ARTICLE DE RECHERCHE - Recherche en IA spécialisée au MIT

ARTICLE DE RECHERCHE

Behavioral and Neural Correlates of Speech Motor
Sequence Learning in Stuttering and Neurotypical
Speakers: An fMRI Investigation

un accès ouvert

journal

Alfonso Nieto-Castañón1, Matthias Heyne1, Saul A. Frankford1

Matthew Masapollo1,2, Jennifer A. Segawa1,3

, Deryk S. Beal1,4

, Jason A. Tourville1
, and Frank H. Guenther1,3,5,6

1Department of Speech, Language, and Hearing Sciences, Université de Boston, Boston, MA, Etats-Unis
2Department of Speech, Language, and Hearing Sciences, University of Florida, Gainesville, FL, Etats-Unis
3Departments of Neuroscience and Biology, Stonehill College, Easton, MA, Etats-Unis
4Department of Speech-Language Pathology, University of Toronto, Toronto, Canada
5Department of Biomedical Engineering, Université de Boston, Boston, MA, Etats-Unis
6Picower Institute for Learning and Memory, Massachusetts Institute of Technology, Cambridge, MA, Etats-Unis

Mots clés: speech motor control, motor sequence learning, stuttering, GODIVA model, IRMf

ABSTRAIT

Stuttering is a neurodevelopmental disorder characterized by impaired production of coordinated
articulatory movements needed for fluent speech. It is currently unknown whether these abnormal
production characteristics reflect disruptions to brain mechanisms underlying the acquisition
and/or execution of speech motor sequences. To dissociate learning and control processes, nous
used a motor sequence learning paradigm to examine the behavioral and neural correlates of
learning to produce novel phoneme sequences in adults who stutter (AWS) and neurotypical
controls. Participants intensively practiced producing pseudowords containing non-native
consonant clusters (par exemple., “GVAZF”) over two days. The behavioral results indicated that although the
two experimental groups showed comparable learning trajectories, AWS performed significantly
worse on the task prior to and after speech motor practice. Using functional magnetic resonance
imaging (IRMf), the authors compared brain activity during articulation of the practiced words
and a set of novel pseudowords (matched in phonetic complexity). FMRI analyses revealed no
differences between AWS and controls in cortical or subcortical regions; both groups showed
comparable increases in activation in left-lateralized brain areas implicated in phonological
working memory and speech motor planning during production of the novel sequences compared
to the practiced sequences. De plus, activation in left-lateralized basal ganglia sites was negatively
correlated with in-scanner mean disfluency in AWS. Collectively, these findings demonstrate
that AWS exhibit no deficit in constructing new speech motor sequences but do show impaired
execution of these sequences before and after they have been acquired and consolidated.

INTRODUCTION

Stuttering is a neurodevelopmental disorder affecting children into adulthood with devastating
social effects that impede speech communication. The presenting clinical signs of the disorder
include involuntary repetitions and prolongations of phonemes, syllables, or words, ainsi que
involuntary silent pauses (Bloodstein & Ratner, 2008). Despite considerable research, chercheurs
still have a relatively poor understanding of the neural deficits underlying the disorder (see Craig-
McQuaide et al., 2014, for a review). Neurobiological accounts of stuttering, such as Alm (2004),

Citation: Masapollo, M., Segawa, J.. UN., Beal,
D. S., Tourville, J.. UN., Nieto-Castañón, UN.,
Heyne, M., Frankford, S. UN., & Guenther, F. H.
(2021). Behavioral and neural correlates
of speech motor sequence learning in
stuttering and neurotypical speakers:
An fMRI investigation. Neurobiology of
Language, 2(1), 106–137. https://est ce que je
.org/10.1162/nol_a_00027

EST CE QUE JE:
https://doi.org/10.1162/nol_a_00027

Informations complémentaires:
https://doi.org/10.1162/nol_a_00027

Reçu: 5 Octobre 2020
Accepté: 11 Novembre 2020

Intérêts concurrents: Les auteurs ont
a déclaré qu'aucun intérêt concurrent
exister.

