Science - USA (2021-11-12)

in syntactic training ( 47 ) and in the implement-
ation of grammatical rules ( 21 , 33 ). Further-
more,itactsasaparserofactionstochunk
motor sequences ( 24 , 48 ). Accurate and effi-
cient tool use requires embedding an exter-
nal object into the motor sequence and thus
relies more on the striatum than on manual
actions to parse the motor primitives ( 4 ). Dur-
ing dexterous tool use, hand movements integ-
rate the functional structure of the tool to
maintain an efficient interaction with the ac-
tion target. The sensorimotor transformations
imposed by the tool ( 28 ) constitute the ad-
ditional level embedded in the manual motor
program. Parsing and hierarchy handling also
support syntactic comprehension of center-
embedded object relatives ( 21 , 33 , 34 ). These
functional similarities are reflected by the neu-
ral overlap that we revealed between tool use
and syntax. However, the complexity of the
hierarchies to be handled might not be strictly
commensurate. It is worth noting that this
overlap was found in the BG but not within
the left IFG. In keeping with its documented
involvement in both tool use ( 35 ) and lan-
guage ( 15 ), the lIFG was recruited by both
functions yet in two separately clustered re-
gions. This occurred more posteriorly in the
ventral premotor cortex for tool use and more
anteriorly in Broca’s area for syntactic com-
prehension, in accordance with the cytoarchi-
tectonic and functional specialization of the
lIFG for motor and linguistic processing, re-
spectively ( 49 , 50 ).
These results explain the cross-domain learn-
ing transfer from tool use to syntactic skills in
language and from linguistic syntax training to
skilled tool use. Learning transfer arises pro-
vided that trained and untrained tasks rely on
overlapping neural networks and shared cog-
nitive processes ( 41 ). Transfer effects have been
demonstrated from trained to untrained tasks
belonging to the same domain: perception ( 51 ),
motor ( 52 ), or cognitive control ( 41 Ð 43 ). Cru-
cially, we extended to different cognitive do-
mains the principle of transfer that had so far
been limited to a single domain ( 41 , 42 , 51 , 52 ).
Thus, the transfer holds true even when dif-
ferent cognitive domains such as action and
language are involved. If trained and un-
trained tasks do not share common neuro-
cognitive resources, then transfer might be
tempered or absent. Indeed, training with
subject-relative structures did not improve
motor performance with the tool, and free-
hand training failed to induce benefits to
syntax in the comprehension of complex struc-
tures. Furthermore, the benefits induced by
tool use over language were not based on the
mere additional sensorimotor complexity of
the action executed with the tool compared
with the free hand. After training with a hand
configuration that involved similar sensorimo-
tor constraints imposed by the tool, the parti-

cipants did not show any advantage in pro-

cessing complex syntactic structures compared

with the participants training with the free

hand. The learning transfer between tool use

and syntactic processes in language occurs

bidirectionally. This finding unambiguously

indicates that the two abilities rely on a com-

mon cognitive component, namely a supra-

modal syntax. It also suggests that the neural

resources underlying the shared function can

be similarly mobilized by either of the two

abilities to improve the other.

What drives this cross-domain transfer? Pre-

activation of common resources and fast plas-

ticity within shared circuitries can underlie

the reciprocal boosting of behavioral perform-

ance in tool use and syntax processing in lan-

guage. Training may act as a functional prime

for the subsequent task: Training-dependent

neuronal responses are elicited by tool use or

sentence processing, thus yielding neuronal

adaptation and more efficient activity ( 53 ).

This in turn facilitates the subsequent behav-

ioral performance for the untrained task rely-

ing on the same neural assemblies ( 54 , 55 ).

Alternatively, cross-domain transfer may rely

on fast plastic changes within common circuit-

ries. Short motor training (i.e., <2 hours) trig-

gers rapid functional ( 44 ) and local structural

changes, which are accompanied by improve-

ments in behavioral performance ( 56 ). The

untrained task may benefit from such plastic

changes and recruit new resources within the

shared territories. These results raise the ques-

tion of whether learning transfer can general-

ize to other linguistic tasks and if an optimal

training duration could maximize the benefits.

In the theoretical framework of the expansion-

renormalization hypothesis ( 57 ), a thrilling op-

portunity would be to take advantage of the

temporal dynamics of plastic changes, for in-

stance, by testing syntax while new neural re-

sources are locally and temporarily available

during the course of tool-use training.

Overall, our findings reignite the hypothesis

of a coevolution of tool use and language

( 58 – 60 ). Longstanding theories have claimed

a motor origin of language during evolution

( 25 , 61 ). The advent and refinement of tool use

may have offered the neural niche for the co-

evolution of new cognitive skills serving both

motor and communicative aims ( 5 , 58 , 62 , 63 ).

According to this hypothesis, the role of tool

use has been twofold. On the one hand, the

sophistication of tool use and tool making has

put forward the need for cognitive functions to

efficiently chunk, temporally parse, and deal

with hierarchies of sequences ( 60 ). On the

other hand, tool use and tool making posed

evolutionary pressure for communication, al-

lowing better social transmission of knowl-

edge ( 63 ). Functions responding to demands

of the motor system would therefore have

met communicative needs and progressively

been exapted and recycled for language ( 62 , 64 ).

Such a coevolution scenario has involved a

large brain network, from parietal ( 58 , 60 ) to

frontal regions ( 60 , 65 ) and including the BG

( 66 ). Here, we provide central human evidence

pointing to the BG in particular as the neural

niche for a supramodal syntactic function

serving both action and language. Our findings

show that the motor system can be exploited to

promote other cognitive functions that partly

share the same neurocognitive foundations.

Materials and Methods

Participants

A total of 244 participants were included in

the study, which consisted of five different ex-

periments. None of the participants took part

in more than one experiment. All participants

were healthy, right-handed, French native

speakers with normal or corrected-to-normal

vision and no known motor, linguistic, or neu-

rological disorders. They gave their written

informed consent before experiment acquisi-

tion. All procedures agreed with the Helsinki

declaration and were approved by an ethical

committee (46/17_2, OUEST IV).

Experiment 1

The fMRI acquisition in Experiment 1 included

24 participants who received a compensation

of 110 euros. Four participants were excluded:

two did not fulfill the pre-set familiarization

performance requirements before any neuro-

imaging acquisition, one dropped out after the

inclusion phase, and one was removed from

analyses because of substantial head move-

ments (several runs with movements >1.5 mm).

Data from 20 participants with the following

sociodemographic characteristics and man-

ual preference were analyzed: 10 males and

10 females; mean age ± SD: 24 ± 4 years; mean

score on the Edinburgh handedness inventory

( 67 ): 0.93 ± 0.09; higher education level (the

number of years of education after a high school

degree):3±2years.

Experiment 2

In this experiment, 85 participants were re-

cruited and paid 10 euros for their participa-

tion. Six participants performed the experiment

but were excluded from analysis because of

performance below chance level in at least one

sentence condition and one was excluded for

having incorrectly used the right hand to de-

liver the button press. Data from 78 partici-

pants with the following characteristics were

analyzed: 27 males and 51 females; mean age:

23 ± 3 years; mean Edinburgh score: 0.9 ±

0.12; higher education level: 4 ± 1 years.

Experiment 3

In this experiment, 46 participants were re-

cruited and paid 15 euros for their participa-

tion. The threshold ofd′>1.38inpretest,as

Thibaultet al.,Science 374 , eabe0874 (2021) 12 November 2021 8 of 14

RESEARCH | RESEARCH ARTICLE