RESEARCH ARTICLE
Conceptual Combination in the LATL With and
Without Syntactic Composition
Keine offenen Zugänge
Tagebuch
Alicia Parrish1
and Liina Pylkkänen1,2,3
1Department of Linguistics, New York University, New York, USA
2Abteilung für Psychologie, New York University, New York, USA
3NYUAD Institute, New York University Abu Dhabi, Abu Dhabi, UAE
Schlüsselwörter: conceptual combination, syntax, semantics, magnetoencephalography
ABSTRAKT
The relationship among syntactic, semantic, and conceptual processes in language
comprehension is a central question to the neurobiology of language. Several studies have
suggested that conceptual combination in particular can be localized to the left anterior
temporal lobe (LATL), while syntactic processes are more often associated with the posterior
temporal lobe or inferior frontal gyrus. Jedoch, LATL activity can also correlate with
syntactic computations, particularly in narrative comprehension. Here we investigated the
degree to which LATL conceptual combination is dependent on syntax, specifically asking
whether rapid (~200 ms) magnetoencephalography effects of conceptual combination in the
LATL can occur in the absence of licit syntactic phrase closure and in the absence of a
semantically plausible output for the composition. We find that such effects do occur: LATL
effects of conceptual combination were observed even when there was no syntactic phrase
closure or plausible meaning. But syntactic closure did have an additive effect such that LATL
signals were the highest for expressions that composed both conceptually and syntactically.
Our findings conform to an account in which LATL conceptual composition is influenced by
local syntactic composition but is also able to operate without it.
EINFÜHRUNG
Language comprehension and production require the combination of smaller units of words
and morphemes into larger phrases. The last several decades of research on language process-
ing has yielded many candidates for hubs of basic combinatorial processes. The combinatory
role of the left anterior temporal lobe (LATL) has been characterized in a series of studies and
appears to be conceptual in nature (z.B., Baron & Osherson, 2011; Baron et al., 2010; Blanco-
Elorrieta & Pylkkänen, 2016; Kim & Pylkkänen, 2019; Poortman & Pylkkänen, 2016;
Westerlund & Pylkkänen, 2014; Zhang & Pylkkänen, 2015, 2018A, 2018B), a finding also sup-
ported by studies of language deficits (z.B., Lukic et al., 2021; Mesulam et al., 2015, 2019;
Wilson et al., 2014). At the same time though, the LATL also correlates with syntactic process-
ing steps during narrative comprehension (Brennan et al., 2016; Humphries et al., 2005),
though the extent to which these results may reflect semantic processing is unclear.
While these studies have concluded that the composition computed in the LATL is not itself
syntactic, it leaves open the question of to what degree the computation interacts with or is
Zitat: Parrish, A., & Pylkkänen, L.
(2022). Conceptual combination in the
LATL with and without syntactic
Komposition. Neurobiology of
Language, 3(1), 46–66. https://doi.org
/10.1162/nol_a_00048
DOI:
https://doi.org/10.1162/nol_a_00048
zusätzliche Informationen:
https://doi.org/10.1162/nol_a_00048
Erhalten: 16 Marsch 2021
Akzeptiert: 15 Juni 2021
Konkurrierende Interessen: Die Autoren haben
erklärte, dass keine konkurrierenden Interessen bestehen
existieren.
Korrespondierender Autor:
Alicia Parrish
alicia.v.parrish@nyu.edu
Handling-Editor:
Yanchao Bi
Urheberrechte ©: © 2021
Massachusetts Institute of Technology
Veröffentlicht unter Creative Commons
Namensnennung 4.0 International
(CC BY 4.0) Lizenz
Die MIT-Presse
l
D
Ö
w
N
Ö
A
D
e
D
F
R
Ö
M
H
T
T
P
:
/
/
D
ich
R
e
C
T
.
M
ich
T
.
e
D
u
N
Ö
/
l
/
l
A
R
T
ich
C
e
–
P
D
F
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
N
Ö
_
A
_
0
0
0
4
8
P
D
.
l
F
B
j
G
u
e
S
T
T
Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3
Conceptual combination with and without syntax
Syntactic composition:
The process of combining syntactic
categories to create a larger syntactic
Struktur.
Conceptual combination:
The process of combining two or
more conceptual representations to
create a more complex conceptual
representation.
dependent on syntactic processes associated with composition and phrase closure, Und
whether other well-studied factors such as semantic plausibility affect the LATL computation.
This study explores both of these questions by systematically varying the type and validity of
syntactic composition and semantic plausibility and investigating whether disruptions to these
cause corresponding disruptions to conceptual combination, as reflected in LATL activity. Wir
will use the term LATL conceptual combination to refer to the computations localized in the
LATL that appear to contribute to conceptual combination. We do not mean to imply that
there may not be other neural reflexes of conceptual combination or additional computational
routines that achieve it (sehen, z.B., Boylan et al., 2015, 2017). We simply suggest that the LATL
is a component of the process of combination. Ähnlich, we do not suggest that the LATL’s
language function is only combinatory, given evidence for its role in single-word semantic
Erinnerung (Binney et al., 2010; Lambon Ralph et al., 2009; Westerlund & Pylkkänen, 2014)
and as a more general semantic hub (Lambon Ralph et al., 2017).
In contrast to the LATL’s role in conceptual, meaning-based composition, a broader range
of hypotheses remain open for the left inferior frontal gyrus (LIFG) and left posterior temporal
lobe (LPTL), often cited as loci of syntactic processing (Friederici, 2011; Grodzinsky & Santi,
2008; Hagoort, 2005, 2017; Matchin & Hickok, 2020; Pallier et al., 2011). Certainly, all three
regions are routinely implicated in normal language processing (Fedorenko & Thompson-
Schill, 2014) and may even be recruited in composition in the absence of licit syntax
(Mollica et al., 2020). Jedoch, the relationship between these processes and conceptual
composition is an open question, as it is not clear whether conceptual combination acts on
syntactically licit phrases during sentence processing, or if it functions fully independently of
syntactic input. Ähnlich, the role of the ventromedial prefrontal cortex (vmPFC), which has
been associated with more semantic manipulations (Pylkkänen et al., 2011), in conceptual
composition remains unclear. VmPFC activation sometimes patterns with the LATL in showing
greater activation for phrases compared to single words or lists and sometimes fails to show
such an effect.
Thus this study fills in a gap in the existing literature, by asking, in a highly controlled way,
to what degree conceptual combination effects are modulated by syntactic and semantic effects,
and to what degree the output of conceptual combination similarly affects other processes. Unser
main focus was to test whether combinatory effects in the LATL can be obtained independent of
syntax and, umgekehrt, whether there are other regions not sensitive to manipulations of con-
ceptual combination that are still modulated by syntactic differences.
We first address the question of whether conceptual combination relies on syntactic con-
stituency by creating minimal pairs of phrases that have conceptual structures that are as sim-
ilar as possible but vary in whether syntactic composition is possible. Perfect minimal pairs of
this type are not possible in English, as changes to the structure very often cause a change in
Bedeutung. Jedoch, within the realm of modifiers, we have the ability to create near minimal
pairs of this type, as the addition of the morpheme -ly can turn an adjective into an adverb, Aber
keeps the conceptual content of the word relatively unchanged (z.B., pleasant and pleasantly
have similar conceptual content). We exploit this property of modifiers to create near minimal
pairs for our adjective manipulation that share conceptual content but differ in whether syn-
tactic composition is possible by using an adjective as a target word, comparing activation on
an adjective like sunny following either pleasantly (forming pleasantly sunny, where there is
local syntactic composition) or following pleasant (forming pleasant, sunny, where this is no
syntactic composition between the two modifiers). In both cases, the phrase is followed by a
noun (z.B., day) to form a reasonable English phrase. If conceptual composition relies on syn-
tactic composition to some degree, as would be the case if LATL composition operates on the
Neurobiology of Language
47
l
D
Ö
w
N
Ö
A
D
e
D
F
R
Ö
M
H
T
T
P
:
/
/
D
ich
R
e
C
T
.
M
ich
T
.
e
D
u
N
Ö
/
l
/
l
A
R
T
ich
C
e
–
P
D
F
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
N
Ö
_
A
_
0
0
0
4
8
P
D
.
l
F
B
j
G
u
e
S
T
T
Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3
Conceptual combination with and without syntax
outputs of a licit syntactic parse, then we expect to see greater activation on sunny in the
phrase pleasantly sunny compared to pleasant sunny and no difference between pleasant sunny
and a non-combinatory control (Figure 1A or 1B). Jedoch, if the process of conceptual com-
bination is separate from or does not rely on syntactic parsing steps, then we expect both
pleasantly sunny and pleasant sunny to show similarly greater LATL activation compared to
a non-combinatory control (Figure 1D).
As the majority of past studies looking at LATL conceptual combination has focused on
adjectival modification of nouns (z.B., Bemis & Pylkkänen, 2011; Blanco-Elorrieta et al.,
2018; Poortman & Pylkkänen, 2016; Zhang & Pylkkänen, 2018A), we also introduced a
noun manipulation into the experiment. Rather than using morphological cues to manipulate
whether direct syntactic composition occurs, we created stimuli where syntactic composi-
tion between two words is not impossible, but rather closure of the phrase as a whole is
blocked due to temporary ungrammaticality. We accomplished this by inserting a modifier
between a determiner and the noun, comparing the composition effect on tea between this
herbal tea (where full phrase closure is possible) to these herbal tea (where full phrase clo-
sure would lead to an ungrammatical parse), exploiting the property of English demonstra-
tives that they must agree in number with the head noun. Each phrase was followed by a
deverbal noun with number agreement to the determiner that caused the entire phrase to be
syntactically licit (in this example, drinker(S) followed tea). Like with the adjective manipula-
tion, if there is some modulation in LATL conceptual combination that comes from having a
complete syntactic phrase (rather than just local syntactic composition), then we would ex-
pect to see greater activation in this region for cases like this herbal tea compared to these
herbal tea. If the stronger possibility is true, that LATL conceptual combination only operates
on the output of a syntactic composition mechanism, then we would further expect that this
herbal tea would show greater activation than these herbal tea, relative to a non-combinatory
Kontrolle (Figure 1A and 1B).
l
D
Ö
w
N
Ö
A
D
e
D
F
R
Ö
M
H
T
T
P
:
/
/
D
ich
R
e
C
T
.
