Cognition 203 (2020) 104348

Contents lists available at ScienceDirect

Cognition
journal homepage: www.elsevier.com/locate/cognit

Lack of selectivity for syntax relative to word meanings throughout the T

language network
⁎
Evelina Fedorenkoa,b, , Idan Asher Blanka,c, Matthew Siegelmana,d, Zachary Mineroffa,e
a
Department of Brain and Cognitive Sciences, MIT, Cambridge, MA 02139, USA
b
McGovern Institute for Brain Research, MIT, Cambridge, MA 02139, USA
c
Department of Psychology, UCLA, Los Angeles, CA 90095, USA
d
Department of Psychology, Columbia University, New York, NY 10027, USA
e
Eberly Center for Teaching Excellence & Educational Innovation, CMU, Pittsburgh, PA 15213, USA

A R T I C LE I N FO A B S T R A C T

Keywords: To understand what you are reading now, your mind retrieves the meanings of words and constructions from a
Lexical semantics linguistic knowledge store (lexico-semantic processing) and identifies the relationships among them to construct
Syntax a complex meaning (syntactic or combinatorial processing). Do these two sets of processes rely on distinct,
Composition specialized mechanisms or, rather, share a common pool of resources? Linguistic theorizing, empirical evidence
Language architecture
from language acquisition and processing, and computational modeling have jointly painted a picture whereby
Cognitive neuroscience
lexico-semantic and syntactic processing are deeply inter-connected and perhaps not separable. In contrast,
many current proposals of the neural architecture of language continue to endorse a view whereby certain brain
regions selectively support syntactic/combinatorial processing, although the locus of such “syntactic hub”, and
its nature, vary across proposals. Here, we searched for selectivity for syntactic over lexico-semantic processing
using a powerful individual-subjects fMRI approach across three sentence comprehension paradigms that have
been used in prior work to argue for such selectivity: responses to lexico-semantic vs. morpho-syntactic viola-
tions (Experiment 1); recovery from neural suppression across pairs of sentences differing in only lexical items
vs. only syntactic structure (Experiment 2); and same/different meaning judgments on such sentence pairs
(Experiment 3). Across experiments, both lexico-semantic and syntactic conditions elicited robust responses
throughout the left fronto-temporal language network. Critically, however, no regions were more strongly en-
gaged by syntactic than lexico-semantic processing, although some regions showed the opposite pattern. Thus,
contra many current proposals of the neural architecture of language, syntactic/combinatorial processing is not
separable from lexico-semantic processing at the level of brain regions—or even voxel subsets—within the
language network, in line with strong integration between these two processes that has been consistently ob-
served in behavioral and computational language research. The results further suggest that the language network
may be generally more strongly concerned with meaning than syntactic form, in line with the primary function
of language—to share meanings across minds.

1. Introduction between elements/substrings in the incoming linguistic signal and these

stored knowledge representations, and compose these retrieved re-
1.1. The functional architecture of language processing presentations, in an attempt to re-construct the intended meaning, and
during production (encoding of linguistic utterances), we search our
What is the functional architecture of human language? A core knowledge store for the right words/constructions and combine and
component is a set of knowledge representations, which include arrange them in a particular way to express a target idea.
knowledge of words/constructions/expressions and their meanings, How is this rich set of representations and computations structured?
and the probabilistic constraints on how these linguistic elements can Specifically, which aspects of language are functionally dissociable
combine to create compound words, phrases, and sentences. During from one another? Traditionally, two principal distinctions have been
comprehension (decoding of linguistic utterances), we look for matches drawn: one is between words (the lexicon) and rules (the grammar)

⁎
Corresponding author at: 43 Vassar Street, Room 46-3037G, Cambridge, MA 02139, USA.
E-mail address: evelina9@mit.edu (E. Fedorenko).

https://doi.org/10.1016/j.cognition.2020.104348
Received 23 November 2018; Received in revised form 14 May 2020; Accepted 31 May 2020
0010-0277/ © 2020 Elsevier B.V. All rights reserved.
E. Fedorenko, et al. Cognition 203 (2020) 104348

Fig. 1. A (non-exhaustive) set of theoretically possible architectures of language. Distinct boxes correspond to distinct brain regions (or sets of brain regions; e.g., in
1a-d, “combinatorial processing” may recruit a single region or multiple regions, but critically, this region or these regions do not support other aspects of language
processing, like understanding word meanings). The architectures differ in whether they draw a (region-level) distinction between the lexicon and grammar (a vs. b-
f), between storage and access of linguistic representations (1a-b vs. 1c-f), and critically, in whether combinatorial processing is a separable component (1a-d vs. 1e-
f).

(e.g., Chomsky, 1965, 1995; Fodor, 1983; Pinker, 1991, 1999; Pinker & exquisitely sensitive to contingencies between particular words and the
Prince, 1988); and another is between linguistic representations constructions they occur in (e.g., Clifton, Frazier, & Connine, 1984;
themselves (i.e., our knowledge of the language) and their online pro- Garnsey, Pearlmutter, Myers, & Lotocky, 1997; Jaeger, 2010;
cessing (i.e., accessing them from memory and combining them to MacDonald, Pearlmutter, & Seidenberg, 1994; Reali & Christiansen,
create new complex meanings and structures) (e.g., Chomsky, 1965; 2007; Roland, Dick, & Elman, 2007; Traxler, Morris, & Seely, 2002;
Fodor, Bever, & Garrett, 1974; Newmeyer, 2003). Because these di- Trueswell, Tanenhaus, & Garnsey, 1994), and ii) store not just atomic
mensions are, in principle, orthogonal, we could have distinct mental elements (like morphemes and non-compositional lexical items), but
capacities associated with i) knowledge of word (lexical) meanings, ii) also compositional phrases (e.g., “I don't know” or “give me a break”;
knowledge of grammar (syntactic/combinatorial rules), iii) access (or e.g., Wray, 2005; Evert, 2008; Arnon & Snider, 2010; Morgan & Levy,
“retrieval”) of lexical representations, iv) access of syntactic/combina- 2016; Christiansen & Arnon, 2017) and constructions (e.g., “the X-er
torial rules, and v) combining retrieved representations into new the Y-er”; Goldberg, 1995; Culicover & Jackendoff, 1999). The latter
complex representations (Fig. 1a). suggested that the linguistic units people store are determined not by
However, both of these distinctions have been long debated. For their nature (i.e., atomic vs. not) but instead, by their patterns of usage
example, as linguistic theorizing evolved and experimental evidence (e.g., Barlow & Kemmer, 2000; Bybee, 1998, 2006; Goldberg, 2006;
accumulated through the 1970s–90s, the distinction between the lex- Langacker, 1986, 1987; Tomasello, 2003). Further, people's lexical
icon and grammar began to blur, both for the storage of linguistic abilities have been shown to strongly correlate with their grammatical
knowledge representations and for online processing (e.g., Fig. 1b; see abilities—above and beyond shared variance due to general fluid in-
Snider & Arnon, 2012, for a summary and discussion). Many have ob- telligence—both developmentally (e.g., Bates, Bretherton, & Snyder,
served that much of our grammatical knowledge does not operate over 1988; Bates, Dale, & Thal, 1995; Bates & Goodman, 1997; Dale, Dionne,
general categories like nouns and verbs, but instead requires reference Eley, & Plomin, 2000; Dale, Harlaar, Haworth, & Plomin, 2010; Dixon &
to particular words or word classes (e.g., verbs that can occur in a Marchman, 2007; Hoff, Quinn, & Giguere, 2018; Marchman & Bates,
particular construction, like the ditransitive) (e.g., Lakoff, 1970; Bybee, 1994; Mcgregor, Sheng, & Smith, 2005; Moyle, Weismer, Evans, &
1985, 1998, 2010; Levin, 1993; Goldberg, 1995, 2002; Jackendoff, Lindstrom, 2007; Snedeker, Geren, & Shafto, 2007, 2012) and in
2002a, 2002b, 2007; Sag et al., 2003; Culicover & Jackendoff, 1999; adulthood (e.g., Dąbrowska, 2018). Thus, linguistic mechanisms that
Levin & Rappaport-Hovav, 2005; Jackendoff and Audring, 2020). As a have been previously proposed to be distinct are instead tightly in-
result, current linguistic frameworks incorporate knowledge of “rules” tegrated or, perhaps, are so cognitively inseparable as to be considered
(i.e., syntactic structures) into the mental lexicon, although they differ a single apparatus.
as to the degree of abstraction that exists above and beyond knowledge The distinction between stored knowledge representations and on-
of how particular words combine with other words (e.g., Ambridge, line computations has also been questioned (see Hasson, Chen, &
2018; see Hudson, 2007, for discussion), and in whether abstract syn- Honey, 2015, for a discussion of this issue in language and other do-
tactic representations (like the double object, passive, or question mains). For example, by using the same artificial network to represent
constructions) are always associated with meanings or functions (e.g., all linguistic experience, connectionist models dispense not only with
Pinker, 1989; Goldberg, 1995; cf. Chomsky, 1987; Branigan & the lexicon-grammar distinction but also the storage-computation one,
Pickering, 2017; see Jackendoff, 2002b, for discussion). and assume that the very same units that represent our linguistic
In line with these changes in linguistic theorizing, experimental and knowledge support its online access and processing (e.g., Rumelhart
corpus work in psycholinguistics have established that humans i) are and McClelland, 1986; Seidenberg, 1994; Devlin, Chang, Lee, &

2
E. Fedorenko, et al. Cognition 203 (2020) 104348

Toutanova, 2019; see also Goldinger, 1996; Bod, 1998, 2006, for ex- Vandenberghe, Nobre, & Price, 2002), with some positing multiple hubs
emplar models, which also abandon the storage-computation divide; cf. (e.g., Pallier, Devauchelle, & Dehaene, 2011; Tyler et al., 2011; Wilson,
Chang, Dell, & Bock, 2006 for a connectionist model of sentence pro- Galantucci, Tartaglia, & Gorno-Tempini, 2012).
duction that includes a separate structural—sequencing—component). Second, at least some syntactic/combinatorial manipulations ap-
pear to engage the entire fronto-temporal language network (e.g., Blank,
1.2. Syntax selectivity in prior cognitive neuroscience investigations? Balewski, Mahowald, & Fedorenko, 2016; Fedorenko, Hsieh, Nieto-
Castanon, Whitfield-Gabrieli, & Kanwisher, 2010; Pallier et al., 2011),
Alongside psycholinguistic studies, which inform debates about putting into question the idea of syntactic processing being focally
linguistic architecture by examining the behaviors generated by lan- carried out. Relatedly, studies of patients with brain damage have failed
guage mechanisms, and computational work, which aims to approx- to consistently link syntactic deficits with a particular region within the
imate human linguistic behavior using formal models, a different, language network. Instead, damage to any component of the network
complementary approach is offered by cognitive neuroscience studies. appears to lead to similar syntactic difficulties—which also mirror
These studies aim to constrain the cognitive architecture by examining patterns observed in neurotypical individuals under cognitive load
how cognitive processes are neurally implemented (e.g., Kanwisher, (Miyake, Carpenter, & Just, 1994; Blackwell & Bates, 1995)—leading
2010; Mather, Cacioppo, & Kanwisher, 2013). The assumption that some to argue that syntactic processing is supported by the language
links neuroimaging data (and neuropsychological patient data) to network as a whole (e.g., Caplan, Hildebrandt, & Makris, 1996; Dick
cognitive hypotheses is as follows: if distinct brain regions or sets of et al., 2001; Mesulam et al., 2014; Mesulam, Thompson, Weintraub, &
regions support the processing of manipulations targeting cognitive Rogalski, 2015; Wilson & Saygin, 2004). Given that the language net-
processes X and Y, we can infer that X and Y are dissociable. Such brain work has to additionally support lexico-semantic processing, these
regions would be expected to show distinct patterns of response in brain findings necessarily imply that at least some, and possibly all, of these
imaging studies, and their damage should lead to distinct patterns of syntax-responsive regions overlap with regions that support lexico-se-
cognitive deficits. mantic processing.
A large number of brain imaging investigations have observed dis- Third, a number of studies have actually failed to observe syntax
tinct loci of activation for manipulations that target (lexico-)semantic selectivity, showing that brain regions that respond to syntactic manip-
vs. syntactic processing, and—following the reasoning above—have ulations also show reliable, and sometimes stronger, responses to
argued for a dissociation between the two (e.g., Dapretto & lexico-semantic manipulations (e.g., Chee et al., 1999; Keller,
Bookheimer, 1999; Embick, Marantz, Miyashita, O'Neil, & Sakai, 2000; Carpenter, & Just, 2001; Roder, Stock, Neville, Bien, & Rosler, 2002;
Kuperberg et al., 2000; Ni et al., 2000; Newman, Pancheva, Ozawa, Luke et al., 2002; Heim, Eickhoff, & Amunts, 2008; Rogalsky & Hickok,
Neville, & Ullman, 2001; Kuperberg et al., 2003; Noppeney & Price, 2009; Fedorenko et al., 2010, Fedorenko, Nieto-Castanon, & Kanwisher,
2004; Cooke et al., 2006; Friederici, Kotz, Scott, & Obleser, 2010; 2012, Fedorenko & Varley, 2016 (PNAS); Bautista & Wilson, 2016;
Glaser, Martin, Van Dyke, Hamilton, & Tan, 2013; Schell, Zaccarella, & Blank et al., 2016; Shain, Blank, van Schijndel, Schuler, & Fedorenko,
Friederici, 2017, inter alia; see Hagoort & Indefrey, 2014, for a meta- 2020; Wang, Rice, & Booth, 2020; see Rodd, Vitello, Woollams, &
analysis). Consequently, many proposals of the neural architecture of Adank, 2015, for a meta-analysis). Relatedly, some studies have re-
language postulate a component that selectively supports syntactic, or ported activations for lexico-semantic manipulations, such as lexical
more general combinatorial, processing relative to the storage/proces- ambiguity manipulations, in what appear to be the same regions as the
sing of individual word meanings (e.g., Grodzinsky & Santi, 2008; ones implicated in other studies in syntactic/combinatorial processing:
Baggio & Hagoort, 2011; Friederici, 2011, 2012; Tyler et al., 2011; i.e., regions in the left inferior fontal and posterior temporal cortex
Duffau, Moritz-Gasser, & Mandonnet, 2014; Ullman, 2016; Matchin & (e.g., Bekinschtein, Davis, Rodd, & Owen, 2011; Bilenko, Grindrod,
Hickok, 2019; Pylkkänen, 2019; cf. Bornkessel-Schlesewsky & Myers, & Blumstein, 2008; Davis et al., 2007; Mason & Just, 2007; Rodd
Schlesewsky, 2009; Bornkessel-Schlesewsky, Schlesewsky, Small, & et al., 2012; Rodd, Davis, & Johnsrude, 2005; Rodd, Longe, Randall, &
Rauschecker, 2015 – we come back to these proposals in the Discus- Tyler, 2010; Zempleni, Renken, Hoeks, Hoogduin, & Stowe, 2007).
sion). Although proposals vary in which component(s) of syntactic And fourth, many prior studies that have reported syntax selectivity
processing are emphasized—from morpho-syntactic agreement, to de- suffer from methodological and statistical limitations. For example, al-
pendency structure building and composition, to word-order-related though diverse paradigms have been used across studies to probe
processes—the general idea of syntax selectivity remains prominent in lexico-semantic vs. syntactic/combinatorial processing, any given study
cognitive neuroscience of language. How can we fit such selectivity has typically used a single paradigm, raising the possibility that the
with current linguistic theorizing, psycholinguistic evidence, and results reflect paradigm-specific differences between conditions rather
computational modeling work, which suggest strong integration be- than a general difference between lexico-semantic and syntactic/com-
tween lexico-semantic and syntactic representations and processing, as binatorial representations or computations. Furthermore, many studies
discussed above? Here, we revisit past evidence and report a new study that claimed to have observed a dissociation have not reported the
to argue against the existence of brain regions that are selective for required region-by-condition interactions, as needed to argue for a
syntactic/combinatorial processing over the processing of word functional dissociation between brain regions (Nieuwenhuis,
meanings. Forstmann, & Wagenmakers, 2011). And some studies have argued for
There already exist reasons to doubt the existence of syntax-selec- syntax selectivity based solely on sensitivity to syntactic complexity
tive brain regions if we take a closer look at the cognitive neuroscience manipulations, without even examining responses to lexico-semantic
of language literature. First, the specific brain regions that have been processing (e.g., Stromswold, Caplan, Alpert, & Rauch, 1996; Ben-
argued to support syntactic/combinatorial processing, and the con- Shachar, Hendler, Kahn, Ben-Bashat, & Grodzinsky, 2003; Fiebach,
strual of these regions' contributions, differ across studies and theoretical Schlesewsky, Lohmann, Von Cramon, & Friederici, 2005; Santi &
proposals (e.g., Baggio & Hagoort, 2011; Bemis & Pylkkanen, 2011; Grodzinsky, 2010; see Friederici, 2011, for a meta-analysis). Although
Duffau et al., 2014; Friederici, 2011, 2012; Matchin & Hickok, 2019; such studies (may) establish that a brain region is engaged in syntactic
Pylkkänen, 2019; Tyler et al., 2011; Ullman, 2004, 2016). For example, processing, they say little about its selectivity for syntactic over lexico-
the proposed location of the “core” syntactic/combinatorial hub varies semantic processing. Finally, some studies have reported sensitivity to
between the inferior frontal cortex (e.g., Friederici, Bahlmann, Heim, syntactic manipulations in regions that fall outside the boundaries of
Schubotz, & Anwander, 2006; Hagoort, 2005, 2013), the posterior the fronto-temporal language network. For example, some studies of
temporal cortex (e.g., Matchin & Hickok, 2019), and the anterior morpho-syntactic violations (e.g., Kuperberg et al., 2003; Nieuwland
temporal cortex (e.g., Bemis & Pylkkanen, 2011; Pylkkänen, 2019; et al., 2012) have reported effects in regions that resemble the domain-

3
E. Fedorenko, et al. Cognition 203 (2020) 104348

general bilateral fronto-parietal network implicated in executive con- in fMRI research). To the best of our knowledge, nobody has demon-
trol (e.g., Duncan, 2010, 2013). This network is sensitive to unexpected strated syntactic selectivity using this kind of a rigorous approach, or
events across domains (e.g., Corbetta & Shulman, 2002), and some of its even attempted to do so. At least two factors have likely contributed to
regions lie in close proximity to the language regions (e.g., Fedorenko & the lack of such attempts, and to the resulting lack of clarity in the field.
Blank, 2020; Fedorenko, Nieto-Castañon, & Kanwisher, 2012). Al- First, until a decade ago, the issue of replicability has not been much
though the precise nature of this network's contribution to language discussed in the fields of psychology/cognitive science (e.g., Ioannidis,
processing remains debated (e.g., Diachek, Siegelman, Blank, Affourtit, Munafo, Fusar-Poli, Nosek, & David, 2014) or cognitive neuroscience
& Fedorenko, 2020; Fedorenko, 2014; Ryskin, Levy, & Fedorenko, (e.g., Poldrack et al., 2017). And at least some of the findings in cog-
2020; Shain et al., 2020; Wehbe et al., 2020), sensitivity of this network nitive neuroscience of language that have been taken at face value
to a linguistic manipulation likely indexes a domain-general process, based on a single report by a single group may not be robust and re-
plausibly related to task demands (Diachek et al., 2020), and does not plicable (e.g., see Siegelman, Blank, Mineroff, & Fedorenko, 2019, for a
inform the question of whether different components of the language recent attempt, and failure, to replicate a much cited report by Dapretto
network support syntactic vs. lexico-semantic processing. & Bookheimer, 1999). This issue is further compounded by the many
hidden degrees of freedom (e.g., Simmons, Nelson, & Simonsohn, 2011)
1.3. Motivation for the current study that characterize the choices during the preprocessing and analysis of
brain imaging data (Botvinik-Nezer et al., 2019) and the common use of
The current study aims to resolve the conflict between i) converging “double dipping” (e.g., Kriegeskorte, Simmons, Bellgowan, & Baker,
evidence from linguistic theorizing, behavioral psycholinguistic work, 2009) in many early studies.
and computational modeling, which have jointly painted a clear picture And second, as we have previously argued (e.g., Fedorenko &
of strong integration between lexico-semantic and syntactic re- Kanwisher, 2009; Fedorenko et al., 2010, Fedorenko, Nieto-Castañon, &
presentations and processing, and ii) cognitive neuroscience studies and Kanwisher, 2012; Blank, Kiran, & Fedorenko, 2017; Fedorenko & Blank,
proposals, many of which continue to suggest the existence of syntax- or 2020; see also Brett, Johnsrude, & Owen, 2002 and Saxe, Brett, &
combinatorics-selective brain regions. To do so, we use fMRI to search Kanwisher, 2006), establishing a cumulative research enterprise in
for syntax selectivity using a robust individual-subjects analytic approach, cognitive neuroscience of language has been challenging due to the
including a well-validated task for identifying the language network difficulty of comparing findings across studies that rely on the tradi-
(i.e., a “functional localizer”; Fedorenko et al., 2010) across three classic tional group-averaging approach (e.g., Holmes & Friston, 1998). In this
paradigms from the literature that contrast lexico-semantic and syn- analytic approach—which has dominated the brain-imaging language
tactic processing. We will now discuss and motivate both of these fea- research in the 1990s and 2000s and is still in common use despite
tures of our study in greater detail, in the historical context of the field. being strongly disfavored in neuroimaging studies in other do-
mains—individual activation maps are aligned in a common space, and
1.3.1. The need for robust, replicable, and cumulative science the output is a set of coordinates in that space for voxels where sig-
Over the years, a large number of paradigms have been used across nificant effects obtain (typically the most reliable peak(s) in each ac-
brain imaging studies to probe lexico-semantic and syntactic processes tivation cluster are reported). The main way to compare results across
and their relationship (see references above and in Materials and such studies is to compare the anatomical locations of these activation
methods section). Some paradigms, discussed below (Section 1.3.2), peaks. However, group-level activation peaks are noisy (e.g.,
have varied the presence or absence of lexico-semantic vs. syntactic in- Fedorenko, Nieto-Castañon, & Kanwisher, 2012), due to the combined
formation in the linguistic signal; others have more strongly taxed the effect of two factors, both especially pronounced in the association
processing of word meanings vs. syntactic structure; still others have cortex, which houses the language system: (1) high inter-individual
made the meaning of a particular word vs. the structure of the sentence variability in the locations of functional areas (e.g., Braga, DiNicola, &
more salient / task-relevant. Any of these kinds of manipulations could be Buckner, 2019; Fedorenko et al., 2010; Fedorenko & Blank, 2020;
informative, but no single manipulation—and certainly not from a Mahowald & Fedorenko, 2016); and (2) lack of correspondence be-
single experiment—suffices to argue for syntax selectivity. To compel- tween functional areas and macroanatomic landmarks, like gyri and
lingly argue that a brain region selectively supports (some aspect of) sulci (e.g., Frost & Goebel, 2012; Tahmasebi et al., 2012; Vázquez-
syntactic processing, one would need to demonstrate both the robustness Rodríguez et al., 2019). And even the most systematic comparisons in
of the syntactic > lexico-semantic effect (e.g., replication in a new the form of meta-analyses of activation peaks from large numbers of
sample of participants, on a new set of experimental materials, and/or studies (e.g., Binder, Desai, Graves, & Conant, 2009; Bookheimer, 2002;
in another imaging modality) and its generalizability to other contrasts Costafreda et al., 2006; Hagoort & Indefrey, 2014; Indefrey & Levelt,
between conditions that engage the hypothesized computation and 2004; Kaan & Swaab, 2002; Lindenberg, Fangerau, & Seitz, 2007;
ones that do not. In particular, the selectivity of a brain region for the Poldrack et al., 1999; Vigneau et al., 2006) are not very informative
critical syntactic condition(s) would need to be established relative to a (Kvarven, Strømland, & Johannesson, 2019), and have been shown to
broad range of control conditions, given that any given syntactic vs. lead to fundamentally wrong conclusions about the functional archi-
lexico-semantic pair of conditions will likely differ in ways beyond the tecture of the human brain in some cases (e.g., Aguirre & Farah, 1998).
target distinction between syntax and semantics. For example, showing So what is a solution?
that morpho-syntactic violations elicit a stronger response than se- A decade ago, we developed an alternative approach to the study of
mantic violations is not sufficient to argue for syntactic selectivity be- language in the brain (Fedorenko et al., 2010)—one that had been
cause these two conditions also differ in whether the error can be ex- successful in other domains, including high-level vision (e.g.,
plained by a plausible noise process within a noisy-channel framework Kanwisher, McDermott, & Chun, 1997) and social cognition (e.g., Saxe
of sentence comprehension (e.g., Ferreira, Bailey, & Ferraro, 2002; Levy & Kanwisher, 2003), and has now become widespread in the study of
et al., 2009; Gibson, Bergen, & Piantadosi, 2013); thus, a stronger re- language (e.g., Axelrod, Bar, Rees, & Yovel, 2015; Braga et al., 2019;
sponse to morpho-syntactic violations could reflect the relevant cor- Lane, Kanjlia, Omaki, & Bedny, 2015; Loiotile, Cusack, & Bedny, 2019;
rection process, which does not get engaged for typical semantic vio- Matchin, Brodbeck, Hammerly, & Lau, 2019; Pant, Kanjlia, & Bedny,
lations (Ryskin et al., 2020). Furthermore, it would be critical to ensure 2020; Poldrack et al., 2015; Wang et al., 2020; Wang, Uhrig, Jarraya, &
that—across studies—the key effect arises in the same brain region, not Dehaene, 2015). In this approach, language-responsive areas are de-
just within the same broad macroanatomic area, like the left inferior fined functionally in individual brains without being constrained to fall
frontal gyrus (LIFG) (see Hong, Yoo, Wager, & Woo, 2019, for a general precisely in the same anatomical locations across participants; and the
discussion of challenges to determining what counts as a “replication” localized regions are then probed for their responses to critical