Auteur correspondant:
Matthew Masapollo
mmasapollo@phhp.ufl.edu

Éditeur de manipulation:
Kate Watkins

droits d'auteur: © 2021 Massachusetts
Institut de technologie. Published
under a Creative Commons Attribution
4.0 International (CC PAR 4.0) Licence.

La presse du MIT

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
n
o

je
/

un
r
t
je
c
e
–
p
d

F
/

2
1
1
0
6
1
8
9
9
6
7
7
n
o
_
un
_
0
0
0
2
7
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

Neural correlates of speech motor sequence learning

Max et al. (2004), and recently Connally et al. (2018) and Chang and Guenther (2020), proposed
that stuttering disfluencies arise from an underlying defect in the (left-lateralized) cortico-basal-
ganglia-thalamocortical (cortico-BG) loops (Alexander et al., 1986), which are hypothesized to
be responsible for selecting and triggering articulatory motor programs, while suppressing others,
at the appropriate time during speech sequencing.

There is an extensive body of evidence (recently reviewed by Chang & Guenther, 2020) dans
favor of the idea that stuttering involves an inability to initiate, sustain, and/or terminate speech
motor programs due to anomalous basal ganglia function. Par exemple, numerous acoustic
investigations have provided evidence that stuttering involves an impaired ability to transition from
the articulatory gestures affiliated with an initial phoneme of a syllable to subsequent gestures, comme
indicated by protracted voice onset times and formant transition rates (par exemple., Boutsen, 1993; Chang
et coll., 2002; Robb & Blomgren, 1997; Robb et al., 1998; Yaruss & Conture, 1993). As for the
underlying neural circuitry, abnormal functional activity in the basal ganglia has frequently been
implicated in stuttering severity (Connally et al., 2018; Giraud et al., 2008; Ingham et al., 2012;
Toyomura et al., 2015). En outre, stuttering-like disfluencies have been reported to emerge in
neurodegenerative diseases that impair the function of the basal ganglia, such as Parkinson’s
maladie (Niethammer et al., 2012), and deep brain stimulation applied to either the subthalamic
nucleus or the globus pallidus pars interna of the basal ganglia has been shown to relieve or
exacerbate stuttering disfluencies (see Sapir, 2014, and Skodda, 2012, for reviews).

De plus, and of particular relevance to the current research, some existing theoretical
models attribute a role in speech motor sequence learning to the cortico-BG loops (par exemple.,
Guenther, 2016, Chapters 8 & 10; Kotz & Schwartze, 2010; see also Fee & Goldberg, 2011,
for a similar account in songbird vocal learning). Par exemple, in the Gradient Order Directions
Into Velocities of the Articulators (GODIVA) model of speech sequencing and initiation (Bohland
et coll., 2010; Civier et al., 2013; Guenther, 2016, Chapter 8; Segawa, Masapollo, et coll., 2019), it
is proposed that continual speech motor practice leads to the formation of optimized motor
programs for frequently occurring kinematic sequences, such as the successive gestures making
up a word or syllable (see also, Cholin et al. 2006). These learned movement “chunks” or
“templates” are then encoded and sequentially activated (or “read out”) by the cortico-BG loops
during production. The fundamental idea is that the cortico-BG loops perform two distinct, mais
related, cognitive operations during the speech production process, namely, forming regularized
motor chunks and initiating/activating motor programs affiliated with those chunks.