M
ich
T
.
e
D
u
N
Ö
/
l
/
l
A
R
T
ich
C
e
–
P
D
F
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
N
Ö
_
A
_
0
0
0
4
8
P
D
.
l
F
B
j
G
u
e
S
T
T
Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3
Figur 1. Diagram of predicted results from the four primary hypotheses. Every hypothesis predicts a difference between the
non-combinatory control and the plausible composition condition.
Neurobiology of Language
48
Conceptual combination with and without syntax
Plausibility:
The degree to which the meaning of
an expression describes a situation or
circumstance that could happen in
the real world.
In order to address the relationship between semantic plausibility and conceptual combi-
nation, we also varied whether the first modifier is likely to co-occur with or modify the target
word. For the adjective manipulation, where sunny was the target word, we used the modifier
sharp/sharply, which tends to only modify concrete rather than abstract nouns, to create the
semantic implausibility (the exact cause of the implausibility varies, but was verified by a
norming study described below). In the noun manipulation, we relied on animacy violations
as the source of semantic implausibility. Using the above example, we compared activation
on tea with a semantically plausible modifier (herbal ) to activation on tea with an implausible
modifier (impolite). In the implausible condition, it was always the case that the modifier
could licitly compose with the following deverbal noun (drinker), leading to a globally plau-
sible sentence.
If LATL conceptual combination relies on a meaningfully licit parse being available for the
Phrase, then we would expect to see greater activation in regions previously identified as par-
ticipating in combinatory processes for the plausible condition, with the implausible condition
patterning with the non-combinatory control (compatible with Figure 1A). Notiz, Jedoch, Das
a “meaningfully licit parse” requires composition to evaluate it (though we remain agnostic
about whether it requires conceptual combination or implicates the LATL). In der Tat, unter
the interpretation that one can only know whether a multiword phrase is meaningful after having
actually combined the words in order to assess the meaning (Pylkkänen et al., 2009), then any
effect of the plausibility manipulation on conceptual combination would be quite unexpected.
Jedoch, if conceptual combination relies on the availability of semantically compatible fea-
tures, then we expect that pleasant(ly) and herbal, which have more compatible features with
sunny and tea respectively compared to the implausible sharp(ly) and impolite, will elicit greater
activation in the LATL. Jedoch, it is also possible that the implausibility leads to a richer
semantic representation of the resulting phrase, as the concept for something that is both sharp
and sunny or a tea that is impolite requires a large number of otherwise unrelated (and likely
nicht überlappend) semantic features. If this is the case, then it is possible that the implausible
condition will show greater activation than the plausible condition in the LATL. It is also pos-
sible that both semantic plausibility and syntactic constituency can drive LATL activation
while not directly modulating the effect of conceptual combination. If these two factors can
independently drive LATL activation, then we would expect to see an additive effect on acti-
vation patterns, with the highest activation for the most syntactically and semantically well-
formed condition, and the lowest activation still for the non-combinatory control, with the three
other test conditions falling somewhere in the middle (Abbildung 1C). The noun manipulation
additionally controlled for potential lexical probability-related effects in this design. We cre-
ated two sets of stimuli that differ based on the transitional probability (TP) of the adjective
and noun. This accomplished two things: Erste, it ensured that we can assess whether there is
an effect of the syntactic manipulation independent of the likelihood of the local combination,
und zweitens, it created a condition in which participants were biased toward preferentially
interpreting the adjective noun combination as the object of the deverbal noun that follows
(as in herbal tea drinker, which is interpreted as “one who drinks [herbal tea]”).
In creating a control condition to compare to the test conditions, we needed to ensure that
the target word’s context was truly non-combinatory. This is a difficult condition to create, als
conceptual combination is a basic component of language processing. Als solche, ensuring that
the context of the critical word in the control condition is non-combinatory necessitates pre-
senting the target word in a context where it does not have any lexical material with which to
compose. Though several minimal composition studies have used a string of consonants (z.B.,
“xgmp”) presented prior to the target word to ensure that there is no lexical content with which
Neurobiology of Language
49
l
D
Ö
w
N
Ö
A
D
e
D
F
R
Ö
M
H
T
T
P
:
/
/
D
ich
R
e
C
T
.
M
ich
T
.
e
D
u
N
Ö
/
l
/
l
A
R
T
ich
C
e
–
P
D
F
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
N
Ö
_
A
_
0
0
0
4
8
P
D
.
l
F
B
j
G
u
e
S
T
T
Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3
Conceptual combination with and without syntax
to compose (z.B., Bemis & Pylkkänen, 2013), we found the paradigm with consonant strings to
be highly unnatural when used in full sentences. Ähnlich, a random or unrelated word presented
in this context would be both an unnatural and a potentially elicit attempt at composition as the
participant tries to make sense of the entire sentence. daher, we decided to use the equivalent
of single-word controls following other studies of simple two-word composition (Bemis &
Pylkkänen, 2011, et seq.). In our non-combinatory control condition, the control sentences begin
with the unmodified target word, sunny or tea. This condition allows us to assess whether each of
the test conditions independently shows an effect of conceptual combination, as all of the test
conditions have lexical material prior to the target word that could potentially trigger conceptual
combination, and the primary question is whether each of these test conditions shows an in-
crease in activation relative to the non-combinatory control. We find it unlikely that the presence
or absence of lexical material in the pre-target word region represents a confound with respect
to interpreting the activation patterns in the LATL as effects related to conceptual combination.
Bemis and Pylkkänen (2011), Zum Beispiel, included both single-word controls (like in our
Studie) and an additional control experiment that compared combinatory phrases in a compo-
sition task to two-word sequences in a list condition. Regardless of whether the control for the
combinatory condition was a single-word control or a two-word list, the combinatory condition
still showed greater activation in the LATL around 200 ms after target-word onset, indicating
that prior lexical material alone is not sufficient to drive LATL conceptual combination.
In the noun manipulation, we additionally included numeral controls (one tea, two tea)
based on findings in past production studies where numeral modification did not lead to an
increase in LATL activity compared to a non-combinatory control (Blanco-Elorrieta &
Pylkkänen, 2016; Del Prato & Pylkkänen, 2014). The purpose of this additional control was
to account for the effects of a syntactic mismatch alone, without any possibility of conceptual
combination related effects. Each set of stimuli in the noun condition was presented with these
two additional control conditions. Jedoch, this study is the first time that a numeral condition is
being tested in comprehension rather than production. In our study, the numeral controls did
show effects consistent with conceptual combination compared to the non-combinatory control
described above, and thus it was not a suitable control for analysis, so we did not include these
comparisons in the main results, though we return to this issue in the discussion section.
MATERIALS AND METHODS
Teilnehmer
Twenty-four right-handed native speakers of North American English participated in this study.
All participants were neurotypical, had normal or corrected-to-normal vision, normal hearing,
reported being native speakers of English, and were naive to the purpose of the study. An ac-
curacy threshold of 70% on the globally plausible items in the behavioral task was set to ensure
that all participants were paying attention. Based on this threshold, no subjects were excluded
from analysis (the lowest participant scored 74.85% accuracy, but all others were well above
80%, and mean accuracy was above 90%). One subject was excluded for average dynamic
statistical parametric mapping (Dale et al., 2000) activation far exceeding the average (d.h.,
by more than a factor of three). All analyses were done on the remaining 23 Teilnehmer.
Materials
In order to test the independent effects of syntactic composition on processes of conceptual
combination, we designed stimuli where syntactic composition is blocked, but conceptual
combination is still a possibility. In the adjective manipulation, we tested this with a condition
Neurobiology of Language
50
l
D
Ö
w
N
Ö
A
D
e
D
F
R
Ö
M
H
T
T
P
:
/
/
D
ich
R
e
C
T
.
M
ich
T
.
e
D
u
N
Ö
/
l
/
l
A
R
T
ich
C
e
–
P
D
F
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
N
Ö
_
A
_
0
0
0
4
8
P
D
.
l
F
B
j
G
u
e
S
T
T
Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3
Conceptual combination with and without syntax
that contrasts adverb-adjective sequences (z.B., pleasantly sunny (day), where there is local
combination between pleasantly and sunny) with adjective-adjective sequences (z.B., pleasant
sunny (day), where there is no local combination between the two adjectives; they both modify
the later day). In the noun manipulation, we tested this by blocking composition (d.h., merge) of a
higher phrasal node through a lack of number agreement between a determiner and noun.
Given a potentially combinatory phrase herbal tea, preceding that phrase with a number-
mismatching determiner will block licit closure of the higher phrasal node (*these herbal tea,
compared to the licit closure with this herbal tea).