4
E. Fedorenko, et al. Cognition 203 (2020) 104348

experimental manipulations. The use of the same “localizer” paradigm processing—cover both morpho-syntactic agreement (Experiment 1),
across individuals, studies, and research groups (and in domains, like and dependency structure building/word-order-related processes (Ex-
vision, this is done across species, too; e.g., Tsao, Moeller, & Freiwald, periments 2 and 3).
2008) provides a straightforward way to directly relate findings to one If any brain region within the language network selectively supports
another. syntactic processing, we would expect stronger responses to the syn-
tactic than the lexico-semantic condition in that region in at least one
paradigm. If this pattern holds—for the same brain region(s)—across
1.3.2. Choice of paradigms two or all three paradigms, that would further help rule out paradigm-
Using the individual-subjects functional localization approach, we specific between-condition differences/confounds and strengthen the
have previously argued for the lack of syntax selectivity based on a conclusion. Note that unlike the paradigms that vary the presence/ab-
paradigm that varies the presence of lexico-semantic vs. syntactic in- sence of syntactic and lexical information in the linguistic signal—-
formation in the linguistic signal (Fedorenko et al., 2010, Fedorenko, where one could, in principle, observe a pattern where a brain region is
Nieto-Castanon, & Kanwisher, 2012, Fedorenko & Varley, 2016 (PNAS); not at all engaged unless structure or meaning is present (although in
for earlier uses of this paradigm, see e.g., Mazoyer et al., 1993; practice, even lists of pseudowords, which lack both structure and
Friederici, Meyer, & von Cramon, 2000; Humphries, Willard, meaning, elicit an above-baseline response across much of the language
Buchsbaum, & Hickok, 2001; Vandenberghe et al., 2002; for another network; Fedorenko et al., 2010)—the current paradigms all use sen-
variant, see Bautista & Wilson, 2016). In particular, we examined the tence materials across conditions, so all conditions are expected to elicit
processing of i) sentences, which have a syntactic structure and consist above-baseline responses throughout the language network. The critical
of real, interpretable words, ii) lists of unconnected words, which lack question is whether any brain region(s) would exhibit stronger re-
structure but are individually interpretable, iii) “Jabberwocky” sen- sponses for the syntactic compared to lexico-semantic condition in one
tences, which preserve a syntactic frame (word order and morpho- or more paradigms. If no brain region within the language network
syntactic endings), but have the words replaced by nonwords, so the shows this pattern, this would strongly reinforce the conclusions drawn
meanings of those strings cannot be interpreted with respect to our from the paradigms that have varied the presence/absence of lexico-
world knowledge, aside from very coarse-level semantics, and finally, semantic and syntactic information in the signal (e.g., Bautista &
iv) lists of unconnected nonwords, which lack both structure and in- Wilson, 2016; Fedorenko et al., 2010).
terpretability. Across three replications with fMRI (Fedorenko et al., To foreshadow the results: using the most sensitive analytic methods
2010; see Mollica et al., in prep for another replication) and, in addi- available in fMRI (e.g., Nieto-Castañón & Fedorenko, 2012), we find
tion, in a more spatially and temporally sensitive method (electro- robust responses to both lexico-semantic and syntactic processing
corticography, ECoG) (Fedorenko et al., 2016 (PNAS)), we found that throughout the language network in each of the three experiments.
any language-responsive brain region or electrode that shows sensi- Critically, every brain region in the language network that responds to
tivity to syntactic structure (i.e., stronger responses to sentences than syntactic manipulations responds at least as strongly to lexico-semantic
word lists, and to Jabberwocky sentences than nonword lists) is at least manipulations. No region—or even set of non-contiguous voxels within
as sensitive, and often more sensitive, to meanings of individual words these regions—shows a consistent preference, in the form of a stronger
(showing stronger responses to sentences than Jabberwocky sentences, response, for syntactic processing (ruling out architectures in
and to word lists than nonword lists). Fig. 1a–d). However, in line with our prior work (e.g., Fedorenko,
However, one could question the findings from this paradigm be- Nieto-Castanon, & Kanwisher, 2012; Fedorenko et al., 2016 (PNAS)),
cause the contrasts are rather crude and the materials are artificial/ some regions show the opposite preference—a stronger response to
unnatural. If the overlap in the brain mechanisms that process in- lexico-semantic processing. We therefore hope that this study brings
dividual word meanings and syntactic structure is a real and robust clarity to the field and helps build stronger bridges, grounded in robust
finding, the results should generalize to other, finer-grained compar- empirical work, with behavioral and computational investigations of
isons between lexico-semantic and syntactic processing. As a result, we language processing.
selected three paradigms from studies that have argued for syntax se-
lectivity or for dissociations between lexico-semantic and syntactic
processing, and that continue to be cited as evidence of such, and at- 2. Materials and methods
tempted to conceptually replicate (Schmidt, 2009) them.
Experiments 1 and 3 are designed to differentially tax lexico-se- 2.1. General description of the paradigms and their use in prior studies
mantic vs. syntactic processing by having a critical word in a sentence
be incompatible with the context in terms of either its meaning or The first paradigm is commonly used in ERP investigations of lan-
morpho-syntactic properties (Experiment 1), or by forcing participants guage processing and relies on violations of expectations about an in-
to focus on the meanings of the critical words or the structure of sen- coming word that are set up by the preceding context. In particular, the
tences (Experiment 3). Experiment 2 relies on the well-established critical word does not conform to either the lexico-semantic or the
neural adaptation to the repetition of a stimulus and recovery from such syntactic expectations (e.g., Hagoort, Brown, & Groothusen, 1993;
adaptation when some relevant feature of the stimulus changes: here, a Kutas & Hillyard, 1980; Osterhout & Holcomb, 1992). This paradigm
change in the individual words (but not the sentence structure) vs. the has been used in a number of prior fMRI studies (e.g., Cooke et al.,
sentence structure (but not the words). Along with the manipulations 2006; Embick et al., 2000; Friederici et al., 2010; Herrmann, Obleser,
varying the presence/absence of lexico-semantic and syntactic in- Kalberlah, Haynes, & Friederici, 2012; Kuperberg et al., 2003; Newman
formation in the linguistic signal discussed above, these manipulations et al., 2001). The second paradigm relies on neural adaptation, wherein
span the space of available manipulations targeting lexico-semantic and repeated exposure to a stimulus leads to a reduction in response, and a
syntactic processing quite comprehensively,1 and—for syntactic change in some feature(s) of the stimulus leads to a recovery of re-
sponse (see e.g., Krekelberg, Boynton, & van Wezel, 2006, for a general
1
One type of manipulation missing here is one that relies on ambiguity. Both
lexical (e.g., Bekinschtein et al., 2011; Davis et al., 2007; Mason & Just, 2007; (footnote continued)
Rodd et al., 2005, 2010, 2012; Zempleni et al., 2007) and structural (e.g., temporal cortex. Because, to the best of our knowledge, no arguments for
Mason, Just, Keller, & Carpenter, 2003) ambiguity have been investigated in syntactic selectivity have been made based on stronger responses to syntactic
prior fMRI studies—although the former has received more attention—and both than lexical ambiguity in some part(s) of the language network, we did not
have been shown to elicit responses in the inferior frontal and posterior include ambiguity manipulations here.

5
E. Fedorenko, et al. Cognition 203 (2020) 104348

Table 1 2.3. Participants

Timing parameters for the different versions of the language localizer task.
Version Forty-nine individuals (age 19–32, 27 females) participated for
payment (Experiment 1: n = 22; Experiment 2: n = 14; and
A B C Experiment 3: n = 15; 2 participants overlapped between Experiments
1 and 3). Forty-seven were right-handed, as determined by the
Number of participants 30 14 5
Task: Button press or memory? BP M M Edinburgh handedness inventory (Oldfield, 1971), or self-report; the
Words/nonwords per trial 12 12 12 two left-handed individuals showed typical left-lateralized language
Trial duration (ms) 6000 6000 6000 activations in the language localizer task (see Willems, der Haegen,
Fixation 100 – – Fisher, & Francks, 2014, for arguments to include left-handers in cog-
Presentation of each word/nonword 450 350 350
nitive neuroscience research). All participants were native speakers of
Fixation – 300 300
Button icon/memory probe 400 1000 1000 English from Cambridge, MA and the surrounding community. One
Fixation 100 500 500 additional participant was scanned (for Experiment 2) but excluded
Trials per block 3 3 3 from the analyses due to excessive sleepiness and poor behavioral
Block duration (s) 18 18 18
performance. All participants gave informed consent in accordance
Blocks per condition (per run) 8 8 6
Conditions Sentences Sentences Sentences
with the requirements of MIT's Committee on the Use of Humans as
Nonwords Nonwords Nonwords Experimental Subjects (COUHES).
Word-listsa
Fixation block duration (s) 14 18 18
Number of fixation blocks 5 5 4 2.4. Design, stimuli, and procedure
Total run time (s) 358 378 396
Number of runs 2 2 2–3 Each participant completed a language localizer task (Fedorenko
a et al., 2010) and a critical task (35 participants performed the localizer
Used for an unrelated experiment.
task in the same session as the critical task, the remaining 14 performed
the localizer in an earlier session; see Mahowald & Fedorenko, 2016 and
overview of the approach). This paradigm has also been used in prior
Braga et al., 2019, for evidence that localizer activations are stable
fMRI studies that examined adaptation, or recovery from adaptation, to
across scanning sessions). Most participants completed one or two ad-
the lexico-semantic vs. syntactic features of linguistic stimuli (e.g.,
ditional tasks for unrelated studies. The entire scanning session lasted
Devauchelle, Oppenheim, Rizzi, Dehaene, & Pallier, 2009; Menenti,
approximately 2 h.
Petersson, & Hagoort, 2012; Noppeney & Price, 2004; Santi &
Grodzinsky, 2010; Segaert, Menenti, Weber, Petersson, & Hagoort,
2012). Finally, the third paradigm was introduced in a classic study by 2.4.1. Language localizer task
Dapretto and Bookheimer (1999): pairs of sentences vary in a single The task used to localize the language network is described in detail
word vs. in word order/syntactic structure. Either of these manipula- in Fedorenko et al. (2010). Briefly, we used a reading task that con-
tions can result in the same meaning being expressed across sentences trasted sentences and lists of unconnected, pronounceable nonwords in
(if a word is replaced by a synonym, or if a syntactic alternation, like a standard blocked design with a counterbalanced order across runs (for
active➔passive, is used) or in a different meaning (if a word is replaced timing parameters, see Table 1). This contrast targets higher-level as-
by a non-synonym, or if the thematic roles are reversed). Participants pects of language including, critically, both lexico-semantic and syn-
make same/different meaning judgments on the resulting sentence tactic/combinatorial processing, to the exclusion of perceptual (speech
pairs. or reading-related) and articulatory processes (see e.g., Fedorenko &
Thompson-Schill, 2014, for discussion). As discussed above, brain re-
2.2. General analytic approaches gions thus localized indeed show sensitivity to lexico-semantic proces-
sing (as evidenced by stronger responses to sentences than Jabber-
In an effort to maximize sensitivity and functional resolution (e.g., wocky sentences, and to word lists than to nonword lists) and to
Nieto-Castañón & Fedorenko, 2012), we adopt an approach where all syntactic processing (as evidenced by stronger responses to sentences
the key contrasts are examined within individual participants. We than word lists, and to Jabberwocky sentences than nonword lists) (e.g.,
perform two kinds of analyses. In one set of analyses, we identify lan- Mollica et al., in prep; Fedorenko et al., 2010; Fedorenko, Nieto-
guage-responsive areas in each individual participant with an in- Castanon, & Kanwisher, 2012; Fedorenko et al., 2016 (PNAS)). Stimuli
dependent language “localizer” task based on a contrast between the were presented one word/nonword at a time. For 19 participants, each
reading of sentences vs. sequences of nonwords (Fedorenko et al., trial ended with a memory probe and they had to indicate, via a button
2010), and compare the responses of these areas to the lexico-semantic press, whether or not that probe had appeared in the preceding se-
vs. syntactic conditions in each critical paradigm. As discussed in the quence of words/nonwords. The remaining 30 participants read the
Introduction, the use of the same language localizer task across ex- materials passively and performed a simple button-press task at the end
periments allows for a straightforward comparison of their results, of each trial (included in order to help participants remain alert). Im-
obviating the need to rely on coarse anatomy and reverse-inference portantly, this localizer has been shown to generalize across different
reasoning (Poldrack, 2006, 2011) for interpreting activations in func- versions: the sentences > nonwords contrast, and similar contrasts be-
tional terms. tween language and a degraded control condition, robustly activates
To further ensure that we are not missing syntax selectivity by the fronto-temporal language network regardless of the task, materials,
averaging across (relatively) large sets of language-responsive voxels and modality of presentation (Fedorenko, 2014; Fedorenko et al., 2010;
(see Friston, Rotshtein, Geng, Sterzer, & Henson, 2006, for discussion of Scott, Gallee, & Fedorenko, 2016). Furthermore, the same network
this potential issue), we supplement these analyses with analyses where robustly emerges from naturalistic-cognition paradigms (e.g., resting
we only use data from the critical paradigms. In particular, we use some state, listening to stories, watching movies) using the functional cor-
of the data from a given critical task to search for the most syntactic- relation approach (e.g., Blank, Kanwisher, & Fedorenko, 2014; Braga
selective (or lexico-semantic-selective, examined for completeness) et al., 2019; Branco, Seixas, & Castro, 2020; Tie et al., 2014), suggesting
voxels (e.g., in Experiment 1, voxels that respond more strongly to that this network constitutes a natural kind in the brain, and our lo-
morpho-syntactic than semantic violations), and then test the replic- calizer contrast is simply a quick and efficient way to identify the re-
ability of this selectivity in left-out data, as detailed below. levant areas.

6
E. Fedorenko, et al. Cognition 203 (2020) 104348

Fig. 2. Sample stimuli for each condition in Experiments 1–3. Two examples are provided for each condition in each experiment. For Experiment 1, the top row
shows the beginning of a sentence, and the next rows show different possible continuations. For Experiments 2–3, the top row shows one sentence from a pair, and
the next rows show different possibilities for the other sentence in that pair. Red: Lexico-semantic condition; Blue: Syntactic condition; Green: other experimental
conditions; Black: control condition. [NB1: For Experiment 1, the task was passive reading for the critical materials, but, as described in the text, a small number of
(filler) trials contained a memory probe task. NB2: For Experiment 2, three versions of the same base item (corresponding to the Lexico-semantic, Syntactic, and
Global meaning conditions) are presented for illustrative purposes. As detailed in the text, in the actual materials, distinct sets of base items were used for the three
critical conditions in order to match the number of trials across conditions while avoiding sentence repetition.] (For interpretation of the references to colour in this
figure legend, the reader is referred to the web version of this article.)

Fig. 3. Trial structure for Experiments 1–3. One sample trial is shown for each experiment.

7
E. Fedorenko, et al. Cognition 203 (2020) 104348

Table 2
Stimulus and procedure details for each experiment.
Experiment 1: Experiment 2: Experiment 3:
Violations Recovery from adaptation Meaning judgmente

Item description 10-word sentence Two transitive sentences A & B Two sentences A & B
Base items 240c 144 × 3 sets 80
Versions per base itema 4 6 4
Lexico-semantic Criticald: A then B/B then A Lexico-semantic, same
Syntactic Same: A then A/B then B Lexico-semantic, different
Font Different: X then A/Y then B Syntactic, same
Control (X,Y: another item) Syntactic, different
Unique trials 240 × 4 = 960 (144 × 3) × 6 = 2592 80 × 4 = 320
Experimental lists 4 6 4
(240 trials each, 60 per version) (432 trials each, 144 per subset, 24 per version per subset) (80 trials each,
20 per version)
Trial breakdown in a single list 60 Lexico-semantic 48 Critical Lexico-semantic 40 Lexico-semantic
60 Syntactic 48 Critical Syntactic 40 Syntactic
60 Font 48 Critical Global meaning
60 Control 48 × 3 = 144 Same
(+10 fillers each) 48 × 3 = 144 Different
Runs per list 5 (sometimes 4) 6 (sometimes 5) 2
Trials per version per run 12 4 10
(+2 fillers) (24 per subset)
Trial length (s)b 6 4 6
Fixation per run (s) 72 32 120
Run duration (s) 408 (6 min 48 s) 320 (5 min 20s) 360 (6 min)
No. condition orders 10 6 8
Subjects per list 4–6 2–4 3–4

a
See Fig. 2.
b
See Fig. 3.
c
Verbs did not repeat across items, except for “practice” and “read”, which were each used twice. 179 base items were adapted from Kuperberg et al. (2003).
d
Each of the three sets of base items had a different critical condition: Lexico-semantic, Syntactic, or Global meaning (see Fig. 2).
e
Stimuli and procedure followed the design of Dapretto and Bookheimer (1999).