Results from a functional magnetic resonance imaging (IRMf) study of speech motor sequence
learning with neurotypical adults (Segawa, Tourville, et coll., 2015) provided empirical support for
this view. In the initial training phase of the study, participants performed a cued motor sequence
learning task (outside the scanner) in which they intensively practiced producing pseudowords
(monosyllabic CCVCC[C], where C = consonant, V = vowel) consisting of consonant clusters that
were either phonotactically legal in their native language of English (par exemple., “BLERK”) or illegal in
English but legal in other natural human languages (par exemple., “GVAZF”). The results indicated that,
behaviorally, motor practice led to measurable performance gains (c'est à dire., reduced error rates and
shorter utterance durations), presumably because the sequences of independent speech move-
ments had become concatenated into larger cohesive chunks (see Segawa, Masapollo, et coll.,
2019, for supporting evidence).

In the subsequent test phase (inside the scanner), the authors measured the blood oxygen-level
dependent (AUDACIEUX) response during production of the practiced native and practiced non-native
words as well as a novel set of non-native words (matched in phonetic complexity) using sparse
sampling fMRI (voir, par exemple., Gracco et al., 2005). In one set of imaging analyses, BOLD activity

Basal ganglia:
Set of nuclei found underneath the
cortex in the medial position of
the brain.

Motor sequence learning:
Learning how to plan and execute an
appropriate order of discrete actions.

Motor chunks:
Cohesive action units composed of
frequently occurring subsequences
of movements.

Functional magnetic resonance
imaging (IRMf):
Imaging that measures changes in
blood oxygenation levels that occur
in response to neural firing, allowing
precise localization of brain activity.

Consonant clusters:
Consonant sequences at the start or
end of a syllable.

Neurobiology of Language

107

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
n
o

je
/

un
r
t
je
c
e
–
p
d

F
/

2
1
1
0
6
1
8
9
9
6
7
7
n
o
_
un
_
0
0
0
2
7
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

Neural correlates of speech motor sequence learning

Anterior insula (aINS):
Brain region that plays a role in speech
motor sequence learning.

associated with production of the novel non-native words was contrasted with that of the practiced
non-native words. The results demonstrated increased activation in (left-lateralized) cortical (c'est à dire.,
pre-supplementary motor area [preSMA], anterior insula [aINS], inferior frontal sulcus [IFS], inferior
parietal sulcus [IPS], ventral premotor cortex [vPMC]), and subcortical basal ganglia (c'est à dire., globus
pallidus [GP]) regions involved in the cortico-BG loops (Alexander et al., 1986). It was further dem-
onstrated that the degree of motor performance improvements between training and test was cor-
related positively with activity in the left aINS. The cortical regions that were found to be recruited in
this task overlap with activation foci that have been identified in meta-analyses of working memory
neuroimaging studies (voir, par exemple., Rottschy et al., 2012). Ainsi, these findings are in keeping with the
hypothesis that the cortico-BG loops play an important role in learning and sequencing speech
movement chunks since a higher number of chunks will have to be concatenated in a working
memory repository (or phonological output store) and then subsequently read out during the pro-
duction of novel sequences compared to practiced ones.

In another set of analyses contrasting BOLD activity associated with production of the prac-
ticed non-native and practiced native words, the authors found increased activation in a similar
network of brain areas, although notably no increased activation was found in the left GP for this
contraste. They did, cependant, report increased activation in the right cerebellum (lobule VI) for this
contraste, suggesting that other subcortical structures that extend beyond the cortico-BG may also
play a role in speech motor sequencing learning and motor control (a point we will return to in
the general discussion).

Nevertheless, if the cortico-BG loops are involved in successively encoding and activating
speech motor chunks during sequence learning and production, then that raises the possibility that
the neural deficits hypothesized to underlie stuttering are related not only to problems with motor
execution, but also to impaired motor learning mechanisms. Consistent with this view are experi-
mental findings, in the behavioral literature, that adults who stutter (AWS) and adults with normal
speech (ANS) often show differences in their ability to learn a variety of novel speech and non-
speech (c'est à dire., finger tapping) movement sequences (par exemple., Ludlow et al., 1997; Namasivayam &
van Lieshout, 2008; Smits-Bandstra et al., 2006). Qualitatively similar findings have also been
found in patients with Parkinson’s disease (Ferraro et al., 1993; Helmuth et al., 2000; Jackson
et coll., 1995; Mollaei et al., 2013; Smits-Bandstra & Gracco, 2013; Vakil et al., 2000).