It was always the case that the sentences were grammatically correct following presentation
of the post-target word. For the noun + syntactic non-composition items, this meant that the
agreement with the plural demonstrative (these or those) occurred on the word immediately
following the target word. For the sake of consistency and to allow us to analyze effects on the
post-target word, it was always the case that there was a deverbal noun following the noun
target words, and we ensured that participants always preferentially interpreted the noun target
word as an argument, d.h., a tea drinker is a person who drinks tea. Note that the syntactic
composition examples are essentially garden path phrases, and that the parser must reanalyze
the assumed structure in these cases once the next word is encountered. It was always the case
that the sentences were grammatically correct following presentation of the post-target word,
but for the syntactic non-composition condition, the phrase could not be grammatically com-
pleted at the target word.
Figure 2A shows the assumed syntactic structures of each of these conditions. We addition-
ally manipulated the semantic plausibility of each combination, yielding the full 2 × 2 × 2
paradigm of word category (adjective, noun) × semantic plausibility (plausible, implausible)
× syntactic composition (Komposition, non-composition). The implausible condition addition-
ally represents a meaning that is incomplete or needs to be reanalyzed after the target word, bei
which point phrases in the noun manipulation were globally plausible.
Stimuli Norming
Prior to the magnetoencephalography (MEG) Studie, we verified (ich) that the experimental stimuli
represented a real contrast in plausibility between the plausible and implausible conditions,
Und (ii) that the syntactic manipulation in the adjective stimuli was not confounded by a dif-
ference in acceptability. We manually constructed 50 potential stimulus sets as three-word
phrases for the noun manipulation with high transitional probability (z.B., herbal/impolite
tea drinker), the noun manipulation with low transitional probability (z.B., raw/engaged
produce buyer), and the adjective manipulation ( pleasant(ly)/sharp(ly) sunny day). Wir
then created the following four phrase types from each stimulus set to norm separately: (ich)
the full three-word phrase (z.B., herbal tea drinker), (ii) just the pre-target word and target
word (z.B., herbal tea), (iii) just the target word and post-target word (z.B., tea drinker), Und
(iv) just the pre-target word and post target word (z.B., herbal drinker). We intentionally
excluded some phrases from within the adjective manipulation that would be uninformative
because the resulting phrase was grammatically illicit (z.B., pleasantly day) or failed to result
in a complete phrase (z.B., pleasant sunny). This resulted in a total of 1,150 unique phrases
to rate (50 noun low-TP × 7 phrase types + 50 noun high-TP × 7 phrase types + 50 adjective ×
9 phrase types).
Via Amazon Mechanical Turk (MTurk; https://www.mturk.com), we conducted an accept-
ability judgment task in which participants rated the two- or three-word phrases on a 7-point
scale, mit 1 defined as “very unnatural” and 7 defined as “very natural.” Each MTurk task
Neurobiology of Language
51
l
D
Ö
w
N
Ö
A
D
e
D
F
R
Ö
M
H
T
T
P
:
/
/
D
ich
R
e
C
T
.
M
ich
T
.
e
D
u
N
Ö
/
l
/
l
A
R
T
ich
C
e
–
P
D
F
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
N
Ö
_
A
_
0
0
0
4
8
P
D
.
l
F
B
j
G
u
e
S
T
T
Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3
Conceptual combination with and without syntax
l
D
Ö
w
N
Ö
A
D
e
D
F
R
Ö
M
H
T
T
P
:
/
/
D
ich
R
e
C
T
.
M
ich
T
.
e
D
u
N
Ö
/
l
/
l
A
R
T
ich
C
e
–
P
D
F
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
N
Ö
_
A
_
0
0
0
4
8
P
D
.
l
F
B
j
G
u
e
S
T
T
Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3
Figur 2.
(A) Und (B) show example stimuli and schematic syntactic structures used in the noun (A) and adjective (B) Manipulationen. Target
words are boxed in each. Green lines indicate that the composition between the words is semantically plausible (z.B., pleasantly sunny), and red
lines indicate that it is implausible given world knowledge (z.B., sharply sunny). Dotted lines indicate that the parser could project upcoming
structure from the grammatical cues available. In the non-combinatory control condition, the target word is the initial word in the structure, Also
there is nothing for it to syntactically compose (or not compose) mit, and there is nothing with which it can form a plausible or implausible
Bedeutung. (C) shows the timing and structure of trials in the two-word category manipulations and the control condition. Trials are arranged such
that the 0 time point is the onset of the target word. The adjective manipulation shows an example of the picture verification task, and the other two
examples show trials in which there was no task. In trials with a task, participants saw the “ready for next” screen after their response to the picture.
contained 30 Zu 41 items to rate, of which 5 oder 8 were attention-check items with acceptability
that was either very high (z.B., large dog) or very low (z.B., cement laughter). All items were
presented on the same page, and all items within a task were either three-word phrases or
two-word phrases. No two items derived from the same stimulus set were displayed within
Neurobiology of Language
52
Conceptual combination with and without syntax
the same task (d.h., herbal tea and impolite tea could never appear on the same page together). Item
presentation order was randomized using a Latin square design; the randomization and HTML
page formatting were done using TurkTools (Erlewine & Kotek, 2016). We provide screenshots
from the task instructions and the rating task in Appendix D. (Appendices for this article are
located in the online Supporting Information at https://doi.org/10.1162/nol_a_00048.)
A total of 990 unique MTurk workers completed at least one online task. We used the
attention-check items to identify and exclude participants who either did not understand the task
or were not completing it felicitously. Accuracy criteria on the attention-check items were set as
a rating of 1 oder 2 on the “unacceptable” items (z.B., cement laughter) and a rating of 6 oder 7 on the
“acceptable” items (z.B., large dog). We applied an accuracy cutoff for each worker of 75%,
which led to us excluding 85 workers (8.59%) for low accuracy. We additionally excluded
30 workers (3.03%) who indicated that they were not native speakers of English, leaving a total
von 875 workers whose data were used in the study. Each item was ultimately rated by 13–18
unique MTurk workers. We also assessed inter-rater reliability with Krippendorff’s alpha
(Krippendorff, 2011) to assess the degree to which raters were using the Likert scale in a similar
way and found an overall agreement of (cid:1) = 0.733. Within the plausible and implausible items,
inter-rater reliability was at (cid:1) = 0.723 Und (cid:1) = 0.718, jeweils, indicating a reasonable
degree of agreement among raters.
Of the 150 initial stimulus sets, we identified the best 35 sets within each category (noun high
TP, noun low TP, and adjective). In determining which sets to include, we applied the following
Kriterien: (ich) in the noun condition, 3-word phrases had a mean rating greater than 4.5, (ii) im
adjective condition, 3-word phrases had a mean rating greater than 4 for the plausible items and
lower than 4 for the implausible items, (iii) in both the noun and adjective conditions, items of the
pre-target word plus the target word had mean ratings greater than 4.5 in the plausible condition
and less than 3.5 in the implausible condition. This selection criteria left us with 112 viable sets
(37 in noun high TP, 40 in noun low TP, Und 35 in adjective), which was further reduced to the
desired set size by excluding sets in which the difference between plausible and implausible
items within a set was the smallest. We present the results of the norming study on the final
105 total stimulus sets that we used in the MEG study in Table 1. Most crucially, the results in
Tisch 1 demonstrate that all stimuli used in the experiment showed a contrast in acceptability
between the two-word phrases that we had labelled as plausible (z.B., pleasantly sunny, herbal
tea) and those that we had labelled implausible (z.B., sharply sunny, impolite tea).
Experiment Design
The full experimental design is a 2 × 2 × 2 design of word category (noun, adjective) × syn-
tactic composition (Komposition, non-composition) × plausibility (plausible, implausible), mit
the noun manipulation additionally including a transitional probability factor (hoch, niedrig). Das
design included one type of control item (non-combinatory) in the adjective manipulation and
three types of control items (non-combinatory, numeral & syntactic composition, numeral &
syntactic non-composition) in the noun manipulation. As the numeral condition turned out to
not be a suitable control, we exclude this manipulation from the main results. The experiment
included three groupings of 35 stimulus sets, with each set created around a single target word,
and with no lexical items serving as the target word in more than one stimulus set. In the two
noun manipulations (high TP, low TP), each of these sets contained seven sentences: the four
test conditions, plus the three control conditions. In the one adjective manipulation, each set
contained five sentences: the four test conditions, plus the one control condition. The entire
experiment thus included a total of 665 sentences presented to each participant. A full list of
Neurobiology of Language
53
l
D
Ö
w
N
Ö
A
D
e
D
F
R
Ö
M
H
T
T
P
:
/
/
D
ich
R
e
C
T
.
M
ich
T
.
e
D
u
N
Ö
/
l
/
l
A
R
T
ich
C
e
–
P
D
F
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
N
Ö
_
A
_
0
0
0
4
8
P
D
.
l
F
B
j
G
u
e
S
T
T
Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3
Trans.
prob.
–
–
–
–
Low
–
Word
category
e
v
ich
T
C
e
J
D
A
N
u
Ö
N
Tisch 1.
Experiment 1 results of the MTurk norming study
Plausibility
Plaus.
Syntactic
Komposition
Compose
Pre-Target
wordv
pleasantly
Target
word
sunny
Post-Target
word
Tage
Pre-Target +
Target
5.871
Target +
Post-Target
Pre-Target +
Post-Target
–
Full 3-word
Phrase
6.060
Non-comp.
pleasant
sunny
Imp.
Compose
sharply
Non-comp.
sharp
sunny
sunny
Plaus.
Imp.
Hoch
Plaus.
–
Imp.
