Makuuchi, M., Bahlmann, J., 2009. The role of the posterior superior
Bastiaansen, M.C., van der Linden, M., Ter Keurs, M., Dijkstra, T., Hagoort, P., 2005.
temporal cortex in sentence comprehension. Neuroreport 20, 563–568.
Theta responses are involved in lexical-semantic retrieval during language proces-
Friederici, A.D., 1981. Production and comprehension of prepositions in aphasia.
sing. J. Cogn. Neurosci. 17, 530–541.
Neuropsychologia 19, 191–199.
Bemis, D.K., Pylkkanen, L., 2011. Simple composition: a magnetoencephalography in-
Friederici, A.D., 1985. Levels of processing and vocabulary types: evidence from on-line
vestigation into the comprehension of minimal linguistic phrases. J. Neurosci. 31,
comprehension in normals and agrammatics. Cognition 19, 133–166.
2801–2814.
Friederici, A.D., 2002. Towards a neural basis of auditory sentence processing. Trends
Bemis, D.K., Pylkkanen, L., 2013. Flexible composition: MEG evidence for the deployment
Cogn. Sci. 6, 78–84.
of basic combinatorial linguistic mechanisms in response to task demands. PLoS One
Friederici, A.D., 2011. The brain basis of language processing: from structure to function.
8, e73949.
Physiol. Rev. 91, 1357–1392.
Ben-Shachar, M., Hendler, T., Kahn, I., Ben-Bashat, D., Grodzinsky, Y., 2003. The neural
Friederici, A.D., 2012. The cortical language circuit: from auditory perception to sentence
reality of syntactic transformations: evidence from functional magnetic resonance
comprehension. Trends Cogn. Sci. 16, 262–268.
imaging. Psychol. Sci. 14, 433–440.
Goldstein, K., 1948. Language and Language Disturbances. Grune & Stratton, New York.
Berndt, R.S., Caramazza, A., 1980. A redefinition of the syndrome of Broca’s aphasia:
Goucha, T., Friederici, A.D., 2015. The language skeleton after dissecting meaning: a
implication for a neuropsychological model of language. Appl. Psycolinguist. 1,
functional segregation within Broca’s Area. Neuroimage 114, 294–302.
225–278.
Goucha, T., Zaccarella, E., Friederici, A.D., 2017. A revival of the Homo loquens as a
Berwick, R.C., Friederici, A.D., Chomsky, N., Bolhuis, J.J., 2013. Evolution, brain, and the
builder of labeled structures: neurocognitive considerations. Neurosci. Biobehav. Rev
nature of language. Trends Cogn. Sci. 17, 89–98.
(advance access published Mar 16, 2017).
Bolhuis, J.J., Tattersall, I., Chomsky, N., Berwick, R.C., 2014. How could language have
Griswold, M.A., Jakob, P.M., Heidemann, R.M., Nittka, M., Jellus, V., Wang, J., Kiefer, B.,
evolved? PLoS Biol. 12, e1001934.
Haase, A., 2002. Generalized autocalibrating partially parallel acquisitions
Bonhage, C.E., Mueller, J.L., Friederici, A.D., Fiebach, C.J., 2015. Combined eye tracking
(GRAPPA). Magn. Reson. Med. 47, 1202–1210.
and fMRI reveals neural basis of linguistic predictions during sentence comprehen-
Grodzinsky, Y., Amunts, K. (Eds.), 2006. Broca’s Region. OUP, New York.
sion. Cortex 68, 33–47.
Grodzinsky, Y., Friederici, A.D., 2006. Neuroimaging of syntax and syntactic processing.
Bookheimer, S., 2002. Functional MRI of language: new approaches to understanding the
Curr. Opin. Neurobiol. 16, 240–246.
cortical organization of semantic processing. Annu. Rev. Neurosci. 25, 151–188.
Grodzinsky, Y., Santi, A., 2008. The battle for Broca’s region. Trends Cogn. Sci. 12,
Bornkessel, I., Schlesewsky, M., 2006. The extended argument dependency model: a
474–480.
neurocognitive approach to sentence comprehension across languages. Psychol. Rev.
Grossman, M., Rhee, J., Moore, P., 2005. Sentence processing in frontotemporal de-
113, 787–821.
mentia. Cortex 41, 764–777.
Bornkessel, I., Zysset, S., Friederici, A.D., von Cramon, D.Y., Schlesewsky, M., 2005. Who
Hagoort, P., Indefrey, P., 2014. The neurobiology of language beyond single words. Annu.
did what to whom? The neural basis of argument hierarchies during language com-
Rev. Neurosci. 37, 347–362.
prehension. Neuroimage 26, 221–233.