The Current Study

In the current research, we performed a functional brain-imaging study to distinguish between
motor execution and motor learning impairments in AWS at both the behavioral and neural
levels, by replicating and extending Segawa, Tourville, et autres. (2015) with a cohort of adult
speakers with persistent developmental stuttering. Toward this end, we trained AWS and ANS
to produce pseudowords containing non-native consonant clusters. Participants were first trained
over several days (outside the scanner) to produce two sets of novel CCVCC pseudowords: (1)
syllables that involved native consonant clusters ( practiced native) et (2) syllables consisting of
non-native consonant clusters that are phonotactically illegal in English ( practiced non-native).
Based on previous results (Segawa, Tourville, et coll., 2015; see also Segawa, Masapollo, et coll.,
2019), we expected to observe significantly larger performance gains throughout the course of
training for the practiced non-native words than for the practiced native words (for which perfor-
mance is already expected to be near ceiling at the beginning of training). During a subsequent
testing phase (inside the scanner), we then contrasted BOLD activity associated with the produc-
tion of the practiced non-native and practiced native words, and the practiced non-native and
practiced native sequences.

Neurobiology of Language

108

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
n
o

je
/

un
r
t
je
c
e
–
p
d

F
/

2
1
1
0
6
1
8
9
9
6
7
7
n
o
_
un
_
0
0
0
2
7
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

Neural correlates of speech motor sequence learning

We hypothesized that, if motor sequence learning mechanisms per se are impaired in stuttering,
alors: (1) At the behavioral level, AWS should show a slower rate of learning over the course of
training and/or poorer learning outcomes compared to ANS; et (2) At the neural level, contrasting
the BOLD signal for the novel non-native–practiced non-native and/or practiced non-native–
practiced native conditions should yield less activity in the regions of the cortico-BG circuit pre-
viously identified in Segawa, Tourville, et autres. (2015) in AWS, especially the left aINS, since they
will not have fully formed cohesive motor chunks for the trained speech sequences.

If, cependant, the only core deficit underlying stuttering is related to motor implementation
rather than to motor learning mechanisms, alors: (1) At the behavioral level, AWS should show
comparable performance gains over time but will display poorer (and/or slower) overall motor
performance compared to ANS; et (2) At the neural level, AWS should show a similar reduction
in activations in the aforementioned regions of the cortico-BG circuit implicated in motor
sequences learning and working memory (the left aINS, preSMA, and IFS) as ANS, but should still
show differences in primary motor (and possibly premotor) cortex; et (3) The size of the BOLD
activity difference in (at least some of ) these brain areas will covary with stuttering severity (and/or
in-scanner disfluency) since mechanisms involved in selecting and initiating successive speech
motor programs will be more impaired in AWS with more severe stuttering.

MATERIALS AND METHODS

Participants
Sixteen AWS (14 male, age range 18–42, median age 26) and fifteen ANS (13 male, age range
18–40, median age 25) were paid for participating in three testing sessions on three separate days.
Participants reported normal (or corrected-to-normal) vision and no history of hearing, speech,
langue, or neurological deficits (apart from stuttering in the AWS group). All were right-handed
(Oldfield, 1971; mean score = 80.46) and native speakers of American English with no previous
experience with any of the languages used in stimulus creation (see following text). Individuals
were excluded from taking part in the study if they were currently on medications that may
have substantial effects on neural activity, or if they had claustrophobia preventing them from
completing the MRI protocol. All participants underwent a magnetic resonance safety
screening.