–
–
–
–
raw
produce
engaged
produce
herbal
impolite
tea
tea
Tage
Tage
Tage
buyer
buyer
drinker
drinker
–
2.414
–
6.594
2.019
6.833
2.014
6.782
6.465
6.129
6.059
–
3.429
3.923
4.932
3.836
4.866
6.067
2.746
2.887
5.993
5.824
6.057
5.826
Notiz. Numbers represent the average raw responses from participants across all items within a given condition, as judged on a 7-point scale. The color indicates the expected acceptability
rating based on the item’s label, with red being stimuli that were expected to be implausible and thus receive a low acceptability rating, and green being stimuli that were expected to be
plausible and thus receive a high acceptability rating. The words under Pre-Target Word, Target Word, and Post-Target Word are examples to illustrate each experimental condition. Trans.
prob = Transitional probability; Syntactic comp. = syntactic composition; Plaus. = Plausible; Imp. = Implausible; Non-comp. = Non-composition.
C
Ö
N
C
e
P
T
u
A
l
C
Ö
M
B
ich
N
A
T
ich
Ö
N
w
ich
T
H
A
N
D
w
ich
T
H
Ö
u
T
S
j
N
T
A
X
l
D
Ö
w
N
Ö
A
D
e
D
F
R
Ö
M
H
T
T
P
:
/
/
D
ich
R
e
C
T
.
M
ich
T
.
e
D
u
N
Ö
/
l
/
l
A
R
T
ich
C
e
–
P
D
F
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
N
Ö
_
A
_
0
0
0
4
8
P
D
.
l
F
B
j
G
u
e
S
T
T
Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3
N
e
u
R
Ö
B
Ö
Ö
G
j
l
ich
Ö
F
L
A
N
G
u
A
G
e
5
4
Conceptual combination with and without syntax
test stimuli, along with the carrier sentences they appeared in, can be found in Appendices A
and B.
These stimuli were split into 12 blocks of 54–56 trials for the experiment, with each trial
consisting of a single sentence presented via rapid serial visual presentation. In assigning trials
to blocks, trials were pseudo-randomized so that no two sentences (experimental or control)
from the same stimulus set appeared in the same block. Das ist, none of the five possible sen-
tences that used the target word sunny ever appeared twice in the same block. All blocks con-
tained at least two sentences from each of the eight main test conditions (word category ×
semantic plausibility × syntactic composition) plus at least two controls from each word cate-
gory manipulation. Trial order within blocks was fully randomized for each participant, und das
order in which the blocks were presented was also fully randomized for each participant.
Verfahren
Following the guidelines set out by the Institutional Review Board of New York University, alle
participants gave informed, written consent before beginning the study. This experiment mea-
sured MEG responses during sentence reading and used a picture verification task. Stimuli
were presented visually, one word at a time, via PsychoPy (Peirce, 2007). Each trial began
with a fixation cross at the center of the screen. This cross lasted for 500 ms and was followed
by a 150 ms blank screen before the first word of the stimulus sentence appeared. Words were
presented with a 450 ms stimulus onset asynchrony (150 ms inter-stimulus interval). Sentences
were 5–9 words in length. Für 28% of the trials, the trial included an image presented after the
Satz, and participants indicated with their left hand if it was a “match” or “not a match”
with the sentence they just read. Each experimental block contained 14–16 task trials (out of
54–56 trials in the block), with approximately equal numbers of match and not-a-match task
trials within the block. None of the images included a recognizable person (z.B., a celebrity),
and none of the images contained words. As a general heuristic, participants were told to imag-
ine whether the sentence would be a possible caption in a newspaper or online article for the
picture in making their judgment. The full trial structure used for this experiment is shown in
Figure 2C with an example of a trial that had a task following it. For trials where there was no
Aufgabe, the final word in the sentence was immediately followed by a prompt asking the partic-
ipant if they were ready for the next word. The experiment lasted approximately 1 hr for each
participant to complete, einschließlich 11 breaks between blocks that were each spaced approx-
imately 5 min apart. The duration of the breaks was left up to the participants to decide.
Data Acquisition and Processing
Participants’ headshapes were digitized using Polhemus FastSCAN system (Polhemus Inc.,
Colchester, USA). This headshape data, along with fiducial landmarks, was co-registered to
the average brain available in FreeSurfer (Dale et al., 1999). Continuous MEG data were re-
corded using a 208-channel axial gradiometer whole-head system using a 1000 Hz sampling
rate. All preprocessing was done using eelbrain version 0.27 (Brodbeck, 2018) and MNE ver-
sion 0.16 (Gramfort et al., 2014). The MEG data was recorded with a 200 Hz low pass filter,
and then filtered offline with a 0–40 Hz bandpass filter. Artifact rejection was completed with
independent component analysis (ICA) for each participant to remove artifacts due to environ-
mental noise, eyeblinks, and heartbeat. Epochs were created starting from 100 ms before the
target word and extending to 1,000 ms after the target word in order to include both the target
and post-target word in each epoch. Epoch rejection was done via an absolute threshold to
eliminate epochs containing amplitudes exceeding 3,000 fT, as these were assumed to contain
Neurobiology of Language
55
l
D
Ö
w
N
Ö
A
D
e
D
F
R
Ö
M
H
T
T
P
:
/
/
D
ich
R
e
C
T
.
M
ich
T
.
e
D
u
N
Ö
/
l
/
l
A
R
T
ich
C
e
–
P
D
F
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
N
Ö
_
A
_
0
0
0
4
8
P
D
.
l
F
B
j
G
u
e
S
T
T
Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3
Conceptual combination with and without syntax
artifacts not removed with ICA. In Summe, 6.75% of all epochs were rejected, with no individual
participant having greater than 30% of their trials rejected.
As the target word occurred in the middle of the sentence and in a different sentence po-
sition across some conditions, we de-meaned the entire epoch from 100 ms prior to the onset
of the target word to 100 ms after the offset of the post-target word (for a total epoch size of
1,100 MS) rather than apply baseline correction before the target word. This allowed us to
directly compare the experimental and control trials in our analysis. Channel noise covariance
was estimated from 3 min of empty-room recording taken immediately prior to each partici-
pant’s recording session.
Region of Interest Selection
We conducted analyses in four regions of interest (ROIs) selected based on prior studies show-
ing these regions’ sensitivity to syntactic, semantic, and combinatorial manipulations. Diese
ROIs were the anterior temporal lobe (ATL), posterior temporal lobe (PTL), inferior frontal gyrus
(IFG), and vmPFC, and we analyzed both the left and right hemisphere homologues for each.
For the ATL, PTL, and IFG, we based the localization of these regions on Pallier et al.’s (2011)
Studie, taking the MNI coordinates that they identified and labelling the 30 mm regions around
those points (for the ATL, we used just the temporal pole coordinates). For the vmPFC ROI, Wir
combined the medial and lateral orbitofrontal labels from the aparc parcellation (Desikan
et al., 2006). These ROIs and their exact coordinates are summarized in the table in
Appendix C.
MEG Analysis
All planned analyses were conducted using temporal cluster permutation tests (Maris &
Oostenveld, 2007) within each predetermined ROI and time window or spatiotemporal cluster
permutation tests across the whole brain. For assessing combinatory effects, we conducted a
2 × 5 ANOVA of word category × condition, and we analyzed the window that spanned
from 100–300 ms after presentation of the target word and post-target word, following the
findings of previous studies that identified the peak of combinatorial activity to begin around
200 ms post-stimulus onset in visually presented comprehension tasks (Bemis & Pylkkänen,
2012). We expanded this window to 100–450 ms for 2 × 2 × 2 comparisons of word category ×
semantic plausibility × syntactic composition (no control items) to take into account additional
syntactic or plausibility-related effects that extend beyond 300 MS.
For the temporal cluster permutation test, we averaged over the sources in each ROI at each
time point, and those time courses were used as input to the cluster-based permutation test. Wir
computed an F statistic for the ANOVAs, and used cluster thresholds set to a minimum temporal
window of at least 25 ms and formed from statistics corresponding to a p value of less than 0.05.
Each test used 10,000 permutations. To correct for multiple comparisons across eight different
ROIs, we assessed significance with the Bonferroni corrected p value (Rice, 1989) to take into
account multiple tests against the same null hypothesis (Cabin & Mitchell, 2000).
As our ROIs were selected based on composition studies, and our stimuli included both plau-
sibility and syntactic manipulations that may not be fully captured in these ROIs, we conducted
exploratory spatiotemporal cluster permutation tests for the 2 × 2 × 2 comparison of word cat-
egory × semantic plausibility × syntactic composition and for the 2 × 2 comparison of transitional
probability × syntactic composition. In each case, we computed an F statistic for these
Neurobiology of Language
56
l
D
Ö
w
N
Ö
A
D
e
D
F
R
Ö
M
H
T
T
P
:
/
/
D
ich
R
e
C
T
.
M
ich
T
.
e
D
u
N
Ö
/
l
/
l
A
R
T
ich
C
e
–
P
D
F
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
N
Ö
_
A
_
0
0
0
4
8
P
D
.
l
F
B
j
G
u
e
S
T
T
Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3
Conceptual combination with and without syntax
comparisons, with minimum cluster thresholds of 25 ms in duration with at least 10 contiguous
sources corresponding to an uncorrected p value of less than 0.05, mit 10,000 permutations.
ERGEBNISSE
Effects of Conceptual Combination as Reflected by Comparison to Non-Combinatory Controls
On the target word (sunny, tea), we observed a main effect of condition ( pcorr = 0.0034) im
LATL characterized by increased activation for all four test conditions compared to the non-
combinatory control, and the highest activation for the plausible + syntactic composition con-
dition. The identified cluster spanned from 155–190 ms post-stimulus onset. None of the other
seven ROIs showed a significant main effect of condition. All four test conditions appeared to
have greater average activation in the identified time window compared to the control. Follow-
up pairwise t tests showed that this difference was significant in three of the four test conditions
(all ps < 0.005), with only the implausible + syntactic non-composition condition not passing the
significance threshold ( p = 0.064). As these were follow-up tests from a significant main effect
and were conducted within a single identified ROI, they were not further controlled for multiple
comparisons. There were additional significant differences between the plausible + syntactic
composition condition and the two syntactic non-composition conditions ( ps < 0.01). These
results are summarized in Figure 3, where we also show that this effect trends in the same direc-
tion for both the noun manipulation and the adjective manipulation (noun and adjective plots
are for visualization purposes, as there was no interaction of word category by condition, so we
did not run comparisons within each word category manipulation).