Hagoort, P., 2005. On Broca, brain, and binding: a new framework. Trends Cogn. Sci. 9,
Bottini, G., Corcoran, R., Sterzi, R., Paulesu, E., Schenone, P., Scarpa, P., Frackowiak, R.S.,
416–423.
Frith, C.D., 1994. The role of the right hemisphere in the interpretation of figurative
Hagoort, P., 2008. The fractionation of spoken language understanding by measuring
aspects of language: a positron emission tomography activation study. Brain 117,
electrical and magnetic brain signals. Philos. Trans. R. Soc. Lond. B Biol. Sci. 363,
1241–1253.
1055–1069.
Bradley, D.C., Garrett, M.F., 1983. Hemisphere differences in the recognition of closed
Hahne, A., Schroger, E., Friederici, A.D., 2002. Segregating early physical and syntactic
and open class words. Neuropsychologia 21, 155–159.
processes in auditory sentence comprehension. Neuroreport 13, 305–309.
Brown, C.M., Hagoort, P., ter Keurs, M., 1999. Electrophysiological signatures of visual
Hashimoto, R., Sakai, K.L., 2002. Specialization in the left prefrontal cortex for sentence
lexical processing: open- and closed-class words. J. Cogn. Neurosci. 11, 261–281.

654
E. Zaccarella et al. Neuroscience and Biobehavioral Reviews 80 (2017) 646–656

comprehension. Neuron 35, 589–597. syntactic analysis: an fMRI study. Brain Lang. 113, 51–58.
Hodges, J.R., Patterson, K., Oxbury, S., Funnell, E., 1992. Semantic dementia. Progressive Nobre, A.C., McCarthy, G., 1994. Language-related ERPs: scalp distributions and mod-
fluent aphasia with temporal lobe atrophy. Brain 115 (Pt. 6), 1783–1806. ulation by word type and semantic priming. J. Cogn. Neurosci. 6, 233–255.
Hoen, M., Pachot-Clouard, M., Segebarth, C., Dominey, P.F., 2006. When Broca experi- Obleser, J., Kotz, S.A., 2010. Expectancy constraints in degraded speech modulate the
ences the Janus syndrome: an ER-fMRI study comparing sentence comprehension and language comprehension network. Cereb. Cortex 20, 633–640.
cognitive sequence processing. Cortex 42, 605–623. Oldfield, R.C., 1971. The assessment and analysis of handedness: the Edinburgh in-
Homae, F., Hashimoto, R., Nakajima, K., Miyashita, Y., Sakai, K.L., 2002. From perception ventory. Neuropsychologia 9, 97–113.
to sentence comprehension: the convergence of auditory and visual information of Osterhout, L., Bersick, M., McKinnon, R., 1997. Brain potentials elicited by words: word
language in the left inferior frontal cortex. Neuroimage 16, 883–900. length and frequency predict the latency of an early negativity. Biol. Psychol. 46,
Humphries, C., Love, T., Swinney, D., Hickok, G., 2005. Response of anterior temporal 143–168.
cortex to syntactic and prosodic manipulations during sentence processing. Hum. Pallier, C., Devauchelle, A.D., Dehaene, S., 2011. Cortical representation of the con-
Brain Mapp. 26, 128–138. stituent structure of sentences. Proc. Natl. Acad. Sci. U. S. A. 108, 2522–2527.
Humphries, C., Binder, J.R., Medler, D.A., Liebenthal, E., 2006. Syntactic and semantic Patterson, K., Nestor, P.J., Rogers, T.T., 2007. Where do you know what you know?: The
modulation of neural activity during auditory sentence comprehension. J. Cogn. representation of semantic knowledge in the human brain. Nat. Rev. Neurosci. 8,
Neurosci. 18, 665–679. 976–987.
Indefrey, P., Hellwig, F., Herzog, H., Seitz, R.J., Hagoort, P., 2004. Neural responses to the Perani, D., Saccuman, M.C., Scifo, P., Anwander, A., Spada, D., Baldoli, C., Poloniato, A.,
production and comprehension of syntax in identical utterances. Brain Lang. 89, Lohmann, G., Friederici, A.D., 2011. Neural language networks at birth. Proc. Natl.
312–319. Acad. Sci. U. S. A. 108, 16056–16061.
Indefrey, P., 2011. Neurobiology of syntax. In: Hogan, P.C. (Ed.), The Cambridge Pulvermuller, F., Lutzenberger, W., Birbaumer, N., 1995. Electrocortical distinction of
Encyclopedia of the Language Sciences. Cambridge University Press, Cambridge, UK/ vocabulary types. Electroencephalogr. Clin. Neurophysiol. 94, 357–370.