2.4.2. Critical experiments often missed during proofreading (e.g., Schotter, Tran, & Rayner,
The key details about the three experiments are presented in Fig. 2 2014), and noticing missing elements seems harder than the extra
(sample stimuli), Fig. 3 (trial structure), and Table 2 (partitioning of ones.)
stimuli into experimental lists and runs). To construct experimental Overall, there were 240 items. They included 139 items with a
stimuli, we first generated for each experiment a set of “base items” and sentence-final critical verb, taken (and sometimes slightly edited) from
then edited each base item to create several, distinct versions corre- Kuperberg et al. (2003), as well as 61 additional items (to increase
sponding to different experimental conditions. The resulting stimuli power) constructed in a similar manner. Further, to render the timing of
were divided into several experimental lists following a Latin Square violations less predictable, we adapted another 40 items from Kuper-
design, such that in each list (i) stimuli were evenly split across ex- berg et al. such that the critical verb appeared before the final (10th)
perimental conditions, and (ii) only one version of each item was used. word: 6 items had the verb in each of the 3rd through 8th positions, and
Each participant saw materials from a single list, divided into a few 4 items had it in the 9th position. Critical verbs were not repeated
experimental runs. All experiments used an event-related design. Con- across the 240 items, with two exceptions (“practice” and “read” were
dition orders were determined with the optseq2 algorithm (Dale, 1999), used twice each). For each participant, 10 additional sentences were
which was also used to distribute inter-trial fixation periods so as to included in each of the four conditions to serve as fillers. These fillers
optimize our ability to de-convolve neural responses to different ex- were followed by a memory-probe task (deciding whether the probe
perimental conditions. The materials for all experiments and the ex- word appeared in the preceding sentence; Fig. 3) to ensure that parti-
perimental scripts are available from OSF (https://osf.io/abjy9/). cipants paid attention to the task; they were excluded from data ana-
lysis.
2.4.2.1. Experiment 1: Lexico-semantic vs. (morpho-)syntactic
violations. Participants passively read stimuli, and their expectations 2.4.2.2. Experiment 2: Recovery from adaptation to word meanings vs.
were violated in several ways. The items were 10-word sentences, with syntactic structure. Participants were asked to attentively read pairs of
four versions each (Fig. 2): the critical verb (i) resulted in a lexico- sequentially presented sentences and perform a memory probe task at
semantic violation (stimuli that typically elicit an N400 component in the end of each pair (i.e., decide whether a probe word appeared in
ERP studies; see Kutas & Federmeier, 2011, for a review); (ii) resulted in either of the two sentences). The sentences were simple transitive
a morpho-syntactic violation (stimuli that typically elicit a P600 sentences consisting of an agent, a verb, and a patient. Because of
component in ERP studies; e.g., Osterhout & Holcomb, 1992; Hagoort constraints on these materials (as elaborated below), we constructed
et al., 1993); (iii) contained no violations (control condition); or (iv) three sets of items (Fig. 2): (i) sentence pairs that differed only in lexical
was presented in a different font (a low-level oddball violation, items (but had the same syntactic structure and global meaning),
included as an additional, stricter, control condition). Lexico-semantic created by replacing the verb and the agent and patient noun phrases
violations were created by shuffling the critical verbs across the base with synonyms or words closely related in meaning; (ii) pairs that
items. Syntactic violations were created by either omitting a required differed only in their syntactic structure (but had the same lexical items
morpheme (30%) or adding an unnecessary morpheme (70%). (The and global meaning), created by using the Active/Passive alternation;
reason we included a higher proportion of added compared to missing and (iii) pairs that differed only in the global meaning (but had the
morphemes is because form-based error correction mechanisms are so same lexical items and syntactic structure), created by switching the
robust during language comprehension that grammatical errors are two noun phrases, leading to opposite thematic role assignment. The

8
E. Fedorenko, et al. Cognition 203 (2020) 104348

third set was included in order to examine sensitivity to overall two sentences within each pair.) In particular, for the Active/Passive
propositional meaning and to probe combinatorial semantic items, either the occupation noun or the verb could be replaced (by a
processing. Overall, there were 432 items (144 per set). synonym or a word with a different meaning); and for the DO/PP items,
In each set, each sentence pair {A,B} had six versions (Table 2): either the occupation noun or the direct object (inanimate) noun could
sentence A followed by sentence B, and sentence B followed by sentence be replaced.
A (“Critical” condition); sentence A followed by sentence A, and sen-
tence B followed by sentence B (“Same” condition); and, finally, sen- 2.5. Data acquisition, preprocessing, and first-level modeling
tence A followed by a completely different sentence X (lexical items,
syntactic structure, and global meaning were all different), and sen- 2.5.1. Data acquisition
tence B followed by a completely different sentence Y, where the pair Whole-brain structural and functional data were collected on a
{X,Y} was taken from another item (“Different” condition). Every whole-body 3 Tesla Siemens Trio scanner with a 32-channel head coil at
sentence was used once in the Different condition of some other item. the Athinoula A. Martinos Imaging Center at the McGovern Institute for
Therefore, within each of the three sets of items, every sentence ap- Brain Research at MIT. T1-weighted structural images were collected in
peared twice in each condition (Critical, Same, Different). Across the 176 axial slices with 1 mm isotropic voxels (repetition time
three sets, there were overall 5 experimental conditions: Critical Lexico- (TR) = 2530 ms; echo time (TE) = 3.48 ms). Functional, blood oxy-
semantic, Critical Syntactic, Critical Global meaning, Same, and Dif- genation level-dependent (BOLD) data were acquired using an EPI se-
ferent. In order to clearly mark the distinctness of the two identical quence with a 90o flip angle and using GRAPPA with an acceleration
sentences in the Same condition, trials across all conditions included a factor of 2; the following parameters were used: thirty-one 4.4 mm
brief visual mask between the two sentences. thick near-axial slices acquired in an interleaved order (with 10% dis-
To keep the materials semantically diverse, items in the first two tance factor), with an in-plane resolution of 2.1 mm × 2.1 mm, FoV in
sets were constructed to be evenly distributed among three types of the phase encoding (A > > P) direction 200 mm and matrix size
agent-patient relationships: (1) animate agent + inanimate patient; (2) 96 × 96 voxels, TR = 2000 ms and TE = 30 ms. The first 10s of each
animate agent + animate patient, where the relationship is biased so run were excluded to allow for steady state magnetization.
that one of the noun phrases is much more likely to be the agent (e.g.,
The hit man killed the politician); and (3) animate agent + animate pa- 2.5.2. Preprocessing
tient, where the two nouns are similarly likely to be the agent (e.g., The Data preprocessing was carried out with SPM5 (using default
protestor quoted the leader). By virtue of the manipulation of global parameters, unless specified otherwise) and supporting, custom
meaning in the third set, all items had to be semantically reversible MATLAB scripts. (Note that SPM was only used for preprocessing and
(i.e., of the third type). basic modeling—aspects that have not changed much in later versions.
For several datasets, we have directly compared the outputs of data
2.4.2.3. Experiment 3: Same-different meaning judgment on sentences that preprocessed and modeled in SPM5 vs. SPM12, and the outputs are
differ in word meanings vs. syntactic structure. This design was adapted nearly identical (e.g., see Fig. SI-4 in Diachek et al., 2020)) Pre-
from Dapretto and Bookheimer (1999). Participants were asked to processing of anatomical data included normalization into a common
decide whether or not a pair of sequentially presented sentences had space (Montreal Neurological Institute (MNI) template) and resampling
roughly the same meaning. The items were 80 sentence pairs, and each into 2 mm isotropic voxels. Preprocessing of functional data included
pair had four versions (Fig. 2; Table 2): two versions in which the motion correction (realignment to the mean image of the first func-
sentences differed in a single word (Lexico-semantic condition), tional run using 2nd-degree b-spline interpolation), normalization (es-
replaced by either a synonym (Same meaning version) or a non- timated for the mean image using trilinear interpolation), resampling
synonym (Different meaning version); and two versions (Syntactic into 2 mm isotropic voxels, smoothing with a 4 mm FWHM Gaussian
condition) in which the sentences were either syntactic alternations filter and high-pass filtering at 200 s.
differing in both structure and word order (Same meaning version), or
in only structure/only word order (Different meaning version). Half of 2.5.3. Data modeling
the items used the Active/Passive constructions (as in Dapretto & For both the language localizer task and the critical tasks, a standard
Bookheimer), and half – the Double Object (DO)/Prepositional Phrase mass univariate analysis was performed in SPM5 whereby a general
Object (PP) constructions. linear model (GLM) estimated the effect size of each condition in each
A number of features varied and were balanced across items (Fig. 2). experimental run. These effects were each modeled with a boxcar
First, the construction was always the same across the two sentences in function (representing entire blocks/events) convolved with the cano-
the Lexico-semantic condition (balanced between active and passive for nical Hemodynamic Response Function (HRF). The model also included
the Active/Passive items, and between DO and PP for the DO/PP items). first-order temporal derivatives of these effects, as well as nuisance
However, in the Syntactic condition, the construction was always dif- regressors representing entire experimental runs and offline-estimated
ferent in the Same-meaning version because this is how the proposi- motion parameters. The individual activation maps for the language
tional meaning was preserved (again, balanced between active and localizer contrast, and for each condition relative to fixation in each of
passive for the Active/Passive items, and between DO and PP for the the three experiments, as well as raw nifti files for the localizer and the
DO/PP items). For the Different-meaning version, the construction critical experiments are available from OSF (https://osf.io/abjy9/).
could either be the same (in which case the order of the two relevant
nouns was switched) or different (in which case the order of the two 2.6. Definition and validation of language-responsive functional regions of
relevant nouns was preserved). In cases where the construction differed interest (fROIs)
across the two sentences, we balanced whether the first sentence was
active vs. passive (for the Active/Passive items), or whether it was DO For each participant (in each experiment), we defined a set of lan-
vs. PP (for the DO/PP items). The second feature that varied across the guage-responsive functional ROIs using group-constrained, participant-
materials was whether the first-mentioned noun was a name or an oc- specific localization (Fedorenko et al., 2010). In particular, each in-
cupation noun. All items contained one instance of each, with order of dividual participant's map for the sentences > nonwords contrast from
presentation balanced across stimuli. And third, for the Lexico-semantic the language localizer task was intersected with a set of six binary
condition, we varied how exactly the words in the second sentence in a masks. These masks were derived from a probabilistic activation
pair differed from the words in the first. (This does not apply to the overlap map for the language localizer contrast in a large set of parti-
Syntactic condition because the content words were identical across the cipants (n = 220) using the watershed parcellation, as described in

9
E. Fedorenko, et al. Cognition 203 (2020) 104348

Fedorenko et al. (2010), and corresponded to relatively large areas each condition (Lexico-semantic and Syntactic) against the low-level
within which most participants showed activity for the target contrast. fixation baseline to ensure robust responses in the language regions to
These masks covered the fronto-temporal language network: three in sentence comprehension. (Note that fixation was used here because,
the left frontal lobe falling within the IFG, its orbital portion, and the unlike in the other two experiments, there was no other baseline con-
MFG, and three in the temporal and parietal cortex (Fig. 5). Within each dition following Dapretto and Bookheimer's (1999) design.) In each of
mask, a participant-specific language fROI was defined as the top 10% these reality-check analyses, the results were FDR-corrected (Benjamini
of voxels with the highest t-values for the localizer contrast. This top n% & Yekutieli, 2001) for the six regions.
approach ensures that fROIs can be defined in every participant and
that their sizes are the same across participants, allowing for general- 2.7.2. Critical analyses (a): Directly comparing the Lexico-semantic and
izable results (e.g., Nieto-Castañón & Fedorenko, 2012). Syntactic conditions in the language fROIs
Before examining the data from the critical experiments, we ensured Next, we directly compared the Lexico-semantic and Syntactic
that the language fROIs show the expected signature response (i.e., that conditions in each region in each experiment. If a brain region is se-
the response is reliably greater to sentences than nonwords). To do so, lectively or preferentially engaged in syntactic processing, then we
we used an across-runs cross-validation procedure (e.g., Nieto-Castañón would expect to observe a reliably stronger response to the Syntactic
& Fedorenko, 2012), where one run of the localizer is used to define the condition than the Lexico-semantic condition. And if a brain region is
fROIs, and the other run to estimate the responses, ensuring in- selectively/preferentially engaged in lexico-semantic processing, we
dependence (e.g., Kriegeskorte et al., 2009). As expected, and re- would expect to observe the opposite pattern. In these analyses, we
plicating prior work (e.g., Blank et al., 2016; Fedorenko et al., 2010; report the results without a correction for the number of regions be-
Fedorenko, Behr, & Kanwisher, 2011; Mahowald & Fedorenko, 2016), cause we wanted to give syntactic selectivity—which we are arguing
the language fROIs showed a robust sentences > nonwords effect against—the best chance to reveal itself. (Of course, if an uncorrected p-
(ts(48) > 8.44; ps < 0.0001), correcting for the number of regions value fails to reach significance, then the corrected one does, too.)
(six) using the False Discovery Rate (FDR) correction (Benjamini &
Yekutieli, 2001). 2.7.3. Critical analyses (b): Searching for voxels selective for syntactic (or
lexico-semantic, for completeness) processing
2.7. Estimating the responses of the language fROIs to the conditions of the One potential concern with the use of language fROIs is that each
critical experiments fROI is relatively large and the responses are averaged across voxels
(e.g., Friston et al., 2006). Thus, fROI-based analyses may obscure un-
We estimated the responses in the language fROIs to the conditions derlying functional heterogeneity and potential selectivity for syntactic
of each critical experiment: the Control condition, Lexico-semantic processing. For example, if a fROI contains a subset of voxels that show
violations, Syntactic violations, and Font violations in Experiment 1; a stronger response to lexico-semantic than syntactic processing, and
the Same condition, Different condition, and three Critical conditions another subset of voxels that show a stronger response to syntactic than
(differing in only lexical items, only syntactic structure, or only global lexico-semantic processing, we may not detect a difference at the level
meaning) in Experiment 2; and the Lexico-semantic and Syntactic of the fROI as a whole. To circumvent this concern, we supplemented
conditions (each collapsed across same and different pairs) in the analyses of language fROIs, with analyses that i) use some of the
Experiment 3. Statistical comparisons were performed on the estimated data from each critical experiment to directly search for voxels—within
percent BOLD signal change (PSC) values in each region in each ex- the same broad masks encompassing the language network—that re-
periment. spond more strongly to syntactic than lexico-semantic processing (i.e.,
We analyzed each experiment separately to allow for the possibility top 10% of voxels based on the Syntactic > Lexico-Semantic contrast),
that syntax selectivity would be observed in just one of the experiments. or vice versa (for completeness), and then ii) examine the replicability
Such a pattern could still be potentially informative and would be of this pattern of response in a left-out portion of the data. We per-
missed in an analysis that pools data from the three experiments. formed this analysis for each of Experiments 1–3. If any (even non-
Furthermore, we examined each region separately, in line with our contiguous) voxels with reliably stronger responses to syntactic pro-
research question: whether any region within the language network is cessing exist anywhere within the fronto-temporal language network,
selective for syntactic over lexico-semantic processing. this analysis should be able to detect them. For these analyses, we used
one-tailed paired-samples t-tests because these hypotheses are direc-
2.7.1. Reality-check analyses: Testing for sensitivity to lexico-semantic and tional. For example, when examining voxels that show stronger re-
syntactic processing sponses to syntactic than lexico-semantic processing to test whether this
First, we tested for basic sensitivity to lexico-semantic and syntactic preference is replicable in left-out data, the critical contrast is Syn-
manipulations. In each region, we used two-tailed paired-samples t- tactic > Lexico-semantic. As in the last set of analyses, these results
tests to compare the response to each critical (lexico-semantic and were not corrected for the number of regions because we wanted to give
syntactic) condition to one or more control conditions. In Experiment 1, syntactic selectivity the best chance to reveal itself.
we compared the response to each critical violation condition (Lexico-
semantic or Syntactic) against a) the Control condition with no viola- 3. Results
tions, and, as an additional, stricter, baseline, b) the Font violation
condition. In Experiment 2, we compared the Same and Different 3.1. Behavioral results
conditions to each other (a reality check to test for recovery from
adaptation in the language regions when all the features of the sentence Error rates and reaction times (RTs), for trials with a recorded re-
change), and then we compared each of the Critical conditions to the sponse, in each of the three experiments are summarized in Fig. 4.
Same condition (where the same sentence is repeated exactly) to test for Performance on the memory probe task in the filler trials in Experiment
recovery from adaptation when the lexical items or the syntactic 1 was close to ceiling (between 95.4% and 96.6% across conditions),
structure changes. The predictions are similar for Experiments 1 and 2: with no reliable difference between the two critical—Lexico-semantic
if a brain region is sensitive to lexical processing, the Lexico-semantic and Syntactic—conditions, in accuracies or RTs (ts(21) < 1, n.s.). In
condition should elicit a stronger response than the control condition Experiment 2, performance on the memory probe task varied between
(s); similarly, if a brain region is sensitive to syntactic processing, the 72.6% and 95.7% across conditions. As expected, participants were
Syntactic condition should elicit a stronger response than the control faster and more accurate in the Same condition, where the same sen-
condition(s). Finally, in Experiment 3, we compared the response to tence was repeated, than in the Different condition, where the two

10
E. Fedorenko, et al. Cognition 203 (2020) 104348

reliably stronger response than the Control (no violations) condition in

each of six language fROIs (ts(21) > 2.77, ps < 0.05), and than the
Font violations condition in five of the six fROIs (ts(21) > 3.43, ps <
0.05), with the MFG fROI not showing a significant effect. The Syn-
tactic condition elicited a reliably stronger response than the Control
condition in two language fROIs: IFGorb and IFG (ts(21) > 2.84,
ps < 0.05). However, the Syntactic condition did not reliably differ
from the Font violations condition in any of the fROIs (ts(21) < 1.45,
n.s.), suggesting that language regions are not recruited more strongly
when people encounter syntactic violations than they are when people
encounter low-level perceptually unexpected features in the linguistic
input (see also Vissers, Chwilla, & Kolk, 2006; van de Meerendonk,
Indefrey, Chwilla, & Kolk, 2011; see Mollica et al., 2020, for extension
to word-order violations). In Experiment 2, the Different condition—-
where the two sentences in a pair differed in lexical items, syntactic
structure, and global meaning—elicited a reliably stronger response
than the Same condition, where the two sentences in a pair were
identical, in four language fROIs: IFG, MFG, AntTemp, and PostTemp
(ts(13) > 3.45, ps < 0.05); the effect was not reliable in the IFGorb
and AngG fROIs. Further, the Lexico-semantic condition elicited a re-
sponse that was reliably stronger than the Same condition in all lan-
guage fROIs, except for the AngG fROI (ts(13) > 2.71, ps < 0.05), and
similarly, the Syntactic condition elicited a response that was reliably
stronger than the Same condition in all language fROIs, except for the
AngG fROI (ts(13) > 2.33, ps < 0.05).2 Finally, in Experiment 3, both
experimental conditions elicited responses that were reliably above the
fixation baseline in all six fROIs (Lexico-semantic: ts(14) > 4.58,
ps < 0.05; Syntactic: ts(14) > 2.66, ps < 0.05).

3.2.2. Critical analyses (a): Directly comparing the Lexico-semantic and

Fig. 4. Summary of the behavioral results from Experiments 1–3. Each column
shows the results from a different experiment: left – Experiment 1: Violations;
Syntactic conditions in the language fROIs
middle – Experiment 2: Recovery from Adaptation; right – Experiment 3: The results for the three experiments are summarized in Fig. 5 and
Meaning Judgment. Top graphs: error rates; bottom graphs: reaction times. Table 3. In Experiment 1, a direct comparison between the Lexico-se-
Here and in all subsequent figures, for each condition and each measure, dots mantic and Syntactic conditions revealed reliably stronger responses to
correspond to individual participants; the bar shows the average across these the Lexico-semantic condition in all language fROIs except for the MFG
participants, and the error bar shows standard error of the mean (across par- fROI (ts(21) > 2.12, ps < 0.05). In Experiment 2, no language fROI
ticipants). Significant differences between the critical, Lexico-semantic (red) showed reliably stronger recovery from adaptation in the Lexico-se-
and Syntactic (blue), conditions are marked with *’s. (For interpretation of the mantic than Syntactic condition or vice versa (ts(13) < 1.44, n.s.). Fi-
references to colour in this figure legend, the reader is referred to the web nally, in Experiment 3, we observed a stronger response to the Lexico-
version of this article.)
semantic than Syntactic condition in two language fROIs—IFGorb and
AntTemp (ts(14) > 2.27, ps < 0.05). No language fROI showed the
sentences in the pair differed in all respects (ts(13) > 15.1, ps < 0.05). opposite pattern.
Furthermore, participants were faster and more accurate in the Syn-
tactic condition than in the Lexico-semantic condition (ts(13) > 3.55,
3.2.3. Critical analyses (b): Searching for voxels selective for syntactic (or
ps < 0.05), plausibly because the words were repeated between the
lexico-semantic, for completeness) processing
two sentences in the pair in the Syntactic, but not Lexico-semantic
The results for the three experiments are summarized in Figs. 6 and
condition. However, the difference was small, with high performance
7 and Table 4. In Experiment 1, when we defined the individual fROIs
(> 89.7%) in both conditions. Finally, in line with the behavioral re-
by the Syntactic > Lexico-semantic contrast, we did not find a replic-
sults in Dapretto and Bookheimer (1999), in Experiment 3, performance
able (across runs) Syntactic > Lexico-semantic effect within any of the
on the meaning judgment task did not differ between the Lexico-se-
masks. In contrast, when we defined the fROIs by the Lexico-se-
mantic and Syntactic conditions, in accuracies or RTs (ts(14) < 1, n.s.).
mantic > Syntactic contrast, we found a replicable Lexico-semantic >
In summary, in each of the three experiments, performance was high
Syntactic effect within all the language masks, except for the MFG mask
across conditions, with no reliable differences between the Lexico-se-
(ts(21) > 2.75, ps < 0.05), consistent with our finding of stronger
mantic and Syntactic conditions in Experiments 1 and 3, and only a
responses for the Lexico-semantic than Syntactic condition in the fROIs
small difference in Experiment 2. We can thus proceed to examine
that were defined by the language localizer contrast.
neural differences between lexico-semantic and syntactic processing
In Experiment 2, when we defined the individual fROIs by the
without worrying about those differences being driven by systematic
Syntactic > Lexico-semantic contrast, we found a slightly higher
differences in processing difficulty.

2
It is worth noting that, similar to the Lexico-semantic and Syntactic condi-
3.2. fMRI results
tions, the Global meaning condition also elicited a response that was reliably
stronger than the Same condition in all language fROIs (ps < 0.05). This effect
3.2.1. Reality-check analyses: Testing for sensitivity to lexico-semantic and provides evidence that language regions are sensitive to differences in complex
syntactic processing meanings above and beyond the meanings of individual words (given that the
The results for the three experiments are summarized in Fig. 5 and only thing that differs between the sentences in a pair in the Global-meaning
Table 3. In Experiment 1, the Lexico-semantic condition elicited a condition is word order).