Additionally, we tested the one tea and two tea numeral controls for the noun manipulation,
but we did not observe any differences between these control sentences and the test sentences in
any of our ROIs. Nor did we observe any patterns of marginally significant differences, even
when applying a looser cluster threshold of p < 0.1.
Factorial Analysis With Plausibility and Syntactic Composition
We compared the test conditions in a 2 × 2 × 2 ANOVA of word category × syntactic com-
position × plausibility. We observed a significant main effect of syntactic composition in the
LATL, characterized by an increase for the syntactic composition condition compared to
syntactic non-composition ( pcorr = 0.0496). The temporal cluster identified extended from 150–
190 ms after onset of the target word. Left vmPFC showed a non-significant trend toward the same
effect of greater activation in the syntactic composition condition compared to the syntactic non-
composition condition, with a cluster that extended 150–190 ms after target word onset ( pcorr =
0.0768). Though the effect in LIFG was in the same direction as the LATL effect, the strongest
cluster was not significant (160–190 ms, pcorr = 0.2568). After multiple comparison correction,
there were no significant main effects of syntactic composition on the post-target word. For both
the target word and post-target word, no significant interactions between syntactic composition
and either word category or semantic plausibility were identified in any of the ROIs, and no
significant effects of syntactic composition were observed in the right hemisphere ROIs.
We observed a significant main effect of semantic plausibility in the LPTL, with an identi-
fied temporal cluster that extended 320–370 ms after onset of the target word ( pcorr = 0.012).
Similar trends were observed in LIFG ( pcorr = 0.02, 325–370 ms) and LATL ( pcorr = 0.0856,
330–360 ms). There were no main effects of semantic plausibility on the post-target word. For
both the target word and post-target word, no significant interactions between semantic plau-
sibility and either word category or syntactic closure were identified in any of the ROIs, and no
Neurobiology of Language
57
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
e
d
u
n
o
/
l
/
l
a
r
t
i
c
e
-
p
d
f
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
n
o
_
a
_
0
0
0
4
8
p
d
.
l
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Conceptual combination with and without syntax
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
e
d
u
n
o
/
l
/
l
a
r
t
i
c
e
-
p
d
f
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
n
o
_
a
_
0
0
0
4
8
p
d
.
l
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Figure 3. Main effect of condition in all LH ROIs from the 2 × 5 ANOVA crossing word category and condition. Only the LATL showed a reliable
effect, with each of the four test conditions showing greater amplitudes than the control for both adjectives and nouns (i.e., no interaction with
word category). Thus the possibility of conceptual combination appears to increase LATL activity even in the absence of local syntactic com-
position or plausibility. The corresponding RH analysis yielded no significant effects. ***p < 0.001, **p < 0.01, *p < 0.05, (cid:1)p < 0.1.
significant effects of semantic plausibility were observed in the right hemisphere ROIs. We also
observed a main effect of word category that persisted even after controlling for the greater
number of noun trials compared to adjective trials on the target word in the left vmPFC
(170–195 ms after presentation of the target word, pcorr = 0.028), with greater activation for
the adjectives (e.g., sunny) compared to the nouns (e.g., tea). There was also a significant dif-
ference on the post-target word that was characterized by greater amplitude in the noun ma-
nipulation (e.g., drinker(s)) compared to the adjective manipulation (e.g., days). This effect was
significant in the LPTL (235–280 ms after presentation of the post-target word, pcorr = 0.0064).
Neurobiology of Language
58
Conceptual combination with and without syntax
No right hemisphere ROIs showed a significant main effect of word category on either the
target word or post-target word, and there were no ROIs that showed a significant interaction
of word category by condition.
Exploratory Spatiotemporal Analyses
As the selected ROIs were not particularly well-suited to identify plausibility effects, and be-
cause the pattern of significant and trending results for both the plausibility manipulation and
the syntactic manipulation indicated that the associated activation patterns may be more dis-
tributed, we conducted an exploratory spatiotemporal test on the whole brain using the full 2 ×
2 × 2 design. In Figure 4, we show the most significant spatiotemporal clusters identified for
these factors. Figure 4A shows the significant cluster associated with the semantic plausibility
manipulation, where the implausible condition shows greater activation than the plausible
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
e
d
u
n
o
/
l
/
l
a
r
t
i
c
e
-
p
d
f
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
n
o
_
a
_
0
0
0
4
8
p
d
.
l
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Figure 4. Main effects of semantic plausibility (A) and syntactic composition (B). Implausible nouns and adjectives showed increased am-
plitudes at 300–390 ms after stimulus onset, widely distributed across the left hemisphere, but most strongly localizing in the middle temporal
lobe, consistent with previous localization of N400-type plausibility effects (Lau et al., 2008; Maess et al., 2006). The main effect of syntactic
composition shown in (B) was only a trend ( p = 0.158), but showed greater activation for the conditions where the syntax fully composes. The
identified cluster ranges from 150 to 200 ms after stimulus presentation and localizes most strongly near the vmPFC, but has a fairly broad
frontal distribution. There were no significant interactions between any of the factors (word category × semantic plausibility × syntactic com-
position) in either the target word window (sunny/tea) or the post-target word window (day/drinker). Significant main effects of word category
(not shown) were all restricted to the left hemisphere after controlling for differences in the number of trials.
Neurobiology of Language
59
Conceptual combination with and without syntax
condition. The cluster is significant in a time window from 300–390 ms after target word onset
( p = 0.017), and the cluster localizes most strongly to the left middle temporal lobe, though
the difference in activation is fairly distributed. Figure 4B shows a non-significant spatiotem-
poral cluster that was identified as the strongest effect of the syntactic manipulation ( p =
0.158). As these analyses were conducted over the whole brain rather than within pre-selected
ROIs, they were corrected within the clustering threshold and not via the post-hoc Bonferroni
correction used for the ROI analyses.
No significant clusters were observed in right hemisphere regions, and no significant inter-
actions were observed between any of the factors. On the post-target word (drinker(s), day), we
did not observe any significant effects for semantic plausibility or syntactic closure. There were
two significant clusters associated with word category. Both clusters extended for the full anal-
ysis window of 100–400 ms after the onset of the post-target word, with one cluster encom-
passing the entire left hemisphere and the other cluster encompassing most of the right
hemisphere, making interpretation of this significant result uninformative.
Effects of Transition Probability in the Noun Manipulation
We conducted two spatiotemporal cluster permutation tests with 2 × 2 factorial analyses of
transition probability × condition and transition probability × syntactic composition to assess if
conceptual combination was modulated by the local probability of the target word. We did
not find any significant main effects of transition probability or any interactions on the target
word or the post-target word. The only cluster identified was a non-significant effect on the
post-target word ( p = 0.191) that was broadly distributed across frontal and temporal regions
of the left hemisphere, but localized most strongly around the STG, showing greater activation
for the high transition probability condition (herbal tea drinker) compared to low (raw produce
buyer). No significant clusters were found in the right hemisphere.
DISCUSSION
We observed that the LATL conceptual composition effect occurred both in the absence of
syntactic composition and in the absence of a plausible semantic interpretation. This finding
suggests, first, that LATL conceptual combination is a processing step that is not dependent on
the syntactic parse. However, we cannot conclude that syntactic information is irrelevant for
the LATL or for conceptual combination more generally, as we observed an effect of the syn-
tactic manipulation in the LATL in a similar time window as the conceptual combination ef-
fect, such that syntactic composition on a given word corresponded to greater activation. In
other words, we effectively found a three-way distinction, such that there is higher activation
on the target word sunny in phrases like pleasantly sunny compared to pleasant sunny, and
both of these phrases show greater activation than the non-combinatory control sunny. While
in the adjective manipulation we could directly manipulate whether local syntactic merge oc-
curred, the difference was subtler in the noun manipulation. The three-way contrast in that
case would be that this herbal tea shows greater activation on tea compared to these herbal
tea (both of which show greater activation than the non-combinatory control tea), even though
in both cases there is a local combination that takes place, namely between herbal and tea. As
there was no interaction in the conceptual combination effects with word category, and as the
same pattern of activation was present within the noun and adjective manipulations, we con-
clude that full licit phrase closure for the DP (determiner phrase), which is possible in this
herbal tea but not these herbal tea is also a driver of whatever syntactic parsing step is respon-
sible for the observed increases in LATL activation.
Neurobiology of Language
60
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
e
d
u
n
o
/
l
/
l
a
r
t
i
c
e
-
p
d
f
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
n
o
_
a
_
0
0
0
4
8
p
d
.
l
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Conceptual combination with and without syntax
There are at least two plausible mechanistic accounts of this syntactic effect co-occurring
with LATL conceptual combination: (i) an earlier processing step that involves syntactic com-
position of constituents feeds into the LATL during a time window consistent with conceptual
combination, or (ii) syntactic feature information is composed in tandem with conceptual
composition, and the highest increase in the syntactic composition condition is an additive
effect of two separate processes. These two possibilities cannot be distinguished on the basis
of just this study, but one aspect of the design hints toward the second possibility. Our second
control condition in noun manipulation used numeral modification as an additional less con-
ceptual but combinatory condition based on the findings of a production study. However, as
we did not observe any differences between the numeral controls and the test conditions, we
consider it possible that, at least for comprehension, numerals may to some extent engage the
LATL conceptual combinatory hub. The syntactic number features do need to be checked, and
if this process implicates the LATL, then it may be that the presence of this effect is the reason
for the lack of a finding when comparing one tea to this herbal tea. It is also possible that in
comprehension, numerals are treated as conceptually composing with the nouns that they
modify, whereas production does not engage such a process. Further study is needed to dis-
entangle these options.