An experimenter interviewed all participants to confirm the diagnosis of persistent develop-
mental stuttering in AWS and to confirm normal speech production in ANS. None of the AWS
were enrolled in a fluency-shaping program at the time of participation. The stuttering severity of
each AWS was assessed using the Stuttering Severity Instrument, Edition 4 (SSI-4; Riley, 2009). Comme
part of this assessment, each AWS was video recorded while reading aloud, conversing with an
experimenter, and speaking on the telephone. A certified speech-language pathologist then
rated the frequency and duration of the stuttering events and the presence of physical concomitants
that accompanied the moments of disfluency (par exemple., eye-blinking). Stuttering severity in the AWS
group ranged from 13 à 48, with a median of 27, and an interquartile range of 17 à 34. Five
of the 16 AWS were categorized as “very mild” (SSI-4 score <17), 1 as “mild” (18–24), 5 as “moderate” (25–31), 3 as “severe” (32–36), and 2 as “very severe” (37–46). Stimuli The speech stimuli consisted of several sets of monosyllabic pseudoword sequences (15 words per set). As shown in Table 1, all items contained either native or non-native syllable-initial (onset) and syllable-final (coda) consonant clusters. In the native sequences (e.g., “BLERK,” “THRIMF,” “TRALP”), the onset and coda clusters are phonotactically legal in English; in the non-native Neurobiology of Language 109 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 / / / / 2 1 1 0 6 1 8 9 9 6 7 7 n o _ a _ 0 0 0 2 7 p d / . l f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Neural correlates of speech motor sequence learning Table 1. stimuli used to elicit the native (left) and non-native (right) target onset and coda clusters International phonetic alphabet (IPA) transcription and orthography for experimental IPA blɚɹk bɹalk dɹalf flIsk fɹʌmp glæntʃ gɹalv klεlθ kɹeInθ plaɹθ pɹʌndʒ ʃ ɹidθ tɹælp θɹImf dwIlm kwanst gwεfθ twɚɹv θwIlb splɚɹst spɹIdθ swarf skεln stIsp Native Orthography BLERK BRALK DRALF FLISK FREMP GLANCH GRALVE KLELTH KRENGTH PLARTH PRENGE SHRIDTH TRALP THRIMF DWILM KWANST GWEFTH TWERVE THWILB SPLERST SPRIDTH SWARF SKELN STISP Phonotactics Non-native IPA fsεfk fʃ Ikp fθæmtʃ fzItʃ b vsεpʃ vðæʃ p zvεktʃ fpεstʃ ftεbstʃ ʃ kεvt ʃ tæzg vbImk vgæmʃ zbæpk zdεbg bvImpf bzInstʃ gvæzf kvætʃ k tfIpʃ tʃ tvItp bdeŋt dkεdv gbεsb kpεʃ tʃ ptætʃst tbæstf tgItk tpIpf zgεkf Orthography FSEFK FSHIKP FTHAMCH FZICHB VSEPSH VTHASHP ZVEKCH FPESCH FTEBSCH SHKEVT SHTAZG VBIMK VGAMSH ZBAPK ZDEBG BVIMPF BZINSCH GVAZF KVACHK TFIPSHCH TVITP BDANGT DKEDV GBESB KPESHCH PTACHST TBASTF TGITK TPIPF ZGEKF 110 Neurobiology of Language 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 / / / / 2 1 1 0 6 1 8 9 9 6 7 7 n o _ a _ 0 0 0 2 7 p d / . l f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Neural correlates of speech motor sequence learning sequences (e.g., “FPESCH,” “GVAZF,” “TPIPF”), the clusters are phonotactically illegal in English, but do occur in some other natural human language. The non-native clusters were taken from a variety of languages and language families including Hebrew, Leti, Taba, Romani, Polish, Lithuanian, Romanian, Georgian, Tepehua, Hungarian, and Pima; participants reported no prior experience with any languages in which these consonant clusters readily occur. None of the stimuli were an orthographic or a phonological word according to the MRC Psycholinguistic Database (Coltheart, 1981). All items containing non-native clusters had a neighborhood size of 0, and no real English words could be created by adding, deleting, or substituting a single phoneme in any subsequence. All of the clusters—both native and non-native alike—were bi- or triconsonantal and used in either onset position or coda position, but not both. Each cluster was used in only one word (i.e., no two words contained the same consonant cluster). The number of phonemes per word was counterbalanced across experimental conditions (see below). To create the prompts for the elicited production task, a female native speaker of American English was recorded producing the words. The model speaker was phonetically trained and had previously practiced producing the sequences until each stimulus could be executed fluently (i.e., without vocoid epenthesis or phoneme omissions, swaps, or substitutions). All recordings took place in a sound-attenuated booth. The speech was recorded directly to