New York, pp. 835–838. Pulvermuller, F., 1995. Agrammatism: behavioral description and neurobiological ex-
Indefrey, P., 2012. Hemodynamic studies of syntactic processing. In: Faust, M. (Ed.), The planation. J. Cogn. Neurosci. 7, 165–181.
Handbook of the Neuropsychology of Language. Blackwell, Malden, MA, pp. Reese, T.G., Heid, O., Weisskoff, R.M., Wedeen, V.J., 2003. Reduction of eddy-current-
209–228. induced distortion in diffusion MRI using a twice-refocused spin echo. Magn. Reson.
Jenkinson, M., Bannister, P., Brady, M., Smith, S., 2002. Improved optimization for the Med. 49, 177–182.
robust and accurate linear registration and motion correction of brain images. Rizzi, L., 2012. Core linguistic computations: how are they expressed in the mind/brain?
Neuroimage 17, 825–841. J. Neurolinguist. 25, 489–499.
Jobard, G., Vigneau, M., Mazoyer, B., Tzourio-Mazoyer, N., 2007. Impact of modality and Roder, B., Stock, O., Neville, H., Bien, S., Rosler, F., 2002. Brain activation modulated by
linguistic complexity during reading and listening tasks. Neuroimage 34, 784–800. the comprehension of normal and pseudo-word sentences of different processing
Kuperberg, G.R., McGuire, P.K., Bullmore, E.T., Brammer, M.J., Rabe-Hesketh, S., Wright, demands: a functional magnetic resonance imaging study. Neuroimage 15,
I.C., Lythgoe, D.J., Williams, S.C., David, A.S., 2000. Common and distinct neural 1003–1014.
substrates for pragmatic, semantic, and syntactic processing of spoken sentences: an Rogalsky, C., Hickok, G., 2009. Selective attention to semantic and syntactic features
fMRI study. J. Cogn. Neurosci. 12, 321–341. modulates sentence processing networks in anterior temporal cortex. Cereb. Cortex
Kutas, M., Hillyard, S.A., 1980. Reading senseless sentences: brain potentials reflect se- 19, 786–796.
mantic incongruity. Science 207, 203–205. Roy, I., Copley, B., Mccune, L., 2015. The emergence of Merge and recursion in the
Laird, A.R., Fox, P.M., Price, C.J., Glahn, D.C., Uecker, A.M., Lancaster, J.L., Turkeltaub, transition from single words to sentences. In: Paper Presented at the 12th Generative
P.E., Kochunov, P., Fox, P.T., 2005. ALE meta-analysis: controlling the false discovery Approaches to Language Acquisition. Nantes, France.
rate and performing statistical contrasts. Hum. Brain Mapp. 25, 155–164. Santi, A., Grodzinsky, Y., 2010. fMRI adaptation dissociates syntactic complexity di-
Lambon Ralph, M.A., Sage, K., Jones, R.W., Mayberry, E.J., 2010. Coherent concepts are mensions. Neuroimage 51, 1285–1293.
computed in the anterior temporal lobes. Proc. Natl. Acad. Sci. U. S. A. 107, Saur, D., Kreher, B.W., Schnell, S., Kummerer, D., Kellmeyer, P., Vry, M.S., Umarova, R.,
2717–2722. Musso, M., Glauche, V., Abel, S., Huber, W., Rijntjes, M., Hennig, J., Weiller, C.,
Lancaster, J.L., Rainey, L.H., Summerlin, J.L., Freitas, C.S., Fox, P.T., Evans, A.C., Toga, 2008. Ventral and dorsal pathways for language. Proc. Natl. Acad. Sci. U. S. A. 105,
A.W., Mazziotta, J.C., 1997. Automated labeling of the human brain: a preliminary 18035–18040.
report on the development and evaluation of a forward-transform method. Hum. Shapiro, L.P., Jensen, L.R., 1986. Processing open and closed class-headed nonwords: left
Brain Mapp. 5, 238–242. hemisphere support for separate vocabularies. Brain Lang. 28, 318–327.