Based on previous studies that have found that syntactic composition drives activity in the
left PTL rather than the left ATL (Flick & Pylkkänen, 2020; Matchin & Hickok, 2020), it is per-
haps surprising that our syntactic manipulations did not show any significant effects in the left
PTL. However, the syntactic manipulation in this study was much subtler than that used by
Flick & Pylkkänen, who were contrasting predication with modification. All of our cases are
strictly modificational, and thus might not be enough of a syntactic distinction to drive mea-
surable LPTL activity. It is worth noting, however, that the time course at which we measured
syntactic modulations in the LATL is roughly aligned with the time course with which Flick &
Pylkkänen observed an effect of their syntactic manipulation in the LPTL, possibly pointing
toward these reflecting similar underlying computations.
Similarly, the lack of a significant effect of the syntactic manipulation in the LIFG may be
surprising given the long history of this region’s association with syntactic processes (e.g.,
Friederici et al., 2006). Again, a number of studies that find support for a syntactic role for
the LIFG are comparing much more distinct computations than what this study used, such
as word lists vs. full sentences (Friederici et al., 2000, i.a.; Zaccarella et al., 2017). In a more
directly comparable study, Schell et al. (2017) found an increase in LIFG activation for min-
imal phrases like this ship compared to single word controls using fMRI. The difference be-
tween their findings and ours may lie in the different types of brain activity measurable from
MEG versus fMRI, or it could be due to the use of a different experimental task, as Schell et al.
used a kind of grammaticality judgment task, which could lead to an increase in syntactic
prediction, as syntactic prediction is a beneficial strategy in grammaticality judgment tasks
and has been associated with greater activation in IFG due to increases in verbal working
memory load (Matchin et al., 2017). It is also possible that, as meaning composition was al-
ways possible in this study, combinatorial processing was always active, masking any effects
of the syntactic manipulation. In a direct comparison of lexico-semantic and syntactic com-
position, Fedorenko et al. (2020) found that no brain regions were selectively sensitive to syn-
tax over semantic processes, and they suggest that combinatorial processes may not be fully
distinguishable between syntactic and semantic/plausiblity manipulations.
We also found that although semantic plausibility alone was not a pre-condition to eliciting
LATL conceptual combination effects, semantic plausibility led to an additional increase in
LATL activation compared to the control in the same time window as conceptual combination.
Neurobiology of Language
61
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
e
d
u
n
o
/
l
/
l
a
r
t
i
c
e
-
p
d
f
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
n
o
_
a
_
0
0
0
4
8
p
d
.
l
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Conceptual combination with and without syntax
The factorial analyses, however, failed to identify semantic plausibility as a significant factor
within this time window. The effects we did observe for semantic plausibility were about 100 ms
later and replicated expected effects of this manipulation. Due to the timing of the identified
temporal cluster associated with the main effect of plausibility, and due to the localization of
the effect near the middle temporal gyrus, it is most likely that we are observing lexical access
related to M350/N400 type activity (Maess et al., 2006; Pylkkänen & Marantz, 2003). This effect
is well-studied and completely expected given our experimental design. The fact that we do not
observe that plausibility independently modulates conceptual combination is consistent with
the idea that a phrase must first be composed in order to assess if it is plausible given world
knowledge (Pylkkänen et al., 2009) and consistent with recent work by Graessner et al.
(2021), who observed a dissociation between phrasal composition and “meaningful composi-
tion” driven by a plausibility manipulation.
Neither syntactic composition nor semantic plausibility were observed to be necessary for
conceptual combination, as one of the syntactic non-composition conditions and one of the
semantically implausible conditions showed significantly greater activation than a non-
combinatory control in the LATL and in a time window consistent with previous findings
for conceptual combination. However, the implausible + syntactic non-composition condition
did not show a statistically significant difference from the non-combinatory control. This find-
ing should not be taken as evidence that this condition was itself non-combinatory, as a failure
to reach significance alone is not an argument in favor of the null hypothesis. It may be that
there was a greater degree of variance in this condition due to individual differences between
participants that led to the lack of a significant difference from the control. However, we take
the view that both syntactic constituency and semantic plausibility show additive effects in the
activation patterns in the LATL alongside conceptual combination. Thus the lack of both syn-
tactically combinable features and semantically compatible features may have rendered this
condition indistinguishable (statistically) from the control, despite the fact that the implausible
+ syntactic non-composition condition appears visually to pattern with the plausible + syntac-
tic non-composition condition. We further propose that this additive effect is responsible for
the only condition that was fully well-formed at the target word, the plausible + syntactic com-
position condition, showing significantly greater activation than many of the other test
conditions.
Considering that the plausible + syntactic composition condition was well-formed on the
target word and that this effectively created a very brief garden path effect, the lack of any
significant differences between conditions on the post-target word may be somewhat surpris-
ing. Many ERP studies have reported that reanalysis effects triggered by garden path sentences
elicit a parietal positivity starting around 400–600 ms after the point of reanalysis (e.g., Gouvea
et al., 2010; Kaan & Swaab, 2003), referred to as the P600 effect. However, many others have
noted that “reanalysis” is a rather broad explanation for the P600 (Bornkessel et al., 2003;
Tanner et al., 2017), and semantic anomalies can also elicit this component (Kim &
Osterhout, 2005, et seq.). Under a similar account, in our study, the plausible + syntactic com-
position condition in the noun manipulation would have required syntactic reanalysis on the
post-target word (drinker), and the implausible + syntactic composition condition may have
required semantic reanalysis on the post-target word in order to arrive at the interpretation that
the drinker is impolite, rather than the tea. However, we did not observe any significant effects
of the syntactic or plausibility manipulations on the post-target word. This may be because the
garden path effect in our study was much less pronounced than what has been used in pre-
vious studies. Many of the existing garden path effects relate to anomalies at the level of the
verb, whereas in our study the temporary anomaly was restricted to the noun phrase. Given
Neurobiology of Language
62
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
e
d
u
n
o
/
l
/
l
a
r
t
i
c
e
-
p
d
f
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
n
o
_
a
_
0
0
0
4
8
p
d
.
l
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Conceptual combination with and without syntax
the argument structure and thematic role related steps necessary at the verb phrase, it may be
that we did not create as strong a garden path effect.
Given the well-attested predictability effects that often co-occur with studies of semantic
plausibility (and which also co-occur in this study), we can also conclude that lexical predict-
ability did not appear to directly modulate conceptual combination in our study. Indeed, our
direct manipulation of lexical probability did not have any effect on LATL conceptual combi-
nation and only appeared as a non-significant later effect on the post-target word. Note, how-
ever, that in this particular design, the contrast between high and low transitional probability
was a relative one: “low” transition probability phrases had a corpus frequency of zero, while
“high” just had to have a probability of greater than 0.05. Whether very highly predictable
phrases still show effects of conceptual combination remains a topic for further study, as
pre-activation of a lexical item in a highly predictable context could still interact with compo-
sition effects. Though we also observed a significant effect of word category on the post-target
word characterized by greater activation for the noun manipulation (tea drinker) compared to
the adjective manipulation (sunny day), as this was a factor in our analysis, we caution against
any interpretation of those findings based on this study. First, the lexical differences between
the noun and adjective manipulations were not very controlled in our stimuli: All the post-
target words in the adjective condition were either plural nouns or mass nouns, while in
the noun manipulation half were plural and half were singular. Additionally, we did not con-
trol for frequency differences on the post-target word between these two conditions. There
were also multiple systematic differences between the noun and adjective manipulations,
any of which could be responsible for the observed effect: (i) the noun manipulation had more
morphologically complex words, as they were all deverbal nouns; (ii) the noun manipulation
had more complex argument structure than the adjective manipulation, as the target word
could always be interpreted as having an object relationship with the post-target word (a
tea drinker is someone who drinks tea); and (iii) the adjective manipulation used adjectival
modification, while the noun manipulation used nominal modification to form noun-noun
compounds. Though the question of which differences in processing arise from different types
of modification or of different levels of complexity in modification relationships is a highly
relevant question, the current design does not allow us to shed any light on the topic, and
we leave this question for future work.
Conclusion
We observed that conceptual composition as reflected in the LATL shows a stronger activation
pattern when there is local syntactic composition, though the lack of a syntactically licit parse
does not completely eradicate the composition effect. This study supports a processing model
where conceptual composition as reflected by LATL activity is influenced by local syntactic
properties, but is able to operate independent of them. Further, we found no evidence that the
local plausibility of the combination drives conceptual combination, supporting a processing
account where conceptual composition is independent of semantic plausibility.
ACKNOWLEDGMENTS
This work is funded by New York University Abu Dhabi Research Institute under Grant
G1001. We thank Brian McElree and Alec Marantz for detailed feedback on earlier drafts,
as well as the members of the Neuroscience of Language Lab at NYU and NYUAD for helpful
discussion and feedback throughout. We also thank our two anonymous reviewers for their
comments and suggestions.
Neurobiology of Language
63
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
e
d
u
n
o
/
l
/
l
a
r
t
i
c
e
-
p
d
f
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
n
o
_
a
_
0
0
0
4
8
p
d
.
l
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Conceptual combination with and without syntax
FUNDING INFORMATION
Liina Pylkkänen, New York University Abu Dhabi Research Institute Grant, Award ID: G1001
(https://dx.doi.org/10.13039/100012025).