11
E. Fedorenko, et al. Cognition 203 (2020) 104348

Fig. 5. Responses in language fROIs to the condi-

tions in Experiments 1–3. Responses (beta weights
from a GLM) are measured as PSC relative to the
fixation baseline. Each panel shows the results from
a different experiment: (A) Experiment 1: Violations;
(B) Experiment 2: Recovery from Adaptation; (C)
Experiment 3: Meaning Judgment. Within each
panel, each group of bars shows data from a different
fROI (legend at the bottom). Significant differences
between the critical, Lexico-semantic (red) and
Syntactic (blue), conditions are marked with *’s.
(The significance of the reality-check analyses that
establish the sensitivity of the language fROIs to the
lexico-semantic and syntactic manipulations relative
to the control conditions is not marked; please see
Results section and Table 3.) Brain images at the
bottom show the broad masks used to constrain the
selection of the individual fROIs (these are not the
fROIs themselves, which were defined using the
sentences > nonwords contrast separately in each
participant, as described in the text, and could thus
vary within the borders of the masks depicted here).
(For interpretation of the references to colour in this
figure legend, the reader is referred to the web ver-
sion of this article.)

response to the Syntactic than Lexico-semantic condition within the 4. Discussion

PostTemp mask (t(14) = 1.84, p < .05). This effect would not survive
correction for the number of regions. And even if, with a larger number 4.1. Summary of the findings and their relationship to earlier proposals of
of participants, the statistical robustness of the Syntactic > Lexico-se- the neural architecture of language
mantic effect increased, the difference between the two conditions
would remain small. When we defined the fROIs by the Lexico- Across three fMRI experiments that used paradigms from the prior
semantic > Syntactic contrast, we did not find a replicable Lexico- language literature but relied on a more sensitive and rigorous analytic
semantic > Syntactic effect in any of the language masks, in line with approach—functional localization at the individual-subject level (Brett
the two critical conditions not eliciting differential responses in the et al., 2002; Fedorenko et al., 2010; Nieto-Castañón & Fedorenko, 2012;
fROIs defined by the language localizer contrast in this experiment. Saxe et al., 2006)—we searched for syntactic selectivity within the
Finally, in Experiment 3, as in Experiment 1, when we defined the fronto-temporal language network, asking whether any part of this
individual fROIs by the Syntactic > Lexico-semantic contrast, we did network responds reliably more strongly to syntactic than lexico-se-
not find a replicable Syntactic > Lexico-semantic effect within any of mantic processing. No such selectivity was found. In particular, no
the masks; but when we defined the fROIs by the Lexico- language fROI, defined with a robust language localizer contrast
semantic > Syntactic contrast, we found a replicable Lexico- (Fedorenko et al., 2010), showed stronger responses to syntactic pro-
semantic > Syntactic effect within all the language masks, except for cessing than lexico-semantic processing in any experiment (Fig. 5).
the MFG mask (ts(14) > 2.14, ps < 0.05). Even sets of voxels defined by their stronger responses to syntactic than
lexico-semantic conditions using half of the data did not show a re-
plicable Syntactic > Lexico-semantic effect in the other half of the data

12
E. Fedorenko, et al. Cognition 203 (2020) 104348

Table 3
Responses of language fROIs in Experiments 1–3: mean PSC with standard error (by participants), effect size (Cohen's d), t-value, and p-value.a
fROI Exp. 1: violations Exp. 2: adaptation recovery Exp. 3: meaning judgments
(size)
LexSem Synt LexSem Synt LexSem Diff LexSem Synt LexSem LexSem Synt LexSem
vs. vs. vs. vs. vs. vs. vs. vs. vs. vs. vs. vs.
Control Control Font Font Synt Same Same Same Synt Baseline Baseline Synt

IFGorb 0.44 0.19 0.38 0.13 0.25 0.09 0.21 0.13 0.07 1.46 1.18 0.28
(37 voxels) ( ± 0.07) ( ± 0.07) ( ± 0.08) ( ± 0.09) ( ± 0.08) ( ± 0.05) ( ± 0.08) ( ± 0.04) ( ± 0.07) ( ± 0.21) ( ± 0.21) ( ± 0.12)
d = 1.37 d = 0.61 d = 0.97 d = 0.31 d = 0.64 d = 0.52 d = 0.73 d = 0.83 d = 0.28 d = 1.80 d = 1.46 d = 0.59
t = 6.40 t = 2.85 t = 4.54 t = 1.44 t = 3.01 t = 1.93 t = 2.72 t = 3.10 t = 1.05 t = 6.96 t = 5.67 t = 2.28
p < 10−5 p = .0096 p = .0002 p = .16 p = .0067 p = .075 p = .18 p = .0084 p = .31 p < 10−5 p < 10−4 p = .039
IFG 0.46 0.22 0.35 0.12 0.24 0.21 0.31 0.22 0.09 1.48 1.32 0.16
(74 voxels) ( ± 0.08) ( ± 0.06) ( ± 0.09) ( ± 0.09) ( ± 0.08) ( ± 0.05) ( ± 0.08) ( ± 0.08) ( ± 0.08) ( ± 0.21) ( ± 0.20) ( ± 0.12)
d = 1.17 d = 0.72 d = 0.88 d = 0.28 d = 0.64 d = 1.08 d = 1.02 d = 0.74 d = 00.29 d = 1.78 d = 1.72 d = 0.34
t = 5.51 t = 3.37 t = 4.12 t = 1.30 t = 3.02 t = 4.02 t = 3.80 t = 2.77 t = 1.09 t = 6.89 t = 6.68 t = 1.30
p < 10−4 p = .0029 p = .0005 p = .21 p = .0065 p = .0014 p = .0022 p = .016 p = .29 p < 10−5 p < 10−4 p = .21
MFG 0.22 0.14 0.16 0.08 0.09 0.23 0.32 0.32 0.00 1.39 1.41 −0.01
(46 voxels) ( ± 0.08) ( ± 0.09) ( ± 0.09) ( ± 0.09) ( ± 0.12) ( ± 0.07) ( ± 0.10) ( ± 0.09) ( ± 0.08) ( ± 0.22) ( ± 0.29) ( ± 0.12)
d = 0.59 d = 0.33 d = 0.39 d = 0.17 d = 0.16 d = 0.93 d = 0.90 d = 0.92 d = 0.01 d = 1.64 d = 1.25 d = −0.03
t = 2.78 t = 1.56 t = 1.81 t = 0.82 t = 0.74 t = 3.47 t = 3.35 t = 3.46 t = 0.04 t = 6.36 t = 4.85 t = −0.12
p = .011 p = .13 p = .084 p = .042 p = .47 p = .0042 p = .0052 p = .0043 p = .97 p < 10−4 p = .0002 p = .90
AntTemp 0.13 −0.02 0.14 −0.01 0.15 0.15 0.17 0.10 0.07 0.54 0.36 0.19
(162 ( ± 0.03) ( ± 0.02) ( ± 0.04) ( ± 0.04) ( ± 0.04) ( ± 0.04) ( ± 0.05) ( ± 0.04) ( ± 0.05) ( ± 0.11) ( ± 0.11) ( ± 0.06)
voxels) d = 0.83 d = −0.15 d = 0.85 d = −0.04 d = 0.87 d = 0.94 d = 0.95 d = 0.62 d = 0.38 d = 1.26 d = 0.87 d = 0.81
t = 3.90 t = −0.71 t = 3.99 t = −0.18 t = 4.10 t = 3.51 t = 3.56 t = 2.34 t = 1.43 t = 4.87 t = 3.36 t = 3.14
p = .0008 p = .49 p = .0007 p = 0.86 p = .0005 p < .0038 p = .0035 p = .036 p = .17 p = .0002 p = .0047 p = .0072
PostTemp 0.20 0.07 0.18 0.06 0.12 0.20 0.25 0.24 0.01 0.98 1.00 −0.02
(294 ( ± 0.04) ( ± 0.04) ( ± 0.05) ( ± 0.07) ( ± 0.06) ( ± 0.04) ( ± 0.06) ( ± 0.06) ( ± 0.07) ( ± 0.10) ( ± 0.14) ( ± 0.08)
voxels) d = 1.16 d = 0.38 d = 0.80 d = 0.18 d = 0.45 d = 1.22 d = 1.05 d = 1.07 d = 0.05 d = 2.43 d = 1.85 d = −0.06
t = 5.42 t = 1.80 t = 3.76 t = 0.85 t = 2.13 t = 4.58 t = 3.91 t = 3.99 t = 0.20 t = 9.43 t = 7.16 t = −0.22
p < 10−4 p = .086 p = .0011 p = .40 p = .045 p = .0005 p = .0018 p = .0015 p = .84 p < 10−6 p < 10−5 p = .83
AngG 0.19 −0.06 0.22 −0.03 0.25 0.07 0.06 0.11 −0.05 0.80 0.61 0.19
(64 voxels) ( ± 0.06) ( ± 0.04) ( ± 0.06) ( ± 0.06) ( ± 0.06) ( ± 0.05) ( ± 0.06) ( ± 0.05) ( ± 0.07) ( ± 0.17) ( ± 0.23) ( ± 0.14)
d = 0.73 d = −0.30 d = 0.73 d = −0.12 d = 0.95 d = 0.43 d = 0.28 d = 0.58 d = −0.21 d = 1.18 d = 0.69 d = 0.34
t = 3.40 t = −1.42 t = 3.44 t = −0.58 t = 4.48 t = 1.59 t = 1.04 t = 2.16 t = −0.79 t = 4.59 t = 2.67 t = 1.33
p = .0027 p = .17 p = .0024 p = .57 p = .0002 p = .13 p = .32 p = .054 p = 0.44 p < .0004 p = .018 p = .20

a
Non-significant effects at α = 0.05 (FDR-corrected for the number of regions) are shaded in grey; marginal effects (0.05 < p < 0.1) are shaded in light grey.

(Fig. 6), with one exception of a small difference between the Syntactic et al., 2014; Friederici, 2011, 2012; Grodzinsky & Santi, 2008; Matchin
and Lexico-semantic conditions in one experiment that would not sur- & Hickok, 2019; Pylkkänen, 2019; Tyler et al., 2011; Ullman, 2016).
vive correction for the number of regions. Across such proposals (see next section), a syntax-/combinatorics-se-
Importantly, this failure to uncover any syntactically-selective re- lective component is argued to not support the storage and processing
gions/voxels within the language network is not due to lack of power. of individual word meanings, as illustrated in the architectures in
Specifically, the lack of syntactic selectivity stands in sharp contrast to Fig. 1a–d. In contrast to these proposals, it appears that any brain re-
a) sensitivity to both lexico-semantic and syntactic manipulations—in gion/set of voxels within the language network that shows sensitivity to
at least a subset of the language fROIs—relative to the control condi- syntactic manipulations also shows sensitivity to manipulations tar-
tions, where present (in Experiments 1 and 2) (Fig. 5, Table 3); and b) geting the processing of individual word meanings.
stronger responses to lexico-semantic than syntactic conditions, re- Not all cognitive neuroscience proposals of the language archi-
plicable across some experiments (Experiments 1 and 3) and analyses tecture postulate a distinction between syntactic and semantic re-
(Figs. 5 and 7, Tables 3–4). And although the lack of syntactic se- presentations/processing, or between combinatorial processing and
lectivity in the current study appears to run contrary to earlier brain stored knowledge representations (e.g., Bornkessel-Schlesewsky &
imaging reports of dissociable effects of syntactic and semantic pro- Schlesewsky, 2009; Bornkessel-Schlesewsky et al., 2015). For example,
cessing (e.g., Cooke et al., 2006; Dapretto & Bookheimer, 1999; Embick Bornkessel-Schlesewsky et al. (2015) have suggested that the language
et al., 2000; Friederici et al., 2010; Glaser et al., 2013; Kuperberg et al., network, as a whole, supports composition—combining smaller lin-
2000; Kuperberg et al., 2003; Newman et al., 2001; Ni et al., 2000; guistic units into larger ones—in the service of meaning extraction (see
Noppeney & Price, 2004; Schell et al., 2017, inter alia), none of those also Mollica et al., 2020, for further empirical support of this idea). The
earlier studies had compellingly established selectivity for syntactic current results align well with these proposals.
processing in a robust (across multiple sets of participants/materials)
and generalizable (across diverse manipulations that aim to isolate the
4.2. The bias of the language network toward lexico-semantic processing
same cognitive process, and relative to diverse control conditions) way,
as discussed in the Introduction.
In two of our experiments, lexico-semantic conditions elicited nu-
The current results—along with earlier results from paradigms that
merically, and sometimes reliably, stronger responses than syntactic
have varied the presence/absence of lexico-semantic vs. syntactic in-
conditions in many of the language fROIs. Furthermore, in Experiment
formation in the linguistic signal (e.g., Bautista & Wilson, 2016; Bedny,
1, unlike the lexico-semantic violation condition, the syntactic violation
Pascual-Leone, Dodell-Feder, Fedorenko, & Saxe, 2011; Fedorenko
condition did not elicit a response that was higher than a low-level (font
et al., 2010; Fedorenko, Nieto-Castanon, & Kanwisher, 2012; Fedorenko
violation) condition in the language regions. These overall stronger
et al., 2016 (PNAS); Pallier et al., 2011)—pose a challenge for proposals
responses to semantic than syntactic conditions are in line with two
of the neural architecture of language that postulate syntax- or com-
prior findings. First, using multivariate analyses, we have previously
binatorics-selective brain regions (e.g., Baggio & Hagoort, 2011; Duffau
found that lexico-semantic information is represented more robustly

13
E. Fedorenko, et al. Cognition 203 (2020) 104348

Fig. 6. Responses in fROIs defined by the Syntactic > Lexico-semantic contrast to the critical conditions in Experiments 1–3. Participant-specific fROIs were defined,
within the borders of each mask (Fig. 5), as the top 10% of voxels showing the strongest Syntactic > Lexico-semantic contrast effect in the corresponding experiment.
These fROIs were defined based on half the data from that experiment, and then the other (independent) half were used to estimate the effect size of this same
contrast (i.e., estimate the replicability of the contrast effect). Conventions are the same as in Fig. 5, with one exception: in panels A and C, parts of the y-axis at the
top or bottom have been cut out (marked by two parallel horizontal tick marks) in order to stretch the bars more and accentuate differences across conditions when
those appeared. In these visually edited cases, distance between the most extreme 1–2 data points and their corresponding bars are not at scale. Those data points are
colored in grey. Differences between the Lexico-semantic (red) and Syntactic (blue) conditions are marked with *’s. (For interpretation of the references to colour in
this figure legend, the reader is referred to the web version of this article.)

than syntactic information in the language system (Fedorenko, Nieto- Tamariz, Cornish, & Smith, 2015), and with the fact that most of our
Castanon, & Kanwisher, 2012; see Wang et al., 2020 for related evi- knowledge of language has to do with lexical semantics (word mean-
dence from children). In particular, pairs of conditions that differ in ings), with only a small number of bits needed to store all of our syn-
whether or not they contain lexico-semantic information (e.g., sen- tactic knowledge (Mollica & Piantadosi, 2019). And it is not consistent
tences vs. Jabberwocky sentences, or lists of words vs. lists of non- with syntax-centric views of language, especially with the construal of
words) are more robustly dissociable in the fine-grained patterns of linguistic syntax as an abstract computation not sensitive to the nature
activity than pairs of conditions that differ in whether or not they are of the units being combined (e.g., Berwick, Friederici, Chomsky, &
structured (e.g., sentences vs. lists of words, or Jabberwocky sentences Bolhuis, 2013; Chomsky & Dinozzi, 1972; Friederici, 2018; Friederici
vs. lists of nonwords). And second, in ECoG, we observed reliably et al 2017; Friederici et al., 2006; Hauser, Chomsky, & Fitch, 2002;
stronger responses to conditions that only contain lexico-semantic in- Pinker, 1995).
formation (word lists) than conditions that only contain syntactic in- One implication of these, and earlier behavioral, results discussed in
formation (Jabberwocky) in many language-responsive electrodes the Introduction is that artificial grammar learning and processing
(Fedorenko et al., 2016 (PNAS)), but no electrodes showed the opposite paradigms (e.g., Reber, 1967)—where structured sequences of mean-
pattern. Along with the current study, these results demonstrate that ingless units (e.g., syllables) are used in an attempt to approximate
the magnitude and spatial organization of responses in the human human syntax (e.g., Friederici et al., 2006; Petersson, Folia, & Hagoort,
language network are determined more by meaning than structure. 2012; Wang et al., 2015)—have limited utility for understanding
This bias toward lexico-semantic processing fits with the view that human language, given that syntactic representations and processing
the goal of language is communication, i.e., the transfer of meanings seem to be inextricably linked with representations of linguistic
across minds (e.g., Gibson et al., 2019; Goldberg, 2006; Hahn, Jurafsky, meaning (see also Fedor, Varga, & Szathmáry, 2012).
& Futrell, 2020; Hurford, 1998, 2007; Jackendoff, 2011; Kirby,

14
E. Fedorenko, et al. Cognition 203 (2020) 104348

Fig. 7. Responses in fROIs defined by the Lexico-semantic > Syntactic contrast to the critical conditions in Experiments 1–3. This figure depicts data from a parallel
analysis to that depicted in Fig. 6; here, participant-specific fROIs were defined as the top 10% of voxels showing the strongest Lexico-semantic > Syntactic contrast
effect in the corresponding experiment, and the size of this contrast was then estimated in an independent part of the data (this is the opposite contrast to the one used
in Fig. 6). Conventions are the same as in Figs. 5, 6, with the addition of the following: non-significant effects with p < .10 are marked with *s above tildes.

4.3. Limitations and thus may not be ideal for isolating syntactic computations. In ad-
dition, the use of a word memory probe task in Experiments 1 and 2
It is worth acknowledging some limitations of the current study. As may have biased participants toward lexico-semantic processing (al-
already noted in the Introduction, no single paradigm developed for though see Diachek et al., 2020 and Ivanova, Siegelman, et al., in prep,
probing lexico-semantic and syntactic processing is perfect. In past for evidence that task demands do not strongly modulate neural re-
work, we have relied on somewhat unnatural stimuli that do or do not sponses in the language network). Importantly, to the extent that syn-
contain lexico-semantic or syntactic information (e.g., Fedorenko et al., tactic selectivity has been inferred from the kinds of paradigms we use
2010; see also Pallier et al., 2011; Bautista & Wilson, 2016, inter alia). here, we show that these findings don't survive when using methods
In the current study, we instead adopted three paradigms from the with superior sensitivity. Furthermore, the lack of syntactic selectivity
literature in an effort to conceptually replicate the previously reported reported here converges with findings from other paradigms we have
dissociations. All three paradigms rely on relatively natural-sounding used in earlier work (Fedorenko et al., 2010; Fedorenko, Nieto-
sentence materials and do not suffer from difficulty confounds. How- Castanon, & Kanwisher, 2012; Fedorenko et al., 2016 (PNAS)).
ever, these paradigms still have limitations. For example, the violations In addition, the study's scope is limited in several ways. First, per-
paradigm used in Experiment 1 uses morpho-syntactic violations in the haps some cells/circuits/cortical-layer-specific areas are selective for
Syntactic condition. As acknowledged earlier, language comprehension (some aspect) of syntactic/combinatorial processing (e.g., the archi-
mechanisms are highly robust to noise (e.g., Ferreira et al., 2002; Levy tecture in Fig. 1e), but we are not able to detect this selectivity due to
et al., 2009; Gibson et al., 2013), and small, form-based, including the relatively coarse spatial resolution of our method. This possibility is
grammatical, errors are often missed during proofreading (e.g., Schotter hard to rule out without resorting to approaches like single-cell re-
et al., 2014). It is therefore possible that this manipulation was too cordings (e.g., Engel, Moll, Fried, & Ojemann, 2005; Mukamel & Fried,
subtle. Note, however, that morpho-syntactic violations like the ones 2012) or laminar imaging (e.g., Norris & Polimeni, 2019). However, it
used here, do consistently elicit a robust P600 effect in ERP investiga- is worth noting that at least for the paradigm that varies the presence/
tions (e.g., Hagoort et al., 1993; Osterhout & Holcomb, 1992), in- absence of lexico-semantic vs. syntactic information in the linguistic
dicating that comprehenders do register these errors. Similarly, syn- signal, the results from an ECoG study (Fedorenko et al., 2016
tactic priming in comprehension is notoriously weak, and—as discussed (PNAS))—where the spatial resolution is substantially higher than in
below—does not reflect purely syntactic processing (e.g., Mahowald, fMRI—closely mirrored the fMRI results (e.g., Fedorenko et al., 2010).
James, Futrell, & Gibson, 2016; Ziegler, Snedeker, & Wittenberg, 2018) Another possibility is that no single brain region is selective for

15
E. Fedorenko, et al. Cognition 203 (2020) 104348

Table 4
Replicability (in left-out data) of the critical contrasts in Experiments 1–3.a,b
fROI Exp. 1: Exp. 2: Exp. 3:
(size) violations adaptation recovery meaning judgments

LexSem > Synt > LexSem > Synt > LexSem > Synt >
Synt LexSem Synt LexSem Synt LexSem

IFGorb 0.24 0.03 0.05 0.00 0.46 −0.10

(37 voxels) ( ± 0.08) ( ± 0.08) ( ± 0.12) ( ± 0.04) ( ± 0.15) ( ± 0.10)
d = 0.75 d = 0.07 d = 0.11 d = −0.01 d = 0.79 d = −0.25
t = 3.54 t = 0.35 t = 0.40 t = −0.03 t = 3.06 t = −0.95
p = .0010 p = .63 p = .35 p = .49 p = .004 p = .82
IFG 0.26 0.00 0.21 −0.03 0.5 −0.08
(74 voxels) ( ± 0.11) ( ± 0.07) ( ± 0.15) ( ± 0.06) ( ± 0.14) ( ± 0.08)
d = 0.59 d = −0.01 d = 0.37 d = −0.13 d = 0.94 d = −0.24
t = 2.76 t = −0.02 t = 1.40 t = −0.47 t = 3.63 t = −0.93
p = .0059 p = .51 p = .09 p = .68 p = .0014 p = .81
MFG 0.13 0.08 0.08 −0.02 0.07 0.03
(46 voxels) ( ± 0.16) ( ± 0.11) ( ± 0.10) ( ± 0.07) ( ± 0.09) ( ± 0.13)
d = 0.20 d = −0.15 d = 0.20 d = −0.07 d = 0.19 d = 0.05
t = 0.93 t = 0.72, t = 0.74 t = −0.25 t = 0.72 t = 0.21
p = .18 p = .24 p = .24 p = .60 p = .24 p = .42
AntTemp 0.13 −0.02 0.03 −0.01 0.22 −0.06
(162 voxels) ( ± 0.04) ( ± 0.03) ( ± 0.06) ( ± 0.03) ( ± 0.06) ( ± 0.04)
d = 0.74 d = −0.10 d = 0.15 d = −0.05 d = 0.97 d = −0.44
t = 3.50 t = −0.48 t = 0.57 t = −0.19 t = 3.77 t = −1.7
p = .0011 p = .68 p = .29 p = .57 p = .0010 p = .94
PostTemp 0.15 0.00 0.08 0.08 0.17 0.06
(294 voxels) ( ± 0.05) ( ± 0.05) ( ± 0.08) ( ± 0.03) ( ± 0.07) ( ± 0.07)
d = 0.63 d = −0.02 d = 0.26 d = 0.80 d = 0.67 d = 0.21
t = 2.97 t = 0–0.08 t = 0.96 t = 2.99 t = 2.59 t = 0.83
p = .0036 p = .53 p = .18 p = .0051 p = .011 p = .21
AngG 0.21 0.05 −0.05 0.07 0.43 −0.03
(64 voxels) ( ± 0.06) ( ± 0.08) ( ± 0.05) ( ± 0.07) ( ± 0.16) ( ± 0.10)
d = 0.71 d = −0.12 d = −0.26 d = 0.26 d = 0.68 d = −0.08
t = 3.34 t = −0.58 t = −0.99 t = 0.96 t = 2.59 t = −0.31
p = .0015 p = .71 p = .83 p = .18 p = .0099 p = .62

a
The Lexico-semantic > Syntactic effect is tested in Lexico-semantic > Syntactic fROIs, and the Syntactic > Lexico-semantic effect is tested in Syntactic > Lexico-
semantic fROIs.
b
Conventions are the same as in Table 3.