Lancaster, J.L., Woldorff, M.G., Parsons, L.M., Liotti, M., Freitas, C.S., Rainey, L., Small, S.L., Flores, D.K., Noll, D.C., 1998. Different neural circuits subserve reading before
Kochunov, P.V., Nickerson, D., Mikiten, S.A., Fox, P.T., 2000. Automated Talairach and after therapy for acquired dyslexia. Brain Lang. 62, 298–308.
atlas labels for functional brain mapping. Hum. Brain Mapp. 10, 120–131. Snijders, T.M., Vosse, T., Kempen, G., Van Berkum, J.J., Petersson, K.M., Hagoort, P.,
Leyton, C.E., Villemagne, V.L., Savage, S., Pike, K.E., Ballard, K.J., Piguet, O., Burrell, 2009. Retrieval and unification of syntactic structure in sentence comprehension: an
J.R., Rowe, C.C., Hodges, J.R., 2011. Subtypes of progressive aphasia: application of FMRI study using word-category ambiguity. Cereb. Cortex 19, 1493–1503.
the International Consensus Criteria and validation using beta-amyloid imaging. Snowden, J.S., Goulding, P.J., Neary, D., 1989. Semantic dementia: a form of circum-
Brain 134, 3030–3043. scribed cerebral atrophy. Bahav. Neurol. 2, 167–182.
Luria, A.R., 1970. Traumatic Aphasia. Mouton, The Hague. Stowe, L.A., Broere, C.A., Paans, A.M., Wijers, A.A., Mulder, G., Vaalburg, W., Zwarts, F.,
Maguire, E.A., Frith, C.D., 2004. The brain network associated with acquiring semantic 1998. Localizing components of a complex task: sentence processing and working
knowledge. Neuroimage 22, 171–178. memory. Neuroreport 9, 2995–2999.
Makuuchi, M., Bahlmann, J., Anwander, A., Friederici, A.D., 2009. Segregating the core Stowe, L.A., Paans, A.M., Wijers, A.A., Zwarts, F., Mulder, G., Vaalburg, W., 1999.
computational faculty of human language from working memory. Proc. Natl. Acad. Sentence comprehension and word repetition: a positron emission tomography in-
Sci. U. S. A. 106, 8362–8367. vestigation. Psychophysiology 36, 786–801.
Matchin, W., Hickok, G., 2016. Syntactic perturbation’ during production activates the Swinney, D.A., Zurif, E.B., Cutler, A., 1980. Effects of sentential stress and word class
right IFG, but not Broca’s area or the ATL. Front. Psychol. 7, 241. upon comprehension in Broca’s aphasics. Brain Lang. 10, 132–144.
Matchin, W., Hammerly, C., Lau, E., 2017. The role of the IFG and pSTS in syntactic Talairach, J., Tournoux, P., 1988. Co-planar Stereotaxic Atlas of the Human Brain: 3-
prediction: evidence from a parametric study of hierarchical structure in fMRI. Cortex Dimensional Proportional System – An Approach to Cerebral Imaging. Thieme
88, 106–123. Medical Publisher, New York, NY.
Mazoyer, B., Tzourio, N., Frak, V., Syrota, A., Murayama, N., Levrier, O., Salamon, G., ter Keurs, M., Brown, C.M., Hagoort, P., Stegeman, D.F., 1999. Electrophysiological
Dehaene, S., Cohen, L., Mehler, J., 1993. The cortical representation of speech. J. manifestations of open- and closed-class words in patients with Broca’s aphasia with
Cogn. Neurosci. 5, 467–479. agrammatic comprehension. An event-related brain potential study. Brain 122 (Pt. 1),
Mazoyer, B., Zago, L., Jobard, G., Crivello, F., Joliot, M., Perchey, G., Mellet, E., Petit, L., 839–854.
Tzourio-Mazoyer, N., 2014. Gaussian mixture modeling of hemispheric lateralization Turkeltaub, P.E., Eden, G.F., Jones, K.M., Zeffiro, T.A., 2002. Meta-analysis of the func-
for language in a large sample of healthy individuals balanced for handedness. PLoS tional neuroanatomy of single-word reading: method and validation. Neuroimage 16,
One 9, e101165. 765–780.
Mellem, M.S., Jasmin, K.M., Peng, C., Martin, A., 2016. Sentence processing in anterior Turkeltaub, P.E., Eickhoff, S.B., Laird, A.R., Fox, M., Wiener, M., Fox, P., 2012.
superior temporal cortex shows a social-emotional bias. Neuropsychologia 89, Minimizing within-experiment and within-group effects in Activation Likelihood
217–224. Estimation meta-analyses. Hum. Brain Mapp. 33, 1–13.