a computer using Audacity® software ( Version 2.0.3, Audacity Team) via a microphone (Samson C01U studio condenser) connected to a pre-amplifier (44.1-kHz sampling rate, 32-bit quantization). The speaker recorded multiple randomized repetitions of each token. From these repetitions, one instance of each token was selected on the basis of clarity and acoustic similarity in voice pitch (f0) to the other stimuli in the set. Using Praat software (Boersma & Weenink, 2020), all recorded tokens were digitally edited to be matched for peak intensity and duration (i.e., 480 ms) without changing f0. Procedure and Design Participants completed a cued sequence production task (see, e.g., Segawa, Masapollo, et al., 2019; Segawa, Tourville, et al., 2015). The experiment consisted of a training phase, during which participants learned to produce 15 words containing native clusters and 15 words con- taining non-native clusters, followed by a test phase, during which participants were tested on their ability to produce all 30 of the previously learned words and 15 novel words containing non-native consonant clusters. FMRI data were only collected during the test phase (see below). For both phases, participants were asked to repeat aloud each of the target words individually, which were presented both auditorily over noise-cancelling headphones and visually using text stimuli (as shown in Table 1). The training phase consisted of eight blocks of trials over two consecutive days (four on day one and four on day two). Each training block contained eight repetitions of each word, for a total of 240 trials per block. The test phase consisted of eight blocks of trials performed on day three (inside the MRI scanner) after completing the training phase (outside the scanner on days one and two). Each test block contained five to six repetitions of each word, for an average of 40 trials per block, from three stimulus categories (conditions): (1) practiced native words (i.e., words comprising native clusters that were previously encoun- tered in the training phase), (2) practiced non-native words (i.e., words comprising non-native clusters that were previously encountered in the training phase), and (3) novel non-native words (i.e., novel words with novel clusters that were not encountered in the training phase). Participants were divided into four groups, each of which practiced producing a different subset of the native and non-native sequences during training. The non-native words that were not practiced during training were used as novel non-native words during the imaging session. Neurobiology of Language 111 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 / / / / 2 1 1 0 6 1 8 9 9 6 7 7 n o _ a _ 0 0 0 2 7 p d . / l f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Neural correlates of speech motor sequence learning Assignment of non-native words to the practiced non-native and novel non-native categories was counterbalanced across participants. The training phase occurred one to two days before the test phase to allow for sleep-mediated memory consolidation of the newly acquired motor traces (see, e.g., Doyon, Albouy, et al., 2015; Doyon, Bellec, et al., 2009; Fenn et al., 2003; Vahdat et al., 2017; cf. Brawn et al., 2010; Pan & Rickard, 2015). During training, participants were seated in a chair in front of a laptop (IBM Lenovo ThinkPad X61s) computer screen in a sound-treated laboratory room that was dimly lit. The auditory speech stimuli were presented over headphones (Sennheiser, HD 280 Pro) at a comfortable listening level, and utterances produced by the participants were recorded with a Samson (Hauppauge, NY) C01U USB studio condenser microphone connected to the computer via a MOTU micro- book audio interface. Utterances were recorded using MATLAB (MathWorks Inc., Natick, MA) at 44.1 kHz. During testing, participants laid supine in an MRI scanner. Instructions and visual stim- uli were projected onto a screen viewed from within the scanner via a mirror attached to the head coil. The auditory stimuli were played over Sensimetrics model S-14 MRI-compatible earphones. Participants’ productions were transduced by a Fibersound model FOM1-MR-30m fiber-optic microphone, sent to a Lenovo ThinkPad X61s, and recorded using MATLAB at 44.1 kHz. The trial structure was identical during training and testing. First, the orthographic display of a given syllable was centrally presented in tandem with