syntactic/combinatorial processing, but inter-region interaction/syn- across domains, including language, arithmetic, music, and action ob-
chronization—perhaps restricted to particular frequency bands (e.g., servation/planning (e.g., Fadiga, Craighero, & D'Ausilio, 2009; Fitch &
Giraud & Poeppel, 2012; Martin & Doumas, 2019; Meyer, 2018)—is Martins, 2014; Koechlin & Jubault, 2006; Tettamanti & Weniger, 2006).
critical for syntactic structure building. Some have argued that the This claim does not find empirical support. In particular, the part of
arcuate/superior longitudinal fasciculus—the dorsal tract that connects Broca's area that responds to the presence of structure in language (e.g.,
posterior temporal and inferior frontal language areas—is critical for showing stronger responses to structured linguistic stimuli, like sen-
syntactic processing (e.g., Friederici, 2009; Brauer, Anwander, & tences, than to lists of unconnected words) is highly selective for lan-
Friederici, 2011; Papoutsi, Stamatakis, Griffiths, Marslen-Wilson, & guage relative to non-linguistic tasks, including ones that involve
Tyler, 2011; Wilson et al., 2011). However, the selectivity of this tract hierarchical structure and/or recursion, like arithmetic and music (e.g.,
for syntactic/combinatorial processing is unclear, as it has also been Fedorenko et al., 2011; Monti, Parsons, & Osherson, 2012; Amalric &
implicated in non-syntactic computations, including, most commonly, Dehaene, 2018; see Fedorenko et al., 2016 (PNAS), for a review). These
articulation (e.g., Duffau et al., 2003; Hickok & Poeppel, 2007; results make sense given the strong links between linguistic structure
Rauschecker & Scott, 2009), but also aspects of semantic processing and meaning discussed in this manuscript: in other words, given what
(e.g., Glasser & Rilling, 2008). Thus, we would argue that, at present, no we now know about linguistic syntax, the idea that it would be sup-
unequivocal evidence of syntax selectivity exists for inter-regional ported by a mechanism that is not sensitive to the nature of the re-
connections either. presentation does not seem tenable.
Second, our research question focused on the “core” fronto-tem- One possible explanation for some of the findings that have been
poral language network, consisting of regions on the lateral surfaces of used as evidence for a domain-general structure processor is that those
left frontal and left temporal cortex (e.g., Fedorenko & Thompson- manipulations activated a domain-general component of Broca's area.
Schill, 2014). Some areas outside of this network's boundaries have been This component belongs to a distinct network—the domain-general
implicated in syntactic processing, including, for example, parts of the multiple demand (MD) network implicated in executive control and
basal ganglia (e.g., Ullman, 2001, 2004; cf. Grossman, Lee, Morris, goal-directed behaviors and robustly sensitive to effort (e.g., Duncan,
Stern, & Hurtig, 2002; Longworth, Keenan, Barker, Marslen-Wilson, & 2010, 2013; Fedorenko et al., 2013). Manipulations of hierarchical
Tyler, 2005) or the cerebellum (see Mariën et al., 2014 for a review). complexity have often been confounded with difficulty, such that the
However, we would argue that, similar to the cortical language regions, more structurally complex conditions required greater cognitive effort.
selectivity for syntactic processing for any brain region outside of the As a result, they would be likely to elicit responses in the MD network,
core language network has not been compellingly established. including its inferior frontal component residing in Broca's area (see
Of relevance to this point is a claim that a region residing in or Fedorenko & Blank, 2020, for additional discussion). Although it is
around Broca's area supports abstract hierarchical structure processing possible that outside of the domain of language—which appears to rely

16
E. Fedorenko, et al. Cognition 203 (2020) 104348

on domain-specific processing mechanisms (Fedorenko et al., temporal cortex: e.g., Borovsky et al., 2007; Halai et al., 2017; Ding
2011)—this component of the MD network, or the MD network as a et al., in press; inferior frontal cortex: e.g., Schnur et al., 2009; Corina
whole, is important for structured behavior or processing hierarchically et al., 2010; Kojima et al., 2013; Python, Glize, & Laganaro, 2018; both:
structured input, we should keep in mind that this network is also Sanai, Mirzadeh, & Berger, 2008), and responses to lexical selection
sensitive to manipulations that don't involve structured/hierarchical demands have been reported across the language network in a recent
representations (e.g., Crittenden & Duncan, 2012). The latter findings ECoG study (Riès et al., 2017).
argue against the idea of complex syntactic operations being the core Some of the challenges in interpreting findings from language pro-
computation of the MD network. duction research are similar to those discussed in the Introduction in
Third, we have here focused on language comprehension. Could the the context of language comprehension research, including common
architecture of language processing be different for language produc- use of a single paradigm in any given study, failure to report region by
tion? Although we plausibly access the same knowledge representations condition interactions when arguing for between-region differences,
to interpret (comprehend) and generate (produce) linguistic utter- lack of all the necessary control conditions needed to establish syntax
ances—in line with substantial overlap that has been observed between selectivity, and challenges in comparing results across studies (in-
comprehension and production in fMRI (e.g., Menenti, Gierhan, cluding the use of broad anatomical areas as units of analysis in spite of
Segaert, & Hagoort, 2011; Silbert, Honey, Simony, Poeppel, & Hasson, their known structural and functional heterogeneity). Language pro-
2014)—the computational demands of language production differ from duction research also faces some additional challenges: the use of un-
those of language comprehension. In particular, the goal of compre- constrained naturalistic production tasks makes it difficult to compare
hension is to infer the intended meaning from the linguistic signal, and data across individuals, and constrained artificial tasks (like picture
abundant evidence now suggests that the representations we extract naming/description) introduce extraneous demands beyond the critical
and maintain during comprehension are probabilistic and often noisy targeted processes of accessing the relevant stored linguistic re-
(e.g., Ferreira et al., 2002; Gibson et al., 2013; Levy, 2008). In contrast, presentations and phrase/sentence construction. Furthermore, in ana-
in production, the target meaning is (typically) clear and precise, and lyzing production errors—critical for many common approaches, in-
the goal is to express that particular meaning. To do so, we have to utter cluding voxel-lesion symptom mapping—the intended utterance can be
a precise sequence of words where each word takes a particular hard to infer unambiguously, which is critical for interpreting an error
morpho-syntactic form, and the words appear in a particular order. This as reflecting a lexical retrieval failure (e.g., resulting in a word sub-
pressure for linearization of words, morphemes, and sounds might lead stitution or circumlocutions) vs. difficulties in syntactic planning/en-
to a clearer temporal, and perhaps spatial, segregation among the dif- coding (e.g., resulting in an incorrect inflection or a word-order error).
ferent stages of the production process compared to comprehension In conclusion, given the reports of apparently selective deficits in
(e.g., Garrett, 2000; Hagoort & Indefrey, 2014; cf. Vigliocco & some aspects of morpho-syntactic production in the patient literature
Hartsuiker, 2002), and/or require additional, production-selective, (e.g., Bastiaanse, 1995; Miceli et al., 1989; Miceli & Caramazza, 1988;
mechanisms implemented in brain regions that do not support com- Thompson, Fix, & Gitelman, 2002), it remains possible that some as-
prehension. Indeed, some dissociations have been reported among pects of language production are implemented in focal and functionally
different aspects of language production in both stroke aphasia (e.g., selective brain regions that do not support lexical access/word-level
Borovsky, Saygin, Bates, & Dronkers, 2007; Casilio, Rising, Beeson, production. However, such selective regions should also be detectable
Bunton, & Wilson, 2019; Ding, Martin, Hamilton, & Schnur, in press; with brain imaging or neurophysiological approaches, and to the best of
Halai, Woollams, & Ralph, 2017; Matchin et al., 2020; Miceli, Silveri, our knowledge, no single brain region has been compellingly estab-
Romani, & Caramazza, 1989; Mirman et al., 2015; Mirman, Kraft, lished as selective for some component(s) of phrase/sentence-level
Harvey, Brecher, & Schwartz, 2019) and primary progressive aphasia production over single-word retrieval, across individuals, paradigms,
(e.g., Mesulam et al., 2014; Wilson et al., 2010, 2011). And some pa- and labs.
tients have been reported to exhibit syntactic production deficits in the Finally, in the current study, we investigated the relationship be-
absence of syntactic comprehension deficits (e.g., Kolk, van Grunsven, tween syntactic and lexico-semantic processing in a single Germanic
& Keyser, 1985; Miceli, Mazzucchi, Menn, & Goodglass, 1983; language: English. Some of the theoretical linguistic work, experimental
Nespoulous et al., 1988, although reverse comprehension-selective psycholinguistic work, neuropsychological patient work, and compu-
syntactic deficits have also been reported: e.g., Caramazza, Basili, tational modeling work have spanned multiple languages (e.g.,
Koller, & Berndt, 1981; Caplan, 1985; Bates et al., 1988). The key Norcliffe, Harris, & Jaeger, 2015). However, much/most cognitive
question relevant to the current investigation is whether any brain re- neuroscience research has been conducted on English and a handful of
gions selectively support some aspect(s) of syntactic processing. We other languages/families (e.g., see Bornkessel-Schlesewsky &
would argue that, as in the comprehension literature, the separation Schlesewsky, 2016 for discussion). Thus, the conclusions drawn here
between lexical access (which could have distinct semantic vs. syntactic remain to be generalized to typologically diverse languages.
contributions; e.g., Gordon & Dell, 2003; Barde, Schwartz, & Boronat,
2006) and syntactic/combinatorial processing remains controversial. 4.4. Other findings that have been interpreted as evidence for syntax
Similar to comprehension, regions that are most commonly im- selectivity
plicated in syntactic/combinatorial processes in language production
include left inferior frontal areas and left posterior temporal areas. For Three other lines of research—on phenomena that are, or have
example, in a recent investigation of acute stroke patients, Ding et al. been, taken as strong evidence for syntax selectivity—deserve discus-
(in press) found that damage to inferior frontal areas is associated with sion. First, the early ERP literature on language processing appeared to
the production of syntactically ill-formed sentences, including word have provided evidence of distinct components associated with lexico-
omissions and agreement errors, and damage to posterior temporal/ semantic processing (N400; Kutas & Hillyard, 1980) vs. with syntactic
temporo-parietal areas is associated with the production of shorter and processing (P600; Osterhout & Holcomb, 1992; Hagoort et al., 1993).
less structurally complex sentences. Intra-operative stimulation studies However, the interpretation of the P600 as an index of syntactic pro-
have also implicated both inferior frontal (Chang, Kurteff, & Wilson, cessing has been challenged from the earliest days following its dis-
2018) and posterior superior temporal (Lee et al., 2018, Fedorenko, covery (e.g., Coulson, King, & Kutas, 1998), and the current dominant
Williams, & Ferreira, 2018) sites in syntactic encoding during produc- interpretation of this component is as a domain-general error detection
tion. However, importantly, across and sometimes within studies, both or correction signal (e.g., Kolk & Chwilla, 2007; Ryskin et al., 2020;
inferior frontal and posterior temporal damage/stimulation have also Sassenhagen, Schlesewsky, & Bornkessel-Schlesewsky, 2014; van de
been shown to affect lexical selection/word-level production (posterior Meerendonk, Kolk, Vissers, & Chwilla, 2010; Vissers, Chwilla, & Kolk,

17
E. Fedorenko, et al. Cognition 203 (2020) 104348

2007). Some other, earlier, ERP components (e.g., eLAN; Friederici, 2016). Further, based on evidence from MEG, parts of the left anterior
2002) have been argued to index syntactic processes. However, the temporal lobe have been implicated in semantic composition, above
robustness and nature of these components have been questioned and beyond the processing of single words (e.g., Bemis & Pylkkanen,
(Steinhauer & Drury, 2012), and the interpretation that seems to cap- 2011; see Pylkkänen, 2019 for a review; cf. Kochari, Lewis, Schoffelen,
ture the empirical findings best has to do with violations of word-form & Schriefers, 2020). The brain imaging evidence therefore suggests that
expectations rather than syntactic parsing per se (e.g., Dikker, this region is engaged in some syntax-/combinatorics-relevant pro-
Rabagliati, Farmer, & Pylkkänen, 2010; Dikker, Rabagliati, & cesses, in addition to lexico-semantic processing. But because the
Pylkkänen, 2009; Rosenfelt et al., 2009; Rosenfelt, Kluender, & Kutas, temporal pole and the anterior ventral temporal cortex are challenging
2011). to study with fMRI given the signal dropout due to proximity to air-
Second, syntactic priming—re-use of a syntactic frame based on filled sinuses (e.g., Devlin et al., 2000), it is possible that we are missing
recent linguistic experiences (Bock, 1986; see Pickering & Ferreira, some areas—anterior to our LAntTemp fROI and/or on the ventral
2008; Branigan & Pickering, 2017 for reviews)—has often been cited as surface of the temporal lobe—that are truly selective for lexico-se-
evidence of abstract syntactic representations independent of meaning mantic/conceptual over syntactic/combinatorial processing. The pa-
(e.g., Bock & Loebell, 1990), including in relatively recent cognitive tient evidence mentioned above implies the existence of such areas,
neuroscience papers (Pallier et al., 2011). However, a large body of although the evidence is not unequivocal (e.g., Bi et al., 2011). More
work has now established that the effect is strongly modulated by work is therefore needed to functionally characterize left anterior
lexical overlap (e.g., Mahowald et al., 2016; Scheepers, Raffray, & temporal areas, including potentially distinct subregions therein, and
Myachykov, 2017; Ziegler, Bencini, Goldberg, & Snedeker, 2019) and their relationship with the core fronto-temporal language network.
driven by the meaning-related aspects of the utterance (e.g., Cai, Going beyond syntax and semantics, lower-level speech perception
Pickering, & Branigan, 2012; Hare & Goldberg, 1999; Ziegler et al., and reading processes as well as speech production (articulation) re-
2018; Ziegler & Snedeker, 2018). cruit areas that are robustly distinct from the high-level areas that we
And third, a class of linguistic phenomena known as “syntactic is- focused on here. In particular, speech perception recruits parts of the
lands” (Ross, 1967) have been argued to be due to abstract properties of auditory cortex in the superior temporal gyrus and sulcus (e.g., Scott,
syntactic structures unrelated to meaning. In particular, some structures Blank, Rosen, & Wise, 2000, Mesgarani et al., 2014; Overath,
are disfavored when a phrase is “extracted” from its “canonical” McDermott, Zarate, & Poeppel, 2015), and these areas are highly se-
structural location in a sentence (e.g., Chomsky, 1973; Schütze, lective for speech over many other types of auditory stimuli (Norman-
Sprouse, & Caponigro, 2015). However, this interpretation has been Haignere, Kanwisher, & McDermott, 2015). Reading recruits a small
challenged: in particular, some researchers (e.g., Erteschik-Shir, 1973; area on the ventral surface of the temporal lobe (see McCandliss,
Goldberg, 2013; Kuno, 1987) have argued that semantic and discourse Cohen, & Dehaene, 2003, for a review), and this “visual word-form
factors can explain these phenomena (see Abeillé, Hemforth, Winckel, area” is highly selective for letters in a familiar script over a broad
& Gibson, 2020, for empirical support). range of other visual stimuli (Baker et al., 2007; Hamamé et al., 2013).
And articulation draws on a set of areas, including portions of the
4.5. Beyond syntax and semantics: dissociations of other linguistic processes precentral gyrus, supplementary motor area, inferior frontal cortex,
superior temporal cortex, and cerebellum (e.g., Basilakos, Smith,
Language processing encompasses a broad array of computations in Fillmore, Fridriksson, & Fedorenko, 2018; Bohland & Guenther, 2006;
both comprehension and production, and some aspects of language are Eickhoff, Heim, Zilles, & Amunts, 2009; Wise, Greene, Büchel, & Scott,
robustly dissociable and supported by distinct sets of brain regions. 1999).
Here, we have argued that during language comprehension the me- On the other end of linguistic processes, discourse-level processing
chanisms that process the structure of phrases and sentences are also draws on areas distinct from those that support word and sentence-level
deeply sensitive to the meanings of individual words. Is the reverse also comprehension (e.g., Blank & Fedorenko, in press; Ferstl & von Cramon,
true? Are there any brain areas that process individual word meanings 2001; Jacoby & Fedorenko, 2018; Lerner, Honey, Silbert, & Hasson,
but are not sensitive to syntactic/combinatorial processing? 2011), and aspects of non-literal language have been argued to draw on
One area that deserves a mention lies in the left temporal pole, brain regions in the right hemisphere (e.g., Joanette, Goulet, &
extending onto the lateral and ventral surface of the temporal lobe. Hannequin, 1990) and on the system that supports social cognition
According to one hypothesis, this area supports lexical retrieval (e.g., (e.g., Hagoort, 2019; Kline, Gallée, Balewski, & Fedorenko, 2018).
Damasio, Tranel, Grabowski, Adolphs, & Damasio, 2004; Drane et al., Thus, many aspects of language are robustly dissociable, in line with
2008; Grabowski et al., 2001; Mesulam et al., 2013, 2015; Tranel, 2006, distinct patterns of deficits reported in the aphasia literature (e.g.,
2009). According to another hypothesis, motivated chiefly by in- Goodglass, 1993). However, syntactic and lexico-semantic processing
vestigations of semantic dementia (Gorno-Tempini et al., 2011), this do not appear to be separable during language comprehension. Some
region has been linked to general object knowlege (e.g., Lambon-Ralph, brain areas not easily accessible to fMRI may be selective for lexico-
McClelland, Patterson, Galton, & Hodges, 2001; Rogers et al., 2004, semantic processing, as discussed above, but no area within the lan-
2006; Patterson, Nestor, & Rogers, 2007; cf. Bi et al., 2011). Critically, guage network appears to be selective for syntactic processing based on
syntactic abilities in patients with damage to this area appear to be both brain imaging studies and patient investigations.
relatively preserved (e.g., Mesulam et al., 2013, 2015; see Hardy,
Segaert, & Wheeldon, 2020 for potentially related evidence of impaired 5. Concluding remarks
lexical access in the presence of intact syntactic planning in healthy
aging). This area overlaps with our LAntTemp fROI, but extends beyond To conclude, across three fMRI experiments, we found robust re-
it. In our experiments, the LAntTemp language fROI showed stronger sponses to both lexico-semantic and syntactic processing throughout
responses during lexico-semantic than syntactic processing in two of the the language network, with generally stronger responses to lexico-se-
three experiments (Fig. 5). However, this region still responds reliably mantic processing, and no regions, or even sets of non-contiguous
to syntactic processing in at least some manipulations: for example, it voxels within those regions, that respond reliably more strongly to
responds more strongly to a) structured but meaningless Jabberwocky syntactic processing than lexico-semantic processing. These results
sentences compared to lists of unconnected nonwords (Fedorenko et al., constrain the space of possible neural architectures of language. In
2010; see Mollica et al., in prep, for a replication), and b) structurally particular, they rule out architectures that postulate a distinct region
more complex sentences with object-extracted relative clauses com- (or set of regions) that selectively supports syntactic/combinatorial
pared to those with subject-extracted relative clauses (Blank et al., processing (i.e., architectures shown in Fig. 1a–d). These findings,

18
E. Fedorenko, et al. Cognition 203 (2020) 104348

illuminating how minds are instantiated in brains, are mirrored by Arnon, I., & Snider, N. (2010). More than words: Frequency effects for multi-word
studies of how minds are implemented in machines, where modern-day phrases. Journal of Memory and Language, 62(1), 67–82.
Axelrod, V., Bar, M., Rees, G., & Yovel, G. (2015). Neural correlates of subliminal lan-
connectionist networks achieve remarkable performance on a wide guage processing. Cerebral Cortex, 25(8), 2160–2169.
variety of language tasks (e.g., Bahdanau, Chorowski, Serdyuk, Brakel, Baggio, G., & Hagoort, P. (2011). The balance between memory and unification in se-
& Bengio, 2016; Mikolov, Karafiát, Burget, Černocký, & Khudanpur, mantics: A dynamic account of the N400. Language and Cognitive Processes, 26(9),
1338–1367.
2010; Sutskever, Vinyals, & Le, 2014), including those that involve Bahdanau, D., Chorowski, J., Serdyuk, D., Brakel, P., & Bengio, Y. (2016). End-to-end
complex syntactic phenomena (e.g., Futrell et al., 2019; Futrell, Wilcox, attention-based large vocabulary speech recognition. Acoustics, speech and signal
Morita, & Levy, 2018; Gulordava, Bojanowski, Grave, Linzen, & Baroni, processing (ICASSP), 2016 IEEE International Conference (pp. 4945–4949). .
Baker, C. I., Liu, J., Wald, L. L., Kwong, K. K., Benner, T., & Kanwisher, N. (2007). Visual
2018; Linzen, Dupoux, & Goldberg, 2016; Prasad, van Schijndel, & word processing and experiential origins of functional selectivity in human extra-
Linzen, 2019; Wilcox, Levy, & Futrell, 2019a, 2019b), apparently striate cortex. Proceedings of the National Academy of Sciences, 104(21), 9087–9092.
without a clearly separable syntax-selective mechanism. Some of these Barde, L. H., Schwartz, M. F., & Boronat, C. B. (2006). Semantic weight and verb retrieval
in aphasia. Brain and Language, 97(3), 266–278.
state-of-the-art neural network models also robustly predict human
Barlow, M., & Kemmer, S. (2000). Usage based models of language.
neural data during language processing (Schrimpf et al., in prep). Basilakos, A., Smith, K. G., Fillmore, P., Fridriksson, J., & Fedorenko, E. (2018).
Taking all the available data into consideration, it therefore seems that Functional characterization of the human speech articulation network. Cerebral
a cognitive architecture whereby syntactic processing is not separable Cortex, 28(5), 1816–1830.
Bastiaanse, R. (1995). Broca’ s Aphasia: A syntactic and/or a morphological disorder? A
from the processing of individual word meanings is most likely. case study. Brain and Language, 48(1), 1–32.
Bates, E., Bretherton, I., & Snyder, L. (1988). From first words to grammar. New York:
CRediT authorship contribution statement Cambridge University Press.
Bates, E., Dale, P. S., & Thal, D. (1995). Individual differences and their implications for
theories of language development. In P. Fletcher, & B. MacWhinnery (Eds.). The
Evelina Fedorenko:Conceptualization, Methodology, handbook of child language (pp. 96–151). Basil Blackwell.
Investigation, Formal analysis, Writing - original draft.Idan Bates, E., & Goodman, J. C. (1997). On the inseparability of grammar and the lexicon:
Evidence from acquisition, aphasia and real time processing. Language and Cognitive
Blank:Investigation, Formal analysis, Methodology, Writing - re- Processes, 12, 507–586.
view & editing.Matthew Siegelman:Investigation, Formal analysis, Bautista, A., & Wilson, S. M. (2016). Neural responses to grammatically and lexically
Methodology, Writing - review & editing.Zachary degraded speech. Language Cognition and Neuroscience, 31(4), 567–574.
Bedny, M., Pascual-Leone, A., Dodell-Feder, D., Fedorenko, E., & Saxe, R. (2011).
Mineroff:Investigation, Formal analysis, Writing - review & editing. Language processing in the occipital cortex of congenitally blind adults. Proceedings of
the National Academy of Sciences, 108(11), 4429–4434.
Acknowledgements Bekinschtein, T. A., Davis, M. H., Rodd, J. M., & Owen, A. M. (2011). Why clowns taste
funny: The relationship between humor and semantic ambiguity. Journal of
Neuroscience, 31(26), 9665–9671.
We would like to acknowledge the Athinoula A. Martinos Imaging Bemis, D. K., & Pylkkanen, L. (2011). Simple composition: A magnetoencephalography
Center at the McGovern Institute for Brain Research at MIT, and its investigation into the comprehension of minimal linguistic phrases. Journal of
support team (Steve Shannon and Atsushi Takahashi). We thank Josef Neuroscience, 31(8), 2801–2814.
Benjamini, Y., & Yekutieli, D. (2001). The control of the false discovery rate in multiple
Affourtit for helping put together the OSF page and organize the raw testing under dependency. Annals of Statistics, 29(4), 1165–1188.
data, former and current EvLab members (especially Zuzanna Balewski Ben-Shachar, M., Hendler, T., Kahn, I., Ben-Bashat, D., & Grodzinsky, Y. (2003). The
and Brianna Pritchett) for their help with data collection, Gina neural reality of syntactic transformations: Evidence from functional magnetic re-
sonance imaging. Psychological Science, 14(5), 433–440.
Kuperberg for providing the materials used in adapted form in Berwick, R. C., Friederici, A. D., Chomsky, N., & Bolhuis, J. J. (2013). Evolution, brain,
Experiment 1, Michael Behr for creating the script for Experiment 1, and the nature of language. Trends in Cognitive Sciences, 17(2), 89–98.
Zuzanna Balewski for help with creating the materials and script for Bi, Y., Wei, T., Wu, C., Han, Z., Jiang, T., & Caramazza, A. (2011). The role of the left
anterior temporal lobe in language processing revisited: Evidence from an individual
Experiment 2, Nancy Kanwisher for discussions of the experimental with ATL resection. Cortex, 47(5), 575–587.
design for all three experiments, Gary Dell and Gary Oppenheim for a Bilenko, N. Y., Grindrod, C. M., Myers, E. B., & Blumstein, S. E. (2008). Neural correlates
helpful email exchange about language production, and Inbal Arnon, of semantic competition during processing of ambiguous words. Journal of Cognitive
Neuroscience, 21(5), 960–975.
Yonatan Belinkov, Ted Gibson, Adele Goldberg, Maryellen MacDonald,
Binder, J. R., Desai, R. H., Graves, W. W., & Conant, L. L. (2009). Where is the semantic
Stephen Wilson, and Jayden Ziegler for comments on the manuscript. system? A critical review and meta-analysis of 120 functional neuroimaging studies.
We also thank the audience at the 2017 CUNY Sentence Processing Cerebral Cortex, 19(12), 2767–2796.
Binder, J. R., Desai, R. H., Graves, W. W., & Conant, L. L. (2009). Where is the semantic
conference (Cambridge, MA) for feedback, as well as Ina Bornkessel-
system? A critical review and meta-analysis of 120 functional neuroimaging studies.
Schlesewsky , three anonymous reviewers, and the editor (Randi Cerebral Cortex, 19(12), 2767–2796.
Martin) whose comments helped to greatly improve the manuscript. EF Blank, I., Balewski, Z., Mahowald, K., & Fedorenko, E. (2016). Syntactic processing is
was supported by NIH awards R00-HD057522, R01-DC016607, R01- distributed across the language system. NeuroImage, 127, 307–323.
Blank, I., Kanwisher, N., & Fedorenko, E. (2014). A functional dissociation between
DC016950, by a grant from the Simons Foundation to the Simons language and multiple-demand systems revealed in patterns of BOLD signal fluc-
Center for the Social Brain at MIT, and by funds from the Brain and tuations. Journal of Neurophysiology, 112(5), 1105–1118.
Cognitive Sciences Department and the McGovern Institute for Brain Blank, I.A. & Fedorenko, E. (in press). No evidence for differences among language re-
gions in their temporal receptive windows. NeuroImage.
Research at MIT. Blank, I. A., Kiran, S., & Fedorenko, E. (2017). Can neuroimaging help aphasia re-
searchers? Addressing generalizability, variability, and interpretability. Cognitive
Declaration of competing interest Neuropsychology, 34(6), 377–393.
Bock, J. K. (1986). Meaning, sound, and syntax: Lexical priming in sentence production.
Journal of Experimental Psychology: Learning, Memory, and Cognition, 12(4), 575–586.
The authors declare no competing financial interests. Bock, K., & Loebell, H. (1990). Framing sentences. Cognition, 35(1), 1–39.
Bod, R. (1998). Beyond grammar: An experience-based theory of language. Stanford, CA:
CSLI Publications.
References
Bod, R. (2006). Exemplar-based syntax: How to get productivity from examples. The
Linguistic Review, 23(3), 291–320.
Abeillé, A., Hemforth, B., Winckel, E., & Gibson, E. (2020). Extraction from subjects: Bohland, J. W., & Guenther, F. H. (2006). An fMRI investigation of syllable sequence
Differences in acceptability depend on the discourse function of the construction. production. Neuroimage, 32(2), 821–841.
Cognition.. Bookheimer, S. (2002). Functional MRI of language: New approaches to understanding
Aguirre, G. K., & Farah, M. J. (1998). Human visual object recognition: What have we the cortical organization of semantic processing. Annual Review of Neuroscience, 25,
learned from neuroimaging? Psychobiology, 26(4), 322–332. 151–188.
Amalric, & Dehaene, S. (2018). A distinct cortical network for mathematical knowledge in Bornkessel-Schlesewsky, I., & Schlesewsky, M. (2016). The importance of linguistic ty-
the human brain. Neuroimage. pology for the neurobiology of language. Linguistic Typology, 20(3), 615–621.
Ambridge, B. (2018). Against stored abstractions: A radical exemplar model of language Bornkessel-Schlesewsky, I., Schlesewsky, M., Small, S. L., & Rauschecker, J. P. (2015).
acquisition. First Language. https://journals.sagepub.com/doi/pdf/10.1177/ Neurobiological roots of language in primate audition: Common computational
0142723719869731 (0142723719869731). properties. Trends in Cognitive Sciences, 19(3), 142–150.