AUTHOR CONTRIBUTIONS
Alicia Parrish: Conceptualization; Investigation; Data curation; Formal analysis; Visualization;
Writing – original draft; Writing – review & editing; Methodology; Project administration. Liina
Pylkkänen: Conceptualization; Supervision; Writing – review & editing; Resources;
Methodology; Funding acquisition.
REFERENCES
Baron, S. G., & Osherson, D. (2011). Evidence for conceptual com-
bination in the left anterior temporal lobe. NeuroImage, 55(40),
1847–1852. https://doi.org/10.1016/j.neuroimage.2011.01.066,
PubMed: 21281723
Baron, S. G., Thompson-Schill, S. L., Weber, M., & Osherson, D.
( 2 0 1 0 ) . A n e a r l y s t a g e o f c o n c e p t u a l c o m b i n a t i o n :
Superimposition of constituent. Cognitive Neuroscience, 1(1),
44–51. https://doi.org/10.1080/17588920903548751, PubMed:
24168244
Bemis, D. K., & Pylkkänen, L. (2011). Simple composition: A mag-
netoencephalography investigation into the comprehension of
minimal linguistic phrases. Journal of Neuroscience, 31(8),
2801–2814. https://doi.org/10.1523/JNEUROSCI.5003-10.2011,
PubMed: 21414902
Bemis, D. K., & Pylkkänen, L. (2012). Basic linguistic composition
recruits the left anterior temporal lobe and left angular gyrus
during both listening and reading. Cerebral Cortex, 23(8),
1859–1873. https://doi.org/10.1093/cercor/ bhs170, PubMed:
22735156
Bemis D. K., & Pylkkänen L. (2013). Flexible composition: MEG
evidence for the deployment of basic combinatorial linguistic
mechanisms in response to task demands. PLoS ONE 8(9),
Article e73949. https://doi.org/10.1371/journal.pone.0073949,
PubMed: 24069253
Binney, R. J., Embleton, K. V., Jefferies, E., Parker, G. J., & Ralph,
M. A. (2010). The ventral and inferolateral aspects of the anterior
temporal lobe are crucial in semantic memory: Evidence from a
novel direct comparison of distortion-corrected fMRI, rTMS, and
semantic dementia. Cerebral Cortex, 20(11), 2728–2738. https://
doi.org/10.1093/cercor/bhq019, PubMed: 20190005
Blanco-Elorrieta, E., Kastner, I., Emmorey, K., & Pylkkänen, L.
(2018). Shared neural correlates for building phrases in signed
and spoken language. Scientific Reports, 8(1), 1–10. https://doi
.org/10.1038/s41598-018-23915-0, PubMed: 29615785
Blanco-Elorrieta, E., & Pylkkänen, L. (2016). Composition of
complex numbers: Delineating the computational role of the
left anterior temporal lobe. NeuroImage, 124(Part A), 194–203.
https://doi.org/10.1016/j.neuroimage.2015.08.049, PubMed:
26325329
Bornkessel, I., Schlesewsky, M., & Friederici, A. D. (2003). Eliciting
thematic reanalysis effects: The role of syntax-independent infor-
mation during parsing. Language and Cognitive Processes, 18(3),
269–298. https://doi.org/10.1080/01690960244000018
Boylan, C., Trueswell, J. C., & Thompson-Schill, S. L. (2015).
Compositionality and the angular gyrus: A multi-voxel similarity
analysis of the semantic composition of nouns and verbs.
Neuropsychologia, 78, 130–141. https://doi.org/10.1016/j
.neuropsychologia.2015.10.007, PubMed: 26454087
Boylan, C., Trueswell, J. C., & Thompson-Schill, S. L. (2017).
Relational vs. attributive interpretation of nominal compounds
differentially engages angular gyrus and anterior temporal lobe.
Brain & Language, 169, 8–21. https://doi.org/10.1016/j.bandl
.2017.01.008, PubMed: 28236762
Brennan, J., Stabler, E., Van Wagenen, S. E., Luh, W. M., & Hale,
J. T. (2016). Abstract linguistic structure correlates with temporal
activity during naturalistic comprehension. Brain & Language,
157, 81–94. https://doi.org/10.1016/j.bandl.2016.04.008,
PubMed: 27208858
Brodbeck, C. (2018). Eelbrain, Zenodo, Version 0.27. https://doi
.org/10.5281/zenodo.1186450
Cabin, R. J., & Mitchell, R. J. (2000). To Bonferroni or not to
Bonferroni: When and how are the questions. Ecological
Society of America, 81(3), 246–248.
Dale, A. M., Fischl, B., & Sereno, M. I. (1999). Cortical surface-
based analysis: I. Segmentation and surface reconstruction.
NeuroImage, 9(2), 179–194. https://doi.org/10.1006/nimg.1998
.0395, PubMed: 9931268
Dale, A. M., Liu, A. K., Fischl, B. R., Buckner, R. L., Belliveau, J. W.,
Lewine, J. D., & Halgren, E. (2000). Dynamic statistical parametric
mapping: Combining fMRI and MEG for high-resolution imaging
of cortical activity. Neuron, 26(1), 55–67. https://doi.org/10.1016
/S0896-6273(00)81138-1, PubMed: 10798392
Del Prato, P., & Pylkkänen, L. (2014). MEG evidence for conceptual
combination but not numeral quantification in the left anterior
temporal lobe during language production. Frontiers in
Psychology, 5, Article 524. https://doi.org/10.3389/fpsyg.2014
.00524, PubMed: 24926273
Desikan, R. S., Ségonne, F., Fischl, B., Quinn, B. T., Dickerson,
B. C., Blacker, D., Buckner, R. L., Dale, A. M., Maguire, R. P.,
Hyman, B. T., Albert, M. S., & Killiany, R. J. (2006). An automated
labeling system for subdividing the human cerebral cortex on MRI
scans into gyral based regions of interest. NeuroImage, 31(3),
968–980. https://doi.org/10.1016/j.neuroimage.2006.01.021,
PubMed: 16530430
Erlewine, M. Y., & Kotek, H. (2016). A streamlined approach to on-
line linguistic surveys. Natural Language and Linguistic Theory,
34, 481–495. https://doi.org/10.1007/s11049-015-9305-9
Fedorenko, E., Blank, I. A., Siegelman, M., & Mineroff, Z. (2020).
Lack of selectivity for syntax relative to word meanings through-
out the language network. Cognition, 203, Article 104348.
https://doi.org/10.1016/j.cognition.2020.104348, PubMed:
32569894
Neurobiology of Language
64
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
e
d
u
n
o
/
l
/
l
a
r
t
i
c
e
-
p
d
f
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
n
o
_
a
_
0
0
0
4
8
p
d
.
l
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Conceptual combination with and without syntax
Fedorenko, E., & Thompson-Schill, S. L. (2014). Reworking the lan-
guage network. Trends in Cognitive Sciences, 18(3), 120–126.
https://doi.org/10.1016/j.tics.2013.12.006, PubMed: 24440115
Flick, G., & Pylkkänen, L. (2020). Isolating syntax in natural lan-
guage: MEG evidence for an early contribution of left posterior
temporal cortex. Cortex, 127, 42–57. https://doi.org/10.1016/j
.cortex.2020.01.025, PubMed: 32160572
Friederici, A. D. (2011). The brain basis of language processing: From
structure to function. Physiological Review, 91(4), 1357–1392.
https://doi.org/10.1152/physrev.00006.2011, PubMed: 22013214
Friederici, A. D., Fiebach, C. J., Schlesewsky, M., Bornkessel, I. D.,
& von Cramon, D. Y. (2006). Processing linguistic complexity
and grammaticality in the left frontal cortex. Cerebral Cortex,
16(12), 1709–1717. https://doi.org/10.1093/cercor/ bhj106,
PubMed: 16400163
Friederici, A. D., Meyer, M., & von Cramon, D. Y. (2000). Auditory
language comprehension: An event-related fMRI study on the
processing of syntactic and lexical information. Brain and
Language, 74(2), 289–300. https://doi.org/10.1006/ brln.2000
.2313, PubMed: 10950920
Gouvea, A. C., Phillips, C., Kazanina, N., & Poeppel, D. (2010).
The linguistic processes underlying the P600. Language and
Cognitive Processes, 25(2), 149–188. https://doi.org/10.1080
/01690960902965951
Graessner, A., Zaccarella, E., & Hartwigsen, G. (2021). Differential
contributions of left-hemispheric language regions to basic seman-
tic composition. Brain Structure and Function, 226, 501–518.
https://doi.org/10.1007/s00429-020-02196-2, PubMed: 33515279
Gramfort, A., Luessi, M., Larson, E., Engemann, D. A., Strohmeier,
D., Brodbeck, C., Parkkonen, L., & Hämäläinen, M. S. (2014).
MNE software for processing MEG and EEG data. NeuroImage,
86, 446–460. https://doi.org/10.1016/j.neuroimage.2013.10
.027, PubMed: 24161808
Grodzinsky, Y., & Santi, A. (2008). The battle for Broca’s region.
Trends in Cognitive Science, 12(12), 474–480. https://doi.org
/10.1016/j.tics.2008.09.001, PubMed: 18930695
Hagoort, P. (2005). On Broca, brain, and binding: A new frame-
work. Trends in Cognitive Sciences, 9(9), 416–423. https://doi
.org/10.1016/j.tics.2005.07.004, PubMed: 16054419
Hagoort, P. (2017). The core and beyond in the language-ready
brain. Neuroscience and Biobehavioral Reviews, 81(Part B),
194–204. https://doi.org/10.1016/j.neubiorev.2017.01.048,
PubMed: 28193452
Humphries, C., Love, T., Swinney, D., & Hickok, G. (2005).