its corresponding auditory prompt. Participants heard each prompt only once on each trial. Then, after the offset of the auditory presentation, a tone was presented for 50 ms. The time between stimulus offset and tone onset was randomly jittered between 500 and 1,000 ms. This tone served as a “go” signal that prompted the participant to go ahead and repeat the token as clearly and accurately as possible. For all phases, participants produced the target syllables in a pseudo-random order. The combination of the auditory and orthographic presentations was necessary because prior studies have shown that listeners tend to perceive non-native consonant clusters as epenthesized disyllabic sequences (e.g., Berent et al., 2007; Davidson & Shaw, 2012; Dupoux, Kakehi, et al., 1999; Dupoux, Parlato, et al., 2011; Pitt, 1998). Moreover, in another study that directly examined the effects of stimulus input modality (audio only vs. audio and text) on speakers’ ability to pro- duce non-native consonant clusters (Davidson, 2010), it was found that the presence of text led to an improvement in overall task performance. For both phases, participants were instructed to repeat the target syllable as clearly and accu- rately as possible as soon as they heard the tone. Several familiarization trials with experimenter feedback were included at the start of the experiment to confirm that participants understood the task instructions and were able to perform the task. The sequences used during these initial prac- tice trials were not used at any point in the rest of the study. Feedback was not provided during the actual training or test phases. Behavioral Data Analyses We conducted several analyses to provide evidence, at the behavioral level, that incremental motor learning occurred throughout the two days of speech motor practice (outside the MRI scanner), and that such learning was retained at test (inside the scanner). In the first analysis, we analyzed three temporally sensitive measures of learning across the eight training blocks (on days 1 and 2): (1) articulatory sequencing error rate, (2) utterance duration (interval from utterance onset to offset), and (3) reaction time (RT) (interval from the offset of the go-signal to utterance onset). These measures are generally believed to quantify the ease or difficulty with which speakers retrieve and execute speech sequences (Sternberg et al., 1978) and are com- monly used in both the motor sequence learning and second language learning literatures as Neurobiology of Language 112 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 / / / / 2 1 1 0 6 1 8 9 9 6 7 7 n o _ a _ 0 0 0 2 7 p d / . l f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Neural correlates of speech motor sequence learning measures of learning extent (see, e.g., Brawn et al., 2010; Buchwald et al., 2019; Doyon, Albouy, et al., 2015; Nakamura et al., 1998; Rauschecker et al., 2008; Segawa, Masapollo, et al., 2019). Sequencing errors were defined as phoneme additions (including schwa insertions), deletions, and substitutions, and utterance repetitions, omissions, and restarts. Based on prior findings reported with neurotypical speakers (Segawa, Masapollo, et al., 2019; Segawa, Tourville, et al., 2015), we expected to observe greater learning (at least in the ANS group) for the non-native sequences because those sequences included both novel words and novel onset and coda clusters, whereas the native sequences included novel words with familiar onset and coda clusters. In the second analysis, to provide evidence of the retention of learning at test, we examined the same three performance measures for each experimental group (ANS vs. AWS) and condition (practiced native vs. practiced non-native vs. novel non-native) during the test phase (on day 3). For each word production, each measure was calculated following the removal of noise associ- ated with the scanner bore echo and peripheral equipment using a Wiener filter (Wiener, 1949). The coder was blind to the condition (practiced vs. novel) of the non-native syllables. We hypoth- esized that if participants formed regularized motor chunks for the previously trained speech sequences, then they should perform more accurately (and/or faster) at generating the practiced native and non-native sequences compared to the novel non-native sequences. In a final analysis, we examined whether, and if so, how, speech motor practice influenced