19
E. Fedorenko, et al. Cognition 203 (2020) 104348

Bornkessel-Schlesewsky, I., & Schlesewsky, M. (2009). The role of prominence informa- Damasio, H., Tranel, D., Grabowski, T., Adolphs, R., & Damasio, A. (2004). Neural sys-
tion in the real-time comprehension of transitive constructions: A cross-linguistic tems behind word and concept retrieval. Cognition, 92(1–2), 179–229.
approach. Linguistics and Language Compass, 3, 19–58. https://doi.org/10.1111/j. Dapretto, M., & Bookheimer, S. Y. (1999). Form and content: Dissociating syntax and
1749-818X.2008.00099.x. semantics in sentence comprehension. Neuron, 24(2), 427–432.
Borovsky, A., Saygin, A. P., Bates, E., & Dronkers, N. (2007). Lesion correlates of con- Davis, M. H., Coleman, M. R., Absalom, A. R., Rodd, J. M., Johnsrude, I. S., Matta, B. F., ...
versational speech production deficits. Neuropsychologia, 45(11), 2525–2533. Menon, D. K. (2007). Dissociating speech perception and comprehension at reduced
Botvinik-Nezer, R., Holzmeister, F., Camerer, C. F., Dreber, A., Huber, J., Johannesson, levels of awareness. Proceedings of the National Academy of Sciences of the United States
M., ... Avesani, P. (2019). Variability in the analysis of a single neuroimaging dataset of America, 104(41), 16032–16037.
by many teams. bioRxiv. https://www.nature.com/articles/s41586-020-2314-. Devauchelle, A. D., Oppenheim, C., Rizzi, L., Dehaene, S., & Pallier, C. (2009). Sentence
Braga, R. M., DiNicola, L. M., & Buckner, R. L. (2019). Situating the left-lateralized lan- syntax and content in the human temporal lobe: an fMRI adaptation study in auditory
guage network in the broader organization of multiple specialized large-scale dis- and visual modalities. Journal of Cognitive Neuroscience, 21(5), 1000–1012.
tributed networks. bioRxiv. Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of deep
Branco, P., Seixas, D., & Castro, S. L. (2020). Mapping language with resting-state func- bidirectional transformers for language understanding. ArXiv. 1810.04805 [Cs]
tional magnetic resonance imaging: A study on the functional profile of the language http://arxiv.org/abs/1810.04805.
network. Human Brain Mapping, 41(2), 545–560. Devlin, J. T., Russell, R. P., Davis, M. H., Price, C. J., Wilson, J., Moss, H. E., ... Tyler, L. K.
Branigan, H. P., & Pickering, M. J. (2017). An experimental approach to linguistic re- (2000). Susceptibility-induced loss of signal: Comparing PET and fMRI on a semantic
presentation. Behavioral and Brain Sciences, 40. task. NeuroImage, 11(6), 589–600.
Brauer, J., Anwander, A., & Friederici, A. D. (2011). Neuroanatomical prerequisites for Diachek, E., Siegelman, M., Blank, I., Affourtit, J., & Fedorenko, E. (2020). The domain-
language functions in the maturing brain. Cerebral Cortex, 21(2), 459–466. general multiple demand (MD) network does not support core aspects of language
Brett, M., Johnsrude, I. S., & Owen, A. M. (2002). The problem of functional localization comprehension: A large-scale fMRI investigation. Journal of Neuroscience. 40(23),
in the human brain. Nature Reviews. Neuroscience, 3(3), 243–249. 4536–4550. https://doi.org/10.1523/JNEUROSCI.2036-19.2020.
Bybee, J. (2006). From usage to grammar: The mind’s response to repetition. Language, Dick, F., Bates, E., Wulfeck, B., Utman, J. A., Dronkers, N., & Gernsbacher, M. A. (2001).
82, 711–733. Language deficits, localization, and grammar: Evidence for a distributive model of
Bybee, J. (2010). Language, usage and cognition. Vol. 98. Cambridge University Press. language breakdown in aphasic patients and neurologically intact individuals.
Bybee, J. L. (1985). Morphology: A study of the relation between meaning and form. Vol. 9. Psychological Review, 108(4), 759–788. https://doi.org/10.1037//0033-295x.108.4.
Amsterdam: John Benjamins Publishing Company. 759.
Bybee, J. L. (1998). A functionalist approach to grammar and its evolution. Evolution of Dikker, S., Rabagliati, H., Farmer, T. A., & Pylkkänen, L. (2010). Early occipital sensitivity
Communication, 2(2), 249–278. to syntactic category is based on form typicality. Psychological Science, 21(5),
Cai, Z. G. G., Pickering, M. J., & Branigan, H. P. (2012). Mapping concepts to syntax: 629–634.
Evidence from structural priming in Mandarin Chinese. Journal of Memory and Dikker, S., Rabagliati, H., & Pylkkänen, L. (2009). Sensitivity to syntax in visual cortex.
Language, 66(4), 833–849. Cognition, 110(3), 293–321.
Caplan, D. (1985). Syntactic and semantic structures in agrammatism. Agrammatism (pp. Ding, J., Martin, R., Hamilton, A. C., & Schnur, T. T. (in press). Dissociation between
125–152). Academic Press. frontal and temporal-parietal contributions to connected speech in acute stroke.
Caplan, D., Hildebrandt, N., & Makris, N. (1996). Location of lesions in stroke patients Brain.
with deficits in syntactic processing in sentence comprehension. Brain, 119(3), Dixon, J. A., & Marchman, V. A. (2007). Grammar and the lexicon: Developmental or-
933–949. dering in language acquisition. Child Development, 78(1), 190–212.
Caramazza, A., Basili, A. G., Koller, J. J., & Berndt, R. S. (1981). An investigation of Drane, D. L., Ojemann, G. A., Aylward, E., Ojemann, J. G., Johnson, L. C., Silbergeld, D.
repetition and language processing in a case of conduction aphasia. Brain and L., ... Tranel, D. (2008). Category-specific naming and recognition deficits in tem-
Language, 14(2), 235–271. poral lobe epilepsy surgical patients. Neuropsychologia, 46(5), 1242–1255.
Casilio, M., Rising, K., Beeson, P. M., Bunton, K., & Wilson, S. M. (2019). Auditory-per- Duffau, H., Capelle, L., Denvil, D., Gatignol, P., Sichez, N., Lopes, M., ... Van Effenterre, R.
ceptual rating of connected speech in aphasia. American Journal of Speech-Language (2003). The role of dominant premotor cortex in language: A study using in-
Pathology, 28(2), 550–568. traoperative functional mapping in awake patients. Neuroimage, 20(4), 1903–1914.
Chang, E. F., Kurteff, G., & Wilson, S. M. (2018). Selective interference with syntactic Duffau, H., Moritz-Gasser, S., & Mandonnet, E. (2014). A re-examination of neural basis of
encoding during sentence production by direct electrocortical stimulation of the in- language processing: Proposal of a dynamic hodotopical model from data provided
ferior frontal gyrus. Journal of Cognitive Neuroscience, 30(3), 411–420. by brain stimulation mapping during picture naming. Brain and Language, 131, 1–10.
Chang, F., Dell, G. S., & Bock, K. (2006). Becoming syntactic. Psychological Review, 113(2), https://doi.org/10.1016/j.bandl.2013.05.011.
234. Duncan, J. (2010). The multiple-demand (MD) system of the primate brain: Mental
Chee, M. W., Caplan, D., Soon, C. S., Sriram, N., Tan, E. W., Thiel, T., & Weekes, B. (1999). programs for intelligent behaviour. Trends in Cognitive Sciences, 14(4), 172–179.
Processing of visually presented sentences in Mandarin and English studied with Duncan, J. (2013). The structure of cognition: Attentional episodes in mind and brain.
fMRI. Neuron, 23(1), 127–137. Neuron, 80(1), 35–50.
Chomsky, N. (1965). Aspects of the theory of syntax. Cambridge, MA: MIT Press. Eickhoff, S. B., Heim, S., Zilles, K., & Amunts, K. (2009). A systems perspective on the
Chomsky, N. (1987). ). Syntactic Structures. The Hague: Mouton. effective connectivity of overt speech production. Philosophical Transactions of the
Chomsky, N. (1973). Conditions on transformations. A festschrift for Morris Halle. Royal Society of London A: Mathematical, Physical and Engineering Sciences, 367(1896),
Chomsky, N. (1995). Language and nature. Mind, 104(413), 1–61 (JSTOR). 2399–2421.
Chomsky, N., & Dinozzi, R. (1972). Language and mind. Harcourt Brace Jovanovich. Embick, D., Marantz, A., Miyashita, Y., O’Neil, W., & Sakai, K. L. (2000). A syntactic
Christiansen, M. H., & Arnon, I. (2017). More than words: The role of multiword se- specialization for Broca’s area. Proceedings of the National Academy of Sciences, 97(11),
quences in language learning and use. Topics in Cognitive Science, 9(3), 542–551. 6150–6154.
Clifton, C., Frazier, L., & Connine, C. (1984). Lexical expectations in sentence compre- Engel, A. K., Moll, C. K., Fried, I., & Ojemann, G. A. (2005). Invasive recordings from the
hension. Journal of Memory and Language, 23(6), 696. human brain: Clinical insights and beyond. Nature Reviews Neuroscience, 6(1), 35–47.
Cooke, A., Grossman, M., DeVita, C., Gonzalez-Atavales, J., Moore, P., Chen, W., ... Detre, Erteschik-Shir, N. (1973). On the nature of island constraints (Doctoral dissertation)
J. (2006). Large-scale neural network for sentence processing. Brain and Language, Massachusetts Institute of Technology.
96(1), 14–36. Evert, S. (2008). Corpora and collocations. Corpus Linguistics. An International Handbook,
Corbetta, M., & Shulman, G. L. (2002). Control of goal-directed and stimulus-driven at- 2, 1212–1248.
tention in the brain. Nature reviews neuroscience, 3(3), 201–215. Fadiga, L., Craighero, L., & D’Ausilio, A. (2009). Broca’s area in language, action, and
Corina, D. P., Loudermilk, B. C., Detwiler, L., Martin, R. F., Brinkley, J. F., & Ojemann, G. music. Annals of the New York Academy of Sciences, 1169(1), 448–458.
(2010). Analysis of naming errors during cortical stimulation mapping: Implications Fedor, A., Varga, M., & Szathmáry, E. (2012). Semantics boosts syntax in artificial
for models of language representation. Brain and Language, 115(2), 101–112. grammar learning tasks with recursion. Journal of Experimental Psychology: Learning,
Costafreda, S. G., Fu, C. H., Lee, L., Everitt, B., Brammer, M. J., & David, A. S. (2006). A Memory, and Cognition, 38(3), 776.
systematic review and quantitative appraisal of fMRI studies of verbal fluency: Role Fedorenko, & Blank (2020). Broca’s area is not a natural kind. Trends in Cognitive Sciences,
of the left inferior frontal gyrus. Human Brain Mapping, 27(10), 799–810. 24(4), 270–284.
Coulson, S., King, J. W., & Kutas, M. (1998). Expect the unexpected: Event-related brain Fedorenko, E. (2014). The role of domain-general cognitive control in language com-
response to morphosyntactic violations. Language and Cognitive Processes, 13(1), prehension. Frontiers in Psychology, 5, 335.
21–58. Fedorenko, E., Behr, M. K., & Kanwisher, N. (2011). Functional specificity for high-level
Crittenden, B. M., & Duncan, J. (2012). Task difficulty manipulation reveals multiple linguistic processing in the human brain. Proceedings of the National Academy of
demand activity but no frontal lobe hierarchy. Cerebral Cortex, 24(2), 532–540. Sciences, 108(39), 16428–16433.
Culicover, P. W., & Jackendoff, R. (1999). The view from the periphery: The English Fedorenko, E., Hsieh, P. J., Nieto-Castanon, A., Whitfield-Gabrieli, S., & Kanwisher, N.
comparative correlative. Linguistic Inquiry, 30(4), 543–571. (2010). New method for fMRI investigations of language: Defining ROIs functionally
Dąbrowska, E. (2018). Experience, aptitude and individual differences in native language in individual subjects. Journal of Neurophysiology, 104(2), 1177–1194.
ultimate attainment. Cognition, 178, 222–235. Fedorenko, E., & Kanwisher, N. (2009). Neuroimaging of language: why hasn't a clearer
Dale, A. M. (1999). Optimal experimental design for event-related fMRI. Human Brain picture emerged? Language and Linguistics Compass, 3(4), 839–865.
Mapping, 8(2–3), 109–114. Fedorenko, E., Nieto-Castanon, A., & Kanwisher, N. (2012). Lexical and syntactic re-
Dale, P. S., Dionne, G., Eley, T. C., & Plomin, R. (2000). Lexical and grammatical de- presentations in the brain: An fMRI investigation with multi-voxel pattern analyses.
velopment: A behavioural genetic perspective. Journal of Child Language, 27, Neuropsychologia, 50(4), 499–513.
619–642. Fedorenko, E., Duncan, J., & Kanwisher, N. (2012). Language-selective and domain-
Dale, P. S., Harlaar, N., Haworth, C. M. A., & Plomin, R. (2010). Two by two: A twin study general regions lie side by side within Broca's area. Current Biology, 22(21),
of second-language acquisition. Psychological Science, 21(5), 635–640. 2059–2062.