Meyer, L., Obleser, J., Anwander, A., Friederici, A.D., 2012. Linking ordering in Broca's Tzourio-Mazoyer, N., Joliot, M., Marie, D., Mazoyer, B., 2016. Variation in homotopic
area to storage in left temporo-parietal regions: the case of sentence processing. areas’ activity and inter-hemispheric intrinsic connectivity with type of language
Neuroimage 62, 1987–1998. lateralization: an FMRI study of covert sentence generation in 297 healthy volun-
Mohr, B., Pulvermuller, F., Zaidel, E., 1994. Lexical decision after left, right and bilateral teers. Brain Struct. Funct. 221, 2735–2753.
presentation of function words, content words and non-words: evidence for inter- Uchiyama, Y., Toyoda, H., Honda, M., Yoshida, H., Kochiyama, T., Ebe, K., Sadato, N.,
hemispheric interaction. Neuropsychologia 32, 105–124. 2008. Functional segregation of the inferior frontal gyrus for syntactic processes: a
Mugler III, J.P., Brookeman, J.R., 1990. Three-dimensional magnetization-prepared rapid functional magnetic-resonance imaging study. Neurosci. Res. 61, 309–318.
gradient-echo imaging (3D MP RAGE). Magn. Reson. Med. 15, 152–157. Van Petten, C., Kutas, M., 1991. Influences of semantic and syntactic context on open- and
Murphy, E., 2015. Labels, cognomes, and cyclic computation: an ethological perspective. closed-class words. Mem. Cogn. 19, 95–112.
Front. Psychol. 6, 715. Vandenberghe, R., Nobre, A.C., Price, C.J., 2002. The response of left temporal cortex to
Neville, H.J., Mills, D.L., Lawson, D.S., 1992. Fractionating language: different neural sentences. J. Cogn. Neurosci. 14, 550–560.
subsystems with different sensitive periods. Cereb. Cortex 2, 244–258. Visser, M., Jefferies, E., Lambon Ralph, M.A., 2010. Semantic processing in the anterior
Newman, S.D., Ikuta, T., Burns Jr., T., 2010. The effect of semantic relatedness on temporal lobes: a meta-analysis of the functional neuroimaging literature. J. Cogn.

655
E. Zaccarella et al. Neuroscience and Biobehavioral Reviews 80 (2017) 646–656

Neurosci. 22, 1083–1094. Hear. Res. 132, 34–42.

Wang, Y., Xiang, J., Kotecha, R., Vannest, J., Liu, Y., Rose, D., Schapiro, M., Degrauw, T., Xu, J., Kemeny, S., Park, G., Frattali, C., Braun, A., 2005. Language in context: emergent
2008. Spatial and frequency differences of neuromagnetic activities between the features of word, sentence, and narrative comprehension. Neuroimage 25,
perception of open- and closed-class words. Brain Topogr. 21, 75–85. 1002–1015.
Warrington, E.K., 1975. The selective impairment of semantic memory. Q. J. Exp. Zaccarella, E., Friederici, A.D., 2015. Merge in the human brain: a sub-region based
Psychol. 27, 635–657. functional investigation in the left pars opercularis. Front. Psychol. 6, 524.
Weisenberg, T.M., McBride, K.E., 1935. Aphasia, A Clinical and Psychological Study. Zaccarella, E., Friederici, A.D., 2016. The neurobiological nature of syntactic hierarchies.
Commonwealth Fund, New York. Neurosci. Biobehav. Rev (advance access published Jul 29, 2016).
Wilson, S.M., Galantucci, S., Tartaglia, M.C., Rising, K., Patterson, D.K., Henry, M.L., Zaccarella, E., Meyer, L., Makuuchi, M., Friederici, A.D., 2017. Building by syntax: the
Ogar, J.M., DeLeon, J., Miller, B.L., Gorno-Tempini, M.L., 2011. Syntactic processing neural basis of minimal linguistic structures. Cereb. Cortex 27, 411–421.
depends on dorsal language tracts. Neuron 72, 397–403. Zhu, Z., Zhang, J.X., Wang, S., Xiao, Z., Huang, J., Chen, H.C., 2009. Involvement of left
Wilson, S.M., Galantucci, S., Tartaglia, M.C., Gorno-Tempini, M.L., 2012. The neural basis inferior frontal gyrus in sentence-level semantic integration. Neuroimage 47,
of syntactic deficits in primary progressive aphasia. Brain Lang. 122, 190–198. 756–763.
Wong, D., Miyamoto, R.T., Pisoni, D.B., Sehgal, M., Hutchins, G.D., 1999. PET imaging of Zhu, Z., Feng, G., Zhang, J.X., Li, G., Li, H., Wang, S., 2013. The role of the left prefrontal
cochlear-implant and normal-hearing subjects listening to speech and nonspeech. cortex in sentence-level semantic integration. Neuroimage 76, 325–331.

656