fluency in the AWS group. Toward this end, we compared the mean number of trials perceptually coded as containing at least one stuttering-like disfluency during both the training and test phases (for the AWS group only) for each stimulus condition ( practiced native vs. practiced non-native vs. novel non-native). Disfluencies were determined by a certified speech-language pathologist. For each of these analyses, custom MATLAB software was used to perceptually rate and acous- tically measure onsets and offsets of syllables by viewing the waveform and spectrogram and listening to the audio files. Listeners first marked, blinded to experimental condition, whether the target phoneme sequence was produced fluently. Disfluent productions were further catego- rized into four possible disfluency types: (1) repetition (i.e., a phoneme or syllable was repeated), (2) prolongation (i.e., a phoneme was prolonged), (3) audible or silent blocking (involuntary filled or unfilled pauses in speaking), and (4) clustered (i.e., more than one stutter type was produced for a given sequence). Each utterance was then marked for four possible sequencing error subtypes: (1) approxima- tion of the target (i.e., phoneme deletion/omission, insertion, substitution, vowel epenthesis); (2) unrecognizable from the target (i.e., an entirely different sequence was produced); (3) unfinished word (i.e., the sequence produced was not completed before the end of the recording window); and (4) silence (i.e., no sequence was produced at the time of recording). Since we were con- cerned with the learning of non-native phonotactics rather than subphonemic allophonic details, productions from either the model speaker or the participants were not judged on how natural they sounded in the language from which they were derived. In addition, it is important to note that it was impossible to reliably distinguish involuntary “silent” pauses or blocks associated with stuttering from intentional speech onset delays with the audio-only recordings that we obtained in the scanner (although there were constraints on what the delay could be since the length of the recording window was fairly short). Thus, we suspect that trials with stuttering pauses were categorized as sequencing error subtype 4 (i.e., no sequence produced). This is a limitation inherent in the current study and all studies of this kind in stuttering (unless video recordings of articulatory behavior inside the scanner are obtained). In order to explore the possibility that some trials containing “silent” stuttering blocks were included in the neuroimaging analyses, we plotted histograms showing the distribution of RT scores for Neurobiology of Language 113 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 / / / / 2 1 1 0 6 1 8 9 9 6 7 7 n o _ a _ 0 0 0 2 7 p d / . l f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Neural correlates of speech motor sequence learning each stuttering speaker in the test phase (inside the scanner). Critically, these plots did not reveal a bimodal distribution with one peak having frequency values clustered to the right. This finding suggests that any potential blocks or pauses were likely categorized as silent errors (sequencing subtype 4 as described above) and were excluded from the imaging analyses, in which case, the BOLD responses would not have been contaminated by potential delays in RT. See the Supplementary Materials in the online supporting information located at https://www .mitpressjournals.org/doi/suppl/10.1162/nol_a_00027 for further details. Mean disfluency rates for each subject were calculated as the percentage of trials that contained one or more disfluency error subtypes, and mean error rates for each subject were calculated as the percentage of trials that contained one or more error subtypes. For each produc- tion containing no sequencing or disfluency errors, utterance onset and offset were automatically labeled based on sound pressure level thresholds, then hand-checked. Note that, in the neuro- imaging component of the study (described below), only trials in which participants produced the target sequence accurately and fluently were analyzed. FMRI Paradigm In addition to the three speaking conditions ( practiced native vs. practiced non-native vs. novel non-native), a silent baseline condition was intermixed during imaging in which participants viewed a series of asterisks on the screen instead of the