20
E. Fedorenko, et al. Cognition 203 (2020) 104348

Fedorenko, E., & Thompson-Schill, S. L. (2014). Reworking the language network. Trends Gordon, J. K., & Dell, G. S. (2003). Learning to divide the labor: An account of deficits in
in Cognitive Sciences, 18(3), 120–126. light and heavy verb production. Cognitive Science, 27(1), 1–40.
Fedorenko, E., Scott, T. L., Brunner, P., Coon, W. G., Pritchett, B., Schalk, G., & Gorno-Tempini, M. L., Hillis, A. E., Weintraub, S., Kertesz, A., Mendez, M., Cappa, S. F., ...
Kanwisher, N. (2016). Neural correlate of the construction of sentence meaning. Grossman, M. (2011). Classification of primary progressive aphasia and its variants.
Proceedings of the National Academy of Sciences, 113(41), E256–E262. Neurology, 76(11), 1006.
Fedorenko, E., Williams, Z. M., & Ferreira, V. S. (2018). Remaining puzzles about mor- Grabowski, T. J., Damasio, H., Tranel, D., Ponto, L. L. B., Hichwa, R. D., & Damasio, A. R.
pheme production in the posterior temporal lobe. Neuroscience, 392, 160–163. (2001). A role for left temporal pole in the retrieval of words for unique entities.
Ferreira, F., Bailey, K. G. D., & Ferraro, V. (2002). Good-enough representations in lan- Human Brain Mapping, 13(4), 199–212.
guage comprehension. Current Directions in Psychological Science, 11(1), 11–15. Grodzinsky, Y., & Santi, A. (2008). The battle for Broca’s region. Trends in Cognitive
Ferstl, E. C., & von Cramon, D. Y. (2001). The role of coherence and cohesion in text Sciences, 12(12), 474–480.
comprehension: an event-related fMRI study. Cognitive Brain Research, 11(3), Grossman, M., Lee, C., Morris, J., Stern, M. B., & Hurtig, H. I. (2002). Assessing resource
325–340. demands during sentence processing in Parkinson’s disease. Brain and Language, 80,
Fiebach, C. J., Schlesewsky, M., Lohmann, G., Von Cramon, D. Y., & Friederici, A. D. 603–616.
(2005). Revisiting the role of Broca’s area in sentence processing: syntactic integra- Gulordava, K., Bojanowski, P., Grave, E., Linzen, T., & Baroni, M. (2018). Colorless green
tion versus syntactic working memory. Human Brain Mapping, 24(2), 79–91. recurrent networks dream hierarchically. arXiv preprint (arXiv:1803.11138).
Fitch, W. T., & Martins, M. D. (2014). Hierarchical processing in music, language, and Hagoort, P. (2005). On Broca, brain, and binding: A new framework. Trends in Cognitive
action: Lashley revisited. Annals of the New York Academy of Sciences, 1316, 87–104. Sciences, 9(9), 416–423.
Fedorenko, E., Duncan, J., & Kanwisher, N. (2013). Broad domain-generality in focal Hagoort, P. (2013). MUC (Memory, Unification, Control) and beyond. Frontiers in
regions of frontal and parietal cortex. Proceedings of the National Academy of Sciences, Psychology, 4.
110(41), 16616–16621. Hagoort, P. (2019). The neurobiology of language beyond single-word processing.
Fodor, J. A. (1983). The modularity of mind. MIT Press. Science, 366(6461), 55–58.
Fodor, J. A., Bever, T. G., & Garrett, M. F. (1974). The psychology of language: An in- Hagoort, P., Brown, C., & Groothusen, J. (1993). The syntactic positive shift (SPS) as an
troduction to psycholinguistics and generative grammar. McGraw-Hill. ERP measure of syntactic processing. Language and Cognitive Processes, 8(4), 439–483.
Friederici, A. D. (2011). The brain basis of language processing: From structure to Hagoort, P., & Indefrey, P. (2014). The neurobiology of language beyond single words.
function. Physiological Reviews, 91(4), 1357–1392. Annual Review of Neuroscience, 37, 347.
Friederici, A. D. (2012). The cortical language circuit: From auditory perception to sen- Hahn, M., Jurafsky, D., & Futrell, R. (2020). Universals of word order reflect optimization
tence comprehension. Trends in Cognitive Sciences, 16(5), 262–268. of grammars for efficient communication. Proceedings of the National Academy of
Friederici, A. D. (2018). The neural basis for human syntax: Broca’s area and beyond. Sciences.. https://www.pnas.org/content/117/5/2347.
Current Opinion in Behavioral Sciences, 21, 88–92. Halai, A. D., Woollams, A. M., & Ralph, M. A. L. (2017). Using principal component
Friederici, A. D., Bahlmann, J., Heim, S., Schubotz, R. I., & Anwander, A. (2006). The analysis to capture individual differences within a unified neuropsychological model
brain differentiates human and non-human grammars: Functional localization and of chronic post-stroke aphasia: Revealing the unique neural correlates of speech
structural connectivity. Proceedings of the National Academy of Sciences, 103(7), fluency, phonology and semantics. Cortex, 86, 275–289.
2458–2463. Hamamé, C. M., Szwed, M., Sharman, M., Vidal, J. R., Perrone-Bertolotti, M., Kahane, P.,
Friederici, A. D., Chomsky, N., Berwick, R. C., Moro, A., & Bolhuis, J. J. (2017). Language, ... Lachaux, J. P. (2013). Dejerine’s reading area revisited with intracranial EEG:
mind and brain. Nature Human Behaviour, 1(10), 713. Selective responses to letter strings. Neurology, 80(6), 602–603.
Friederici, A. D., Kotz, S. A., Scott, S. K., & Obleser, J. (2010). Disentangling syntax and Hardy, S. M., Segaert, K., & Wheeldon, L. (2020). Healthy aging and sentence production:
intelligibility in auditory language comprehension. Human Brain Mapping, 31(3), Disrupted lexical access in the context of intact syntactic planning. Frontiers in
448–457. Psychology, 11, 257.
Friederici, A. D. (2009). Pathways to language: fiber tracts in the human brain. Trends in Hare, M. L., & Goldberg, A. E. (1999). Structural priming: Purely syntactic? Proceedings of
cognitive sciences, 13(4), 175–181. the Twenty First Annual Conference of the Cognitive Science Society, 208–211.
Friederici, A. D., Meyer, M., & von Cramon, D. Y. (2000). Auditory language compre- Hasson, U., Chen, J., & Honey, C. J. (2015). Hierarchical process memory: Memory as an
hension: An event-related fMRI study on the processing of syntactic and lexical in- integral component of information processing. Trends in Cognitive Sciences, 19(6),
formation. Brain and Language, 74(2), 289–300. 304–313.
Friederici, A. D. (2002). Towards a neural basis of auditory sentence processing. Trends in Hauser, M. D., Chomsky, N., & Fitch, W. T. (2002). The faculty of language: What is it,
cognitive sciences, 6(2), 78–84. who has it, and how did it evolve? Science, 298(5598), 1569–1579.
Friston, K. J., Rotshtein, P., Geng, J. J., Sterzer, P., & Henson, R. N. (2006). A critique of Heim, S., Eickhoff, S. B., & Amunts, K. (2008). Specialisation in Broca’s region for se-
functional localisers. Neuroimage, 30(4), 1077–1087. mantic, phonological, and syntactic fluency? Neuroimage, 40(3), 1362–1368.
Frost, M. A., & Goebel, R. (2012). Measuring structural-functional correspondence: Herrmann, B., Obleser, J., Kalberlah, C., Haynes, J. D., & Friederici, A. D. (2012).
Spatial variability of specialised brain regions after macro-anatomical alignment. Dissociable neural imprints of perception and grammar in auditory functional ima-
Neuroimage, 59(2), 1369–1381. ging. Human Brain Mapping, 33(3), 584–595.
Futrell, R., Wilcox, E., Morita, T., & Levy, R. (2018). RNNs as psycholinguistic subjects: Hickok, G., & Poeppel, D. (2007). The cortical organization of speech processing. Nature
Syntactic state and grammatical dependency. arXiv preprint (arXiv:1809.01329). Reviews Neuroscience, 8(5), 393–402.
Futrell, R., Wilcox, E., Morita, T., Qian, P., Ballesteros, M., & Levy, R. (2019). Neural Hoff, E., Quinn, J. M., & Giguere, D. (2018). What explains the correlation between
language models as psycholinguistic subjects: Representations of syntactic state. arXiv growth in vocabulary and grammar? New evidence from latent change score analyses
preprint (arXiv:1903.03260). of simultaneous bilingual development. Developmental Science, 21(2), e12536.
Garnsey, S. M., Pearlmutter, N. J., Myers, E., & Lotocky, M. A. (1997). The contributions Holmes, A., & Friston, K. (1998). Generalisability, random effects and population in-
of verb bias and plausibility to the comprehension of temporarily ambiguous sen- ference. NeuroImage, 7(4), S754.
tences. Journal of Memory and Language, 37(1), 58–93. Hong, Y., Yoo, Y., Wager, T., & Woo, C. W. (2019). False-positive neuroimaging:
Garrett, M. (2000). Remarks on the architecture of language processing systems. In Y. Undisclosed flexibility in testing spatial hypotheses allows presenting anything as a
Grodzinsky, L. Shapiro, & D. Swinney (Eds.). Language and the brain (pp. 31–69). San replicated finding. bioRxiv, 514–521.
Diego, CA: Academic Press. Hudson, R. (2007). Language networks: The new word grammar. Oxford University
Gibson, E., Bergen, L., & Piantadosi, S. T. (2013). Rational integration of noisy evidence Press UK.
and prior semantic expectations in sentence interpretation. Proceedings of the National Humphries, C., Willard, K., Buchsbaum, B., & Hickok, G. (2001). Role of anterior tem-
Academy of Sciences of the United States of America, 110(20), 8051–8056. poral cortex in auditory sentence comprehension: An fMRI study. NeuroReport, 12(8),
Gibson, E., Futrell, R., Piandadosi, S. T., Dautriche, I., Mahowald, K., Bergen, L., & Levy, 1749–1752.
R. (2019). How efficiency shapes human language. Trends in Cognitive Sciences. 23(5), Hurford, J. R. (1998). Approaches to evolution language: Social and cognitive bases.
389–407. Cambridge University Press.
Giraud, A. L., & Poeppel, D. (2012). Cortical oscillations and speech processing: emerging Hurford, J. R. (2007). The origins of meaning. Oxford University Press.
computational principles and operations. Nature Neuroscience, 15(4), 511. Indefrey, P., & Levelt, W. J. (2004). The spatial and temporal signatures of word pro-
Glaser, Y. G., Martin, R. C., Van Dyke, J. A., Hamilton, A. C., & Tan, Y. (2013). Neural duction components. Cognition, 92(1–2), 101–144.
basis of semantic and syntactic interference in sentence comprehension. Brain and Ivanova, A.*, Siegelman, M.*, Pongos, A., Kean, H., & Fedorenko, E. (in preparation).
Language, 126(3), 314–326. Language activations in the fronto-temporal network are robust to task manipula-
Glasser, M. F., & Rilling, J. K. (2008). DTI tractography of the human brain’s language tions.
pathways. Cerebral Cortex, 18(11), 2471–2482. Ioannidis, J. P. A., Munafo, M. R., Fusar-Poli, P., Nosek, B. A., & David, S. P. (2014).
Goldberg, A. (2002). Construction grammar. Encyclopedia of cognitive science. Macmillan Publication and other reporting biases in cognitive sciences: Detection, prevalence,
Reference Limited Nature Publishing Group. and prevention. Trends in Cognitive Sciences, 18(5), 235–241.
Goldberg, A. E. (1995). Constructions: A construction grammar approach to argument Jackendoff, R. (2002a). English particle constructions, the lexicon, and the autonomy of
structure. University of Chicago Press. syntax. Verb-particle Explorations, 67–94.
Goldberg, A. E. (2006). Constructions at work: The nature of generalization in language. Jackendoff, R. (2002b). Foundations of language: Brain, meaning, grammar, evolution.
Oxford University Press on Demand. Oxford University Press.
Goldberg, A. E. (2013). 10 Backgrounded constituents cannot be “extracted”. Experimental Jackendoff, R. (2007). A Parallel Architecture perspective on language processing. Brain
Syntax and Island Effects, 221. Research, 1146, 2–22.
Goldinger, S. D. (1996). Words and voices: episodic traces in spoken word identification Jackendoff, R. (2011). What is the human language faculty? Two views. Language, 87(3),
and recognition memory. Journal of Experimental Psychology: Learning, Memory, and 586–624.
Cognition, 22(5), 1166. Jackendoff, R., & Audring, J. (2020). The texture of the lexicon: Relational morphology and
Goodglass, H. (1993). Understanding aphasia. Academic Press. the Parallel Architecture. Oxford University Press.

21
E. Fedorenko, et al. Cognition 203 (2020) 104348

Jacoby, N., & Fedorenko, E. (2018). Discourse-level comprehension engages medial Brain and Language, 102(1), 22–29.
frontal Theory of Mind brain regions even for expository texts. Language, Cognition Linzen, T., Dupoux, E., & Goldberg, Y. (2016). Assessing the ability of LSTMs to learn
and Neuroscience, 0(0), 1–17. syntax-sensitive dependencies. Transactions of the Association for Computational
Jaeger, T. F. (2010). Redundancy and reduction: Speakers manage syntactic information Linguistics, 4, 521–535.
density. Cognitive Psychology, 61(1), 23–62. Loiotile, R. E., Cusack, R., & Bedny, M. (2019). Naturalistic audio-movies and narrative
Joanette, Y., Goulet, P., & Hannequin, D. (1990). Right hemisphere and verbal commu- synchronize “visual” cortices across congenitally blind but not sighted individuals.
nication. Springer-Verlag. Journal of Neuroscience, 39(45), 8940–8948.
Kaan, E., & Swaab, T. Y. (2002). The brain circuitry of syntactic comprehension. Trends in Longworth, C. E., Keenan, S. E., Barker, R. A., Marslen-Wilson, W. D., & Tyler, L. K.
Cognitive Sciences, 6(8), 350–356. (2005). The basal ganglia and rule-governed language use: Evidence from vascular
Kanwisher, N., McDermott, J., & Chun, M. M. (1997). The fusiform face area: A module in and degenerative conditions. Brain, 128(3), 584–596.
human extrastriate cortex specialized for face perception. Journal of Neuroscience, Luke, K. K., Liu, H. L., Wai, Y. Y., Wan, Y. L., & Tan, L. H. (2002). Functional anatomy of
17(11), 4302–4311. syntactic and semantic processing in language comprehension. Human brain mapping,
Kanwisher, N. (2010). Functional specificity in the human brain: a window into the 16(3), 133–145.
functional architecture of the mind. Proceedings of the National Academy of Sciences, MacDonald, M. C., Pearlmutter, N. J., & Seidenberg, M. S. (1994). The lexical nature of
107(25), 11163–11170. syntactic ambiguity resolution. Psychological Review, 101, 676–703.
Keller, T. A., Carpenter, P. A., & Just, M. A. (2001). The neural bases of sentence com- Mahowald, K., & Fedorenko, E. (2016). Reliable individual-level neural markers of high-
prehension: A fMRI examination of syntactic and lexical processing. Cerebral Cortex, level language processing: A necessary precursor for relating neural variability to
11(3), 223–237. behavioral and genetic variability. Neuroimage, 139, 74–93.
Kirby, S., Tamariz, M., Cornish, H., & Smith, K. (2015). Compression and communication Mahowald, K., James, A., Futrell, R., & Gibson, E. (2016). A meta-analysis of syntactic
in the cultural evolution of linguistic structure. Cognition, 141, 87–102. priming in language production. Journal of Memory and Language, 91, 5–27.
Kline, M., Gallée, J., Balewski, Z., & Fedorenko, E. (2018). Understanding jokes draws most Marchman, V. A., & Bates, E. (1994). Continuity in lexical and morphological develop-
heavily on the Theory of Mind brain network. PsyArXiv preprinthttps://doi.org/10. ment: A test of the critical mass hypothesis. Journal of Child Language, 21(2),
31234/osf.io/h2nyx. 339–366.
Kochari, A., Lewis, A., Schoffelen, J. M., & Schriefers, H. (2020). Semantic and syntactic Mariën, P., Ackermann, H., Adamaszek, M., Barwood, C. H., Beaton, A., Desmond, J., ...
composition of minimal adjective-noun phrases in Dutch: An MEG study. bioRxiv. Leggio, M. (2014). Consensus paper: language and the cerebellum: An ongoing en-
Koechlin, E., & Jubault, T. (2006). Broca's area and the hierarchical organization of igma. The Cerebellum, 13(3), 386–410.
human behavior. Neuron, 50(6), 963–974. Martin, A. E., & Doumas, L. A. (2019). Predicate learning in neural systems: using os-
Kojima, K., Brown, E. C., Matsuzaki, N., Rothermel, R., Fuerst, D., Shah, A., ... Asano, E. cillations to discover latent structure. Current Opinion in Behavioral Sciences, 29,
(2013). Gamma activity modulated by picture and auditory naming tasks: 77–83.
Intracranial recording in patients with focal epilepsy. Clinical Neurophysiology, Mason, R. A., & Just, M. A. (2007). Lexical ambiguity in sentence comprehension. Brain
124(9), 1737–1744. Research, 1146, 115–127.
Kolk, H., & Chwilla, D. (2007). Late positivities in unusual situations. Brain and Language, Mason, R. A., Just, M. A., Keller, T. A., & Carpenter, P. A. (2003). Ambiguity in the brain:
100(3), 257–261. What brain imaging reveals about the processing of syntactically ambiguous sen-
Kolk, H. H., van Grunsven, M. J., & Keyser, A. (1985). On parallelism between production tences. Journal of Experimental Psychology-Learning Memory and Cognition, 29(6),
and comprehension in agrammatism. Agrammatism (pp. 165–206). Academic Press. 1319–1338.
Krekelberg, B., Boynton, G. M., & van Wezel, R. J. A. (2006). Adaptation: From single cells Matchin, W., Basilakos, A., Stark, B. C., den Ouden, D. B., Fridriksson, J., & Hickok, G.
to BOLD signals. Trends in Neurosciences, 29(5), 250–256. (2020). Agrammatism and paragrammatism: A cortical double dissociation revealed
Kriegeskorte, N., Simmons, W. K., Bellgowan, P. S., & Baker, C. I. (2009). Circular analysis by lesion-symptom mapping. Neurobiology of Language, 1–47.
in systems neuroscience: The dangers of double dipping. Nature Neuroscience, 12(5), Matchin, W., Brodbeck, C., Hammerly, C., & Lau, E. (2019). The temporal dynamics of
535–540. structure and content in sentence comprehension: Evidence from fMRI-constrained
Kuno, S. (1987). Functional syntax: Anaphora, discourse and empathy. University of Chicago MEG. Human Brain Mapping, 40(2), 663–678.
Press. Matchin, W., & Hickok, G. (2019). The cortical organization of syntax. Cerebral Cortex.
Kuperberg, G. R., Holcomb, P. J., Sitnikova, T., Greve, D., Dale, A. M., & Caplan, D. 30(3), 1481–1498.
(2003). Distinct patterns of neural modulation during the processing of conceptual Mather, M., Cacioppo, J. T., & Kanwisher, N. (2013). How fMRI can inform cognitive
and syntactic anomalies. Journal of Cognitive Neuroscience, 15(2), 272–293. theories. Perspectives on Psychological Science, 8(1), 108–113.
Kuperberg, G. R., McGuire, P. K., Bullmore, E. T., Brammer, M. J., Rabe-Hesketh, S., Mazoyer, B. M., Tzourio, N., Frak, V., Syrota, A., Murayama, N., Levrier, O., ... Mehler, J.
Wright, I. C., ... David, A. S. (2000). Common and distinct neural substrates for (1993). The cortical representation of speech. Journal of Cognitive Neuroscience, 5(4),
pragmatic, semantic, and syntactic processing of spoken sentences: An fMRI study. 467–479.
Journal of Cognitive Neuroscience, 12(2), 321–341. McCandliss, B. D., Cohen, L., & Dehaene, S. (2003). The visual word form area: Expertise
Kutas, M., & Federmeier, K. D. (2011). Thirty years and counting: Finding meaning in the for reading in the fusiform gyrus. Trends in Cognitive Sciences, 7(7), 293–299.
N400 component of the event-related brain potential (ERP). Annual Review of Mcgregor, K. K., Sheng, L. I., & Smith, B. (2005). The precocious two-year-old: Status of
Psychology, 62(62), 621–647. the lexicon and links to the grammar. Journal of Child Language, 32(3), 563–585.
Kutas, M., & Hillyard, S. A. (1980). Reading senseless sentences: Brain potentials reflect Menenti, L., Gierhan, S. M. E., Segaert, K., & Hagoort, P. (2011). Shared language:
semantic incongruity. Science, 207(4427), 203–205. Overlap and segregation of the neuronal infrastructure for speaking and listening
Kvarven, A., Strømland, E., & Johannesson, M. (2019). Comparing meta-analyses and revealed by functional MRI. Psychological Science, 22(9), 1173–1182.
preregistered multiple-laboratory replication projects. Nature Human Behaviour, Menenti, L., Petersson, K. M., & Hagoort, P. (2012). From reference to sense: How the
1–12. brain encodes meaning for speaking. Frontiers in Psychology, 3.
Lakoff, G. (1970). Irregularity in syntax. New York: Holt, Rinehart, and Winston. Mesgarani, N., Cheung, C., Johnson, K., & Chang, E. F. (2014). Phonetic feature encoding
Lambon-Ralph, M. A., McClelland, J. L., Patterson, K., Galton, C. J., & Hodges, J. R. in human superior temporal gyrus. Science, 343(6174), 1006–1010.
(2001). No right to speak? The relationship between object naming and semantic Mesulam, M. M., Rogalski, E. J., Wieneke, C., Hurley, R. S., Geula, C., Bigio, E. H., ...
impairment: Neuropsychological evidence and a computational model. Journal of Weintraub, S. (2014). Primary progressive aphasia and the evolving neurology of the
Cognitive Neuroscience, 13(3), 341–356. language network. Nature Reviews Neurology, 10(10), 554.
Lane, C., Kanjlia, S., Omaki, A., & Bedny, M. (2015). “Visual” cortex of congenitally blind Mesulam, M.-M., Thompson, C. K., Weintraub, S., & Rogalski, E. J. (2015). The Wernicke
adults responds to syntactic movement. Journal of Neuroscience, 35(37), conundrum and the anatomy of language comprehension in primary progressive
12859–12868. aphasia. Brain, 138(8), 2423–2437.
Langacker, R. W. (1986). An introduction to cognitive grammar. Cognitive Science, 10(1), Mesulam, M. M., Wieneke, C., Hurley, R., Rademaker, A., Thompson, C. K., Weintraub, S.,
1–40. & Rogalski, E. J. (2013). Words and objects at the tip of the left temporal lobe in
Langacker, R. W. (1987). Foundations of cognitive grammar: Theoretical prerequisites. Vol. 1. primary progressive aphasia. Brain, 136(2), 601–618.
Stanford University Press. Meyer, L. (2018). The neural oscillations of speech processing and language compre-
Lee, D. K., Fedorenko, E., Simon, M. V., Curry, W. T., Nahed, B. V., Cahill, D. P., & hension: State of the art and emerging mechanisms. European Journal of Neuroscience,
Williams, Z. M. (2018). Neural encoding and production of functional morphemes in 48(7), 2609–2621.
the posterior temporal lobe. Nature Communications, 9(1), 1–12. Miceli, G., & Caramazza, A. (1988). Dissociation of inflectional and derivational mor-
Lerner, Y., Honey, C. J., Silbert, L. J., & Hasson, U. (2011). Topographic mapping of a phology. Brain and Language, 35(1), 24–65.
hierarchy of temporal receptive windows using a narrated story. Journal of Miceli, G., Mazzucchi, A., Menn, L., & Goodglass, H. (1983). Contrasting cases of Italian
Neuroscience, 31(8), 2906–2915. agrammatic aphasia without comprehension disorder. Brain and Language, 19(1),
Levin, B. (1993). English verb classes and alternations: A preliminary investigation. University 65–97.
of Chicago press. Miceli, G., Silveri, M. C., Romani, C., & Caramazza, A. (1989). Variation in the pattern of
Levin, B., & Rappaport-Hovav, M. (2005). Argument Realization. Cambridge, UK: omissions and substitutions of grammatical morphemes in the spontaneous speech of
Cambridge University Press. so-called agrammatic patients. Brain and Language, 36(3), 447–492.
Levy, R. (2008). A noisy-channel model of human sentence comprehension under un- Mikolov, T., Karafiát, M., Burget, L., Černocký, J., & Khudanpur, S. (2010). Recurrent
certain input. Proceedings of the 2008 conference on empirical methods in natural lan- neural network based language model. Eleventh Annual Conference of the International
guage processing (pp. 234–243). . Speech Communication Association.
Levy, R., Bicknell, K., Slattery, T., & Rayner, K. (2009). Eye movement evidence that Mirman, D., Chen, Q., Zhang, Y., Wang, Z., Faseyitan, O. K., Coslett, H. B., & Schwartz, M.
readers maintain and act on uncertainty about past linguistic input. Proceedings of the F. (2015). Neural organization of spoken language revealed by lesion–symptom
National Academy of Sciences, 106(50), 21086–21090. mapping. Nature Communications, 6(1), 1–9.
Lindenberg, R., Fangerau, H., & Seitz, R. J. (2007). “Broca’s area” as a collective term? Mirman, D., Kraft, A. E., Harvey, D. Y., Brecher, A. R., & Schwartz, M. F. (2019). Mapping