Response of anterior temporal cortex to syntactic and prosodic
manipulations during sentence processing. Human Brain
Mapping, 26(2), 128–138. https://doi.org/10.1002/hbm.20148,
PubMed: 15895428
Kaan, E., & Swaab, T. Y. (2003). Repair, revision and complexity in
syntactic analysis: An electrophysiological differentiation.
Journal of Cognitive Neuroscience, 15, 98–110. https://doi.org
/10.1162/089892903321107855, PubMed: 12590846
Kim, S., & Ostrehout, L. (2005). The independence of combinatory
semantic processing: Evidence from event-related potentials.
Journal of Memory and Language, 52(2), 205–225. https://doi.org
.org/10.1016/j.jml.2004.10.002
Kim, S., & Pylkkänen, L. (2019). Composition of event concepts:
Evidence for distinct roles for the left and right anterior temporal
lobes. Brain and Language, 188, 18–27. https://doi.org/10.1016/j
.bandl.2018.11.003, PubMed: 30530286
Krippendorff, K. (2011). Computing Krippendorff’s alpha-reliability.
University of Pennsylvania repository. https://repository.upenn
.edu/asc_papers/43
Lambon Ralph, M. A., Jefferies, E., Patterson, K., & Rogers, T. T.
(2017). The neural and computational bases of semantic cogni-
tion. Nature Reviews Neuroscience, 18(1), 42–55. https://doi.org
/10.1038/nrn.2016.150, PubMed: 27881854
Lambon Ralph, M. A., Pobric, G., & Jefferies, E. (2009). Conceptual
knowledge is underpinned by the temporal pole bilaterally:
Convergent evidence from rTMS. Cerebral Cortex, 19(4), 832–838.
https://doi.org/10.1093/cercor/bhn131, PubMed: 18678765
Lau, E. F., Phillips, C., & Poeppel, D. (2008). A cortical network for
semantics: (de)constructing the N400. Nature Reviews
Neuroscience, 9(12), 920–933. https://doi.org/10.1038
/nrn2532, PubMed: 19020511
Lukic, S., Thompson, C. K., Barbieri, E., Chiappetta, B., Bonakdarpour,
B., Kiran, S., Rapp, B., Parrish, T. B., & Caplan, D. (2021). Common
and distinct neural substrates of sentence production and compre-
hension. NeuroImage, 224, Article 117374. https://doi.org/10.1016
/j.neuroimage.2020.117374, PubMed: 32949711
Maess, B., Herrmann, C. S., Hahne, A., Nakamura, A., & Friederici,
A. D. (2006). Localizing the distributed language network re-
sponsible for the N400 measured by MEG during auditory sen-
tence processing. Brain Research, 1096(1), 163–172. https://doi
.org/10.1016/j.brainres.2006.04.037, PubMed: 16769041
Maris, E., & Oostenveld, R. (2007). Nonparametric statistical testing
of EEG-and MEG-data. Journal of Neuroscience Methods, 164(1),
177–190. https://doi.org/10.1016/j.jneumeth.2007.03.024,
PubMed: 17517438
Matchin, W., Hammerly, C., & Lau, E. (2017). The role of the IFG and
pSTS in syntactic prediction: Evidence from a parametric study of
hierarchical structure in fMRI. Cortex, 88, 106–123. https://doi.org
/10.1016/j.cortex.2016.12.010, PubMed: 28088041
Matchin, W., & Hickok, G. (2020). The cortical organization of
syntax. Cerebral Cortex, 30(3), 1481–1498. https://doi.org/10
.1093/cercor/bhz180, PubMed: 31670779
Mesulam, M.-M., Rader, B. M., Sridhar, J., Nelson, M. J., Hyun, J.,
Rademaker, A., Geula, C., Bigio, E. H., Thompson, C. K., Gefen,
T. D., Weintraub, S., & Rogalski, E. J. (2019). Word comprehen-
sion in temporal cortex and Wernicke area: A PPA perspective.
Neurology, 92(3), e224–e233. https://doi.org/10.1212/ WNL
.0000000000006788, PubMed: 30578374
Mesulam, M.-M., Thompson, C. K., Weintraub, S., & Rogalski, E. J.
(2015). The Wernicke conundrum and the anatomy of language
comprehension in primary progressive aphasia. Brain, 138(8),
2423–2437. https://doi.org/10.1093/ brain/awv154, PubMed:
26112340
Mollica, F., Siegelman, M., Diachek, E., Piantadosi, S. T., Mineroff,
Z., Futrell, R., Kean, H., Qian, P., & Fedorenko, E. (2020).
Composition is the core driver of the language-selective network.
Neurobiology of Language, 1(1), 104–134. https://doi.org/10
.1162/nol_a_00005
Pallier, C., Devauchelle, A.-D., & Dehaene, S. (2011). Cortical rep-
resentation of the constituent structure. Proceedings of the
National Academy of Sciences, 108(6), 2522–2527. https://doi
.org/10.1073/pnas.1018711108, PubMed: 21224415
Peirce, J. W. (2007). Peirce, Jonathan W. PsychoPy—psychophysics
software in Python. Journal of Neuroscience Methods, 162(1–2),
8–13. https://doi.org/10.1016/j.jneumeth.2006.11.017, PubMed:
17254636
Poortman, E. B., & Pylkkänen, L. (2016). Adjective conjunction as a
window into the LATL’s contribution to conceptual combination.
Brain & Language, 160, 50–60. https://doi.org/10.1016/j.bandl
.2016.07.006, PubMed: 27475469
Pylkkänen, L., Brennan, J., & Bemis, D. K. (2011). Grounding the
cognitive neuroscience of semantics in linguistic theory.
Neurobiology of Language
65
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
e
d
u
n
o
/
l
/
l
a
r
t
i
c
e
-
p
d
f
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
n
o
_
a
_
0
0
0
4
8
p
d
.
l
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Conceptual combination with and without syntax
Language and Cognitive Processes, 26(9), 1317–1337. https://doi
.org/10.1080/01690965.2010.527490
Pylkkänen, L., & Marantz, A. (2003). Tracking the time course of
word recognition with MEG. Trends in Cognitive Science, 7,
187–189. https://doi.org/10.1016/S1364-6613(03)00092-5,
PubMed: 12757816
Pylkkänen, L., Oliveri, B., & Smart, A. J. (2009). Semantics vs. world
knowledge in prefrontal cortex. Language and Cognitive
Processes, 24(9), 1313–1334. https://doi.org/10.1080
/01690960903120176
Rice, W. R. (1989). Analyzing tables of statistical tests. Society for
the Study of Evolution, 43(1), 223–225. https://doi.org/10.1111/j
.1558-5646.1989.tb04220.x, PubMed: 28568501
Schell, M., Zaccarella, E., & Friederici, A. D. (2017). Differential
cortical contribution of syntax and semantics: An fMRI study
on two-word phrasal processing. Cortex, 96, 105–120. https://
doi.org/10.1016/j.cortex.2017.09.002, PubMed: 29024818
Tanner, D., Grey, S., & van Hell, J. G. (2017). Dissociating retrieval
interference and reanalysis in the P600 during sentence compre-
hension. Psychophysiology, 54(2), 248–259. https://doi.org/10
.1111/psyp.12788, PubMed: 27859315
Westerlund, M., & Pylkkänen, L. (2014). The role of the left anterior
temporal lobe in semantic composition vs. semantic memory.
Neuropsychologia, 57, 59–70. https://doi.org/10.1016/j
.neuropsychologia.2014.03.001, PubMed: 24631260
Wilson, S. M., DeMarco, A. T., Henry, M. L., Gesierich, B., Babiak,
M., Mandelli, M. L., Miller, B. L., & Gorno-Tempini, M. L. (2014).
What role does the anterior temporal lobe play in sentence-level
processing? Neural correlates of syntactic processing in semantic
variant primary progressive aphasia. Journal of Cognitive
Neuroscience, 26(5), 970–985. https://doi.org/10.1162/jocn_a
_00550, PubMed: 24345172
Zaccarella, E., Meyer, L., Makuuchi, M., & Friederici, A. D. (2017).
Building by syntax: The neural basis of minimal linguistic struc-
tures. Cerebral Cortex, 27(1), 411–421. https://doi.org/10.1093
/cercor/bhv234, PubMed: 26464476
Zhang, L., & Pylkkänen, L. (2015). The interplay of composition
and concept specificity in the left anterior temporal lobe: An
MEG study. NeuroImage, 111, 228–240. https://doi.org/10
.1016/j.neuroimage.2015.02.028, PubMed: 25703829
Zhang, L., & Pylkkänen, L. (2018a). Composing lexical versus
functional adjectives: Evidence for uniformity in the left tempo-
ral lobe. Psychonomic Bulletin & Review, 25, 2309–2322.
https://doi.org/10.3758/s13423-018-1469-y, PubMed:
29693207
Zhang, L., & Pylkkänen, L. (2018b). Semantic composition of
sentences word by word: MEG evidence for shared processing
of conceptual and logical elements. Neuropsychologia, 119,
392–404. https://doi.org/10.1016/j.neuropsychologia.2018
.08.016, PubMed: 30138672
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
e
d
u
n
o
/
l
/
l
a
r
t
i
c
e
-
p
d
f
/
/
/
/
/
3
1
4
6
1
9
8
6
8
7
0
n
o
_
a
_
0
0
0
4
8
p
d
.
l
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Neurobiology of Language
66