22
E. Fedorenko, et al. Cognition 203 (2020) 104348

articulatory and grammatical subcomponents of fluency deficits in post-stroke Poldrack, R. A., Baker, C. I., Durnez, J., Gorgolewski, K. J., Matthews, P. M., Munafò, M.
aphasia. Cognitive, Affective, & Behavioral Neuroscience, 19(5), 1286–1298. R., & Yarkoni, T. (2017). Scanning the horizon: Towards trans- parent and re-
Miyake, A., Carpenter, P. A., & Just, M. A. (1994). A capacity approach to syntactic producible neuroimaging research. Nature Reviews. Neuroscience, 18(2), 115.
comprehension disorders: Making normal adults perform like aphasic patients. Poldrack, R. A., Laumann, T. O., Koyejo, O., Gregory, B., Hover, A., Chen, M.-Y., ...
Cognitive Neuropsychology, 11, 671–717. Mumford, J. A. (2015). Long-term neural and physiological phenotyping of a single
Mollica, F., & Piantadosi, S. T. (2019). Humans store about 1.5 megabytes of information human. Nature Communications, 6(1), 1–15.
during language acquisition. Royal Society Open Science, 6(3), 181393. Poldrack, R. A., Wagner, A. D., Prull, M. W., Desmond, J. E., Glover, G. H., & Gabrieli, J.
Mollica, F., Shain, C., Affourtit, J., Kean, H., Siegelman, M., and Fedorenko, E. (in prep) D. E. (1999). Functional specialization for semantic and phonological processing in
Another look at the constituent structure of sentences in the human brain. the left inferior prefrontal cortex. NeuroImage, 10(1), 15–35.
Mollica, F., Siegelman, M., Diachek, E., Piantadosi, S. T., Mineroff, Z., Futrell, R., ... Prasad, G., van Schijndel, M., & Linzen, T. (2019). Using priming to uncover the organization
Fedorenko, E. (2020). Composition is the core driver of the language-selective net- of syntactic representations in neural language models. arXiv preprint
work. Neurobiology of Language, 1–67. (arXiv:1909.10579).
Monti, M. M., Parsons, L. M., & Osherson, D. N. (2012). Thought beyond language: Neural Pylkkänen, L. (2019). The neural basis of combinatory syntax and semantics. Science,
dissociation of algebra and natural language. Psychological Science, 23(8), 914–922. 366(6461), 62–66.
Morgan, E., & Levy, R. (2016). Abstract knowledge versus direct experience in processing Python, G., Glize, B., & Laganaro, M. (2018). The involvement of left inferior frontal and
of binomial expressions. Cognition, 157, 384–402. middle temporal cortices in word production unveiled by greater facilitation effects
Moyle, M. J., Weismer, S. E., Evans, J. L., & Lindstrom, M. J. (2007). Longitudinal re- following brain damage. Neuropsychologia, 121, 122–134.
lationships between lexical and grammatical development in typical and late-talking Rauschecker, J. P., & Scott, S. K. (2009). Maps and streams in the auditory cortex:
children. Journal of Speech, Language, and Hearing Research. 50(2), 508–528. https:// Nonhuman primates illuminate human speech processing. Nature Neuroscience, 12(6),
doi.org/10.1044/1092-4388(2007/035). 718–724.
Mukamel, R., & Fried, I. (2012). Human intracranial recordings and cognitive neu- Reali, F., & Christiansen, M. H. (2007). Processing of relative clauses is made easier by
roscience. Annual Review of Psychology, 63, 511–537. frequency of occurrence. Journal of Memory and Language, 57(1), 1–23.
Newman, A. J., Pancheva, R., Ozawa, K., Neville, H. J., & Ullman, M. T. (2001). An event- Reber, A. S. (1967). Implicit learning of artificial grammars. Journal of Verbal Learning and
related fMRI study of syntactic and semantic violations. Journal of Psycholinguistic Verbal Behavior, 6(6), 855–863.
Research, 30(3), 339–364. Riès, S. K., Dhillon, R. K., Clarke, A., King-Stephens, D., Laxer, K. D., Weber, P. B., ... Lin,
Newmeyer, F. J. (2003). Grammar is grammar and usage is usage. Language, 79(4), J. J. (2017). Spatiotemporal dynamics of word retrieval in speech production re-
682–707. vealed by cortical high-frequency band activity. Proceedings of the National Academy
Ni, W., Constable, R. T., Mencl, W. E., Pugh, K. R., Fulbright, R. K., Shaywitz, S. E., ... of Sciences, 114(23), E4530–E4538.
Shankweiler, D. (2000). An event-related neuroimaging study distinguishing form Rodd, J. M., Berriman, R., Landau, M., Lee, T., Ho, C., Gaskell, M. G., & Davis, M. H.
and content in sentence processing. Journal of Cognitive Neuroscience, 12(1), 120–133. (2012). Learning new meanings for old words: Effects of semantic relatedness.
Nieto-Castañón, A., & Fedorenko, E. (2012). Subject-specific functional localizers increase Memory & Cognition, 40(7), 1095–1108.
sensitivity and functional resolution of multi-subject analyses. NeuroImage, 63(3), Rodd, J. M., Davis, M. H., & Johnsrude, I. S. (2005). The neural mechanisms of speech
1646–1669. comprehension: FMRI studies of semantic ambiguity. Cerebral Cortex (New York, N.Y.:
Nieuwenhuis, S., Forstmann, B. U., & Wagenmakers, E. J. (2011). Erroneous analyses of 1991), 15(8), 1261–1269.
interactions in neuroscience: A problem of significance. Nature Neuroscience, 14(9), Rodd, J. M., Longe, O. A., Randall, B., & Tyler, L. K. (2010). The functional organisation of
1105–1107. the fronto-temporal language system: Evidence from syntactic and semantic ambi-
Nieuwland, M. S., Martin, A. E., & Carreiras, M. (2012). Brain regions that process case: guity. Neuropsychologia, 48(5), 1324–1335.
evidence from Basque. Human brain mapping, 33(11), 2509–2520. Rodd, J. M., Vitello, S., Woollams, A. M., & Adank, P. (2015). Localising semantic and
Nespoulous, J. L., Dordain, M., Perron, C., Ska, B., Bub, D., Caplan, D., ... Lecours, A. R. syntactic processing in spoken and written language comprehension: An Activation
(1988). Agrammatism in sentence production without comprehension deficits: Likelihood Estimation meta-analysis. Brain and Language, 141, 89–102.
Reduced availability of syntactic structures and/or of grammatical morphemes? A Roder, B., Stock, O., Neville, H., Bien, S., & Rosler, F. (2002). Brain activation modulated
case study. Brain and language, 33(2), 273–295. by the comprehension of normal and pseudo-word sentences of different processing
Noppeney, U., & Price, C. J. (2004). An fMRI study of syntactic adaptation. Journal of demands: A functional magnetic resonance Imaging study. Neuroimage, 15(4),
Cognitive Neuroscience, 16(4), 702–713. 1003–1014.
Norcliffe, E., Harris, A. C., & Jaeger, T. F. (2015). Cross-linguistic psycholinguistics and its Rogalsky, C., & Hickok, G. (2009). Selective attention to semantic and syntactic features
critical role in theory development: Early beginnings and recent advances. modulates sentence processing networks in anterior temporal cortex. Cerebral Cortex,
Norman-Haignere, S., Kanwisher, N. G., & McDermott, J. H. (2015). Distinct cortical 19(4), 786–796.
pathways for music and speech revealed by hypothesis-free voxel decomposition. Rogers, T. T., Hocking, J., Noppeney, U. T. A., Mechelli, A., Gorno-Tempini, M. L.,
Neuron, 88(6), 1281–1296. Patterson, K., & Price, C. J. (2006). Anterior temporal cortex and semantic memory:
Norris, D. G., & Polimeni, J. R. (2019). Laminar (f)MRI: A short history and future pro- Reconciling findings from neuropsychology and functional imaging. Cognitive,
spects. Neuroimage, 197, 643. Affective, & Behavioral Neuroscience, 6(3), 201–213.
Oldfield, R. C. (1971). The assessment and analysis of handedness: the Edinburgh in- Rogers, T. T., Lambon Ralph, M. A., Garrard, P., Bozeat, S., McClelland, J. L., Hodges, J.
ventory. Neuropsychologia, 9(1), 97–113. R., & Patterson, K. (2004). Structure and deterioration of semantic memory: A neu-
Osterhout, L., & Holcomb, P. J. (1992). Event-related brain potentials elicited by syntactic ropsychological and computational investigation. Psychological Review, 111(1), 205.
anomaly. Journal of Memory and Language, 31(6), 785–806. Roland, D., Dick, F., & Elman, J. L. (2007). Frequency of basic English grammatical
Overath, T., McDermott, J. H., Zarate, J. M., & Poeppel, D. (2015). The cortical analysis of structures: A corpus analysis. Journal of Memory and Language, 57(3), 348–379.
speech-specific temporal structure revealed by responses to sound quilts. Nature Rosenfelt, L., Barkley, C., Belvin, K., Lee, C., Federmeier, K., Kluender, R., & Kutas, M.
Neuroscience, 18(6), 903–911. (2009). No ERP evidence for first pass structure building: Pure word category violations do
Pallier, C., Devauchelle, A. D., & Dehaene, S. (2011). Cortical representation of the not elicit early negativity. CUNY Presentation (Davis, CA).
constituent structure of sentences. Proceedings of the National Academy of Sciences of Rosenfelt, L., Kluender, R., & Kutas, M. (2011). Severing the tie between the eLAN and
the United States of America, 108(6), 2522–2527. automatic syntactic processing. (AMLaP presentation (Paris, France)).
Pant, R., Kanjlia, S., & Bedny, M. (2020). A sensitive period in the neural phenotype of Ross, J. R. (1967). Constraints on variables in syntax.
language in blind individuals. Developmental Cognitive Neuroscience, 41, 100744. Rumelhart, D., & McClelland, J. (1986). On learning the past tenses of English verbs. In D.
Papoutsi, M., Stamatakis, E. A., Griffiths, J., Marslen-Wilson, W. D., & Tyler, L. K. (2011). Rumelhart, & J. McClelland (Eds.). Parallel Distributed Processing: Explorations in the
Is left fronto-temporal connectivity essential for syntax? Effective connectivity, Microstructure of Cognition (pp. 216–271). MIT Press.
tractography and performance in left-hemisphere damaged patients. NeuroImage, Ryskin, R., Levy, R. P., & Fedorenko, E. (2020). Do domain-general executive resources
58(2), 656–664. play a role in linguistic prediction? Re-evaluation of the evidence and a path forward.
Patterson, K., Nestor, P. J., & Rogers, T. T. (2007). Where do you know what you know? Neuropsychologia, 136, 107258.
The representation of semantic knowledge in the human brain. Nature Reviews. Sag, I. A., & Wasow, T. (2000). Syntactic theory: A formal introduction. Computational
Neuroscience, 8(12), 976–987. Linguistics, 26(2), 295.
Petersson, K. M., Folia, V., & Hagoort, P. (2012). What artificial grammar learning reveals Sanai, N., Mirzadeh, Z., & Berger, M. S. (2008). Functional outcome after language
about the neurobiology of syntax. Brain and Language, 120(2), 83–95. mapping for glioma resection. New England Journal of Medicine, 358(1), 18–27.
Pickering, M. J., & Ferreira, V. S. (2008). Structural priming: A critical review. Santi, A., & Grodzinsky, Y. (2010). fMRI adaptation dissociates syntactic complexity di-
Psychological Bulletin, 134(3), 427–459. mensions. Neuroimage, 51(4), 1285–1293.
Pinker, S. (1989). Learnability and cognition: The acquisition of argument structure. Sassenhagen, J., Schlesewsky, M., & Bornkessel-Schlesewsky, I. (2014). The P600-as-P3
Cambridge, MA: MIT Press. hypothesis revisited: Single-trial analyses reveal that the late EEG positivity following
Pinker, S. (1991). Rules of language. Science, 253(5019), 530–535. linguistically deviant material is reaction time aligned. Brain and Language, 137,
Pinker, S. (1995). The language instinct. William Morrow and Company. 29–39.
Pinker, S. (1999). Words and rules: The ingredients of language. London: Weidenfeld & Saxe, R., Brett, M., & Kanwisher, N. (2006). Divide and conquer: A defense of functional
Nicolson. localizers. NeuroImage, 30(4), 1088–1096.
Pinker, S., & Prince, A. (1988). On language and connectionism: Analysis of a parallel Saxe, R., & Kanwisher, N. (2003). People thinking about thinking people: The role of the
distributed processing model of language acquisition. Cognition, 28(1), 73–193. temporo-parietal junction in “theory of mind”. Neuroimage, 19(4), 1835–1842.
Poldrack, R. A. (2006). Can cognitive processes be inferred from neuroimaging data? Scheepers, C., Raffray, C., & Myachykov, A. (2017). The lexical boost effect is not diag-
Trends in Cognitive Sciences, 10(2), 59–63. nostic of lexically-specific syntactic representations. Memory and Language, 95,
Poldrack, R. A. (2011). Inferring mental states from neuroimaging data: From reverse 102–115.
inference to large-scale decoding. Neuron, 72(5), 692–697. Schell, M., Zaccarella, E., & Friederici, A. D. (2017). Differential cortical contribution of

23
E. Fedorenko, et al. Cognition 203 (2020) 104348

syntax and semantics: An fMRI study on two-word phrasal processing. Cortex, 96, macaques and humans. Proceedings of the National Academy of Sciences, 105(49),
105–120. 19514–19519.
Schmidt, S. (2009). Shall we really do it again? The powerful concept of replication is Tyler, L. K., Marslen-Wilson, W. D., Randall, B., Wright, P., Devereux, B. J., Zhuang, J., ...
neglected in the social sciences. Review of General Psychology, 13, 90–100. Stamatakis, E. A. (2011). Left inferior frontal cortex and syntax: Function, structure
Schnur, T. T., Schwartz, M. F., Kimberg, D. Y., Hirshorn, E., Coslett, H. B., & Thompson- and behaviour in patients with left hemisphere damage. Brain, 134, 415–431.
Schill, S. L. (2009). Localizing interference during naming: Convergent neuroimaging https://doi.org/10.1093/brain/awq369.
and neuropsychological evidence for the function of Broca’s area. Proceedings of the Ullman, M. T. (2001). A neurocognitive perspective on language: The declarative/pro-
National Academy of Sciences, 106(1), 322–327. cedural model. Nature Reviews Neuroscience, 2, 717–726.
Schotter, E. R., Tran, R., & Rayner, K. (2014). Don’t believe what you read (only once): Ullman, M. T. (2004). Contributions of memory circuits to language: The declarative/
Comprehension is supported by regressions during reading. Psychological Science, procedural model. Cognition, 92(1–2), 231–270.
25(6), 1218–1226. Ullman, M. T. (2016). The declarative/procedural model: A neurobiological model of
Schrimpf, M., Blank, I.A., Tuckute, G., Kauf, C., Hosseini, E., Kanwisher, N., Tenenbaum, language learning, knowledge, and use. Neurobiology of language. Academic Press.
J., & Fedorenko, E. (in prep). Machine learning models accurately predict language van de Meerendonk, N., Indefrey, P., Chwilla, D. J., & Kolk, H. H. J. (2011). Monitoring in
processing in the human brain. language perception: Electrophysiological and hemodynamic responses to spelling
Schütze, C. T., Sprouse, J., & Caponigro, I. (2015). Challenges for a theory of islands: A violations. NeuroImage, 54(3), 2350–2363.
broader perspective on Ambridge, Pine, and Lieven. Language, 91(2), e31–e39. van de Meerendonk, N., Kolk, H. H. J., Vissers, C., & Chwilla, D. J. (2010). Monitoring in
Scott, S. K., Blank, C. C., Rosen, S., & Wise, R. J. (2000). Identification of a pathway for language perception: Mild and strong conflicts elicit different ERP patterns. Journal of
intelligible speech in the left temporal lobe. Brain: A Journal of Neurology, 123(Pt 12), Cognitive Neuroscience, 22(1), 67–82.
2400–2406. Vandenberghe, R., Nobre, A. C., & Price, C. J. (2002). The response of left temporal cortex
Scott, T. L., Gallee, J., & Fedorenko, E. (2016). A new fun and robust version of an fMRI to sentences. Journal of Cognitive Neuroscience, 14(4), 550–560.
localizer for the frontotemporal language system. Cognitive Neuroscience, 1–10. Vázquez-Rodríguez, B., Suárez, L. E., Markello, R. D., Shafiei, G., Paquola, C., Hagmann,
Segaert, K., Menenti, L., Weber, K., Petersson, K. M., & Hagoort, P. (2012). Shared syntax P., ... Misic, B. (2019). Gradients of structure–function tethering across neocortex.
in language production and language comprehension - An fMRI Study. Cerebral Proceedings of the National Academy of Sciences, 116(42), 21219–21227.
Cortex, 22(7), 1662–1670. Vigliocco, G., & Hartsuiker, R. J. (2002). The interplay of meaning, sound, and syntax in
Seidenberg, M. S. (1994). Language and connectionism: The developing interface. sentence production. Psychological Bulletin, 128(3), 442–472.
Cognition, 50, 385–401. Vigneau, M., Beaucousin, V., Hervé, P. Y., Duffau, H., Crivello, F., Houdé, O., ... Tzourio-
Shain, C., Blank, I. A., van Schijndel, M., Schuler, W., & Fedorenko, E. (2020). fMRI Mazoyer, N. (2006). Meta-analyzing left hemisphere language areas: Phonology, se-
reveals language-specific predictive coding during naturalistic sentence comprehen- mantics, and sentence processing. NeuroImage, 30(4), 1414–1432.
sion. Neuropsychologia, 107307. Vissers, C., Chwilla, D. J., & Kolk, H. H. J. (2006). Monitoring in language perception: The
Siegelman, M., Blank, I. A., Mineroff, Z., & Fedorenko, E. (2019). An attempt to con- effect of misspellings of words in highly constrained sentences. Brain Research, 1106,
ceptually replicate the dissociation between syntax and semantics during sentence 150–163. https://doi.org/10.1016/j.brainres.2006.05.012.
comprehension. Neuroscience, 413, 219–229. Vissers, C., Chwilla, D. J., & Kolk, H. H. J. (2007). The interplay of heuristics and parsing
Silbert, L. J., Honey, C. J., Simony, E., Poeppel, D., & Hasson, U. (2014). Coupled neural routines in sentence comprehension: Evidence from ERPs and reaction times.
systems underlie the production and comprehension of naturalistic narrative speech. Biological Psychology, 75(1), 8–18.
Proceedings of the National Academy of Sciences, 111(43), E4687–E4696. Wang, J., Rice, M. L., & Booth, J. R. (2020). Syntactic and semantic specialization and
Simmons, J. P., Nelson, L. D., & Simonsohn, U. (2011). False-positive psychology: integration in 5-to 6-year-old children during auditory sentence processing. Journal of
Undisclosed flexibility in data collection and analysis allows presenting anything as Cognitive Neuroscience, 32(1), 36–49.
significant. Psychological Science, 22(11), 1359–1366. Wang, L., Uhrig, L., Jarraya, B., & Dehaene, S. (2015). Representation of numerical and
Snedeker, J., Geren, J., & Shafto, C. L. (2007). Starting over: International adoption as a sequential patterns in macaque and human brains. Current Biology, 25(15),
natural experiment in language development. Psychological Science, 18(1), 79–87. 1966–1974.
Snedeker, J., Geren, J., & Shafto, C. L. (2012). Disentangling the effects of cognitive Wehbe, L., Blank, I., Shain, C., Futrell, R., Levy, R., Malsburg, T., ... Fedorenko, E. (2020).
development and linguistic expertise: A longitudinal study of the acquisition of Incremental language comprehension difficulty predicts activity in the language
English in internationally-adopted children. Cognitive Psychology, 65(1), 39–76. network but not the multiple demand network. bioRxiv. https://doi.org/10.1101/
Snider, N., & Arnon, I. (2012). A unified lexicon and grammar? Compositional and non- 2020.04.15.043844.
compositional phrases in the lexicon. Frequency effects in language representation, Wilcox, E., Levy, R., & Futrell, R. (2019a). Hierarchical representation in neural language
127–163. models: Suppression and recovery of expectations. arXiv preprint (arXiv:1906.04068).
Steinhauer, K., & Drury, J. E. (2012). On the early left-anterior negativity (ELAN) in Wilcox, E., Levy, R., & Futrell, R. (2019b). What syntactic structures block dependencies in
syntax studies. Brain and Language, 120(2), 135–162. RNN language models? arXiv preprint (arXiv: 1905.10431).
Stromswold, K., Caplan, D., Alpert, N., & Rauch, S. (1996). Localization of syntactic Willems, R. M., der Haegen, L. V., Fisher, S. E., & Francks, C. (2014). On the other hand:
comprehension by positron emission tomography. Brain and Language, 52(3), Including left-handers in cognitive neuroscience and neurogenetics. Nature Reviews
452–473. Neuroscience, 15(3), 193–201.
Sutskever, I., Vinyals, O., & Le, Q. V. (2014). Sequence to sequence learning with neural Wilson, S. M., Galantucci, S., Tartaglia, M. C., & Gorno-Tempini, M. L. (2012). The neural
networks. ArXiv. (1409.3215 [Cs]) http://arxiv.org/abs/1409.3215. basis of syntactic deficits in primary progressive aphasia. Brain and Language, 122(3),
Tahmasebi, A. M., Davis, M. H., Wild, C. J., Rodd, J. M., Hakyemez, H., Abolmaesumi, P., 190–198.
& Johnsrude, I. S. (2012). Is the link between anatomical structure and function Wilson, S. M., Galantucci, S., Tartaglia, M. C., Rising, K., Patterson, D. K., Henry, M. L., ...
equally strong at all cognitive levels of processing? Cerebral Cortex (New York, N.Y.: Gorno-Tempini, M. L. (2011). Syntactic processing depends on dorsal language tracts.
1991), 22(7), 1593–1603. Neuron, 72(2), 397–403.
Tettamanti, M., & Weniger, D. (2006). Broca's area: A supramodal hierarchical processor? Wilson, S. M., Henry, M. L., Besbris, M., Ogar, J. M., Dronkers, N. F., Jarrold, W., & Gorno-
Cortex, 42(4), 491–494. Tempini, M. L. (2010). Connected speech production in three variants of primary
Thompson, C. K., Fix, S., & Gitelman, D. (2002). Selective impairment of morphosyntactic progressive aphasia. Brain, 133, 2069–2088.
production in a neurological patient. Journal of Neurolinguistics, 15(3–5), 189–207. Wilson, S. M., & Saygin, A. P. (2004). Grammaticality judgment in aphasia: Deficits are
Tie, Y., Rigolo, L., Norton, I. H., Huang, R. Y., Wu, W., Orringer, D., ... Golby, A. J. (2014). not specific to syntactic structures, aphasic syndromes, or lesion sites. Journal of
Defining language networks from resting-state fMRI for surgical planning—A feasi- Cognitive Neuroscience, 16(2), 238–252.
bility study. Human Brain Mapping, 35(3), 1018–1030. Wise, R., Greene, J., Büchel, C., & Scott, S. (1999). Brain regions involved in articulation.
Tomasello, M. (2003). Constructing a language: A usage-based theory of language acquisition. The Lancet, 353(9158), 1057–1061.
Cambridge, MA: Harvard University Press. Wray, A. (2005). Formulaic language and the lexicon. Cambridge University Press.
Tranel, D. (2006). Impaired naming of unique landmarks is associated with left temporal Zempleni, M.-Z., Renken, R., Hoeks, J. C. J., Hoogduin, J. M., & Stowe, L. A. (2007).
polar damage. Neuropsychology, 20(1), 1. Semantic ambiguity processing in sentence context: Evidence from event-related
Tranel, D. (2009). The left temporal pole is important for retrieving words for unique fMRI. NeuroImage, 34(3), 1270–1279.
concrete entities. Aphasiology, 23(7–8), 867–884. Ziegler, J., Bencini, G., Goldberg, A., & Snedeker, J. (2019). How abstract is syntax?
Traxler, M. J., Morris, R. K., & Seely, R. E. (2002). Processing subject and object relative Evidence from structural priming. Cognition, 193, 104045.
clauses: Evidence from eye movements. Journal of Memory and Language, 47(1), Ziegler, J., & Snedeker, J. (2018). How broad are thematic roles? Evidence from struc-
69–90. tural priming. Cognition, 179, 221–240.
Trueswell, J. C., Tanenhaus, M. K., & Garnsey, S. M. (1994). Semantic influences on Ziegler, J., Snedeker, J., & Wittenberg, E. (2018). Event structures drive semantic struc-
parsing: Use of thematic role information in syntactic ambiguity resolution. Journal of tural priming, not thematic roles: Evidence from idioms and light verbs. Cognitive
Memory and Language, 33(3), 285–318. Science..
Tsao, D. Y., Moeller, S., & Freiwald, W. A. (2008). Comparing face patch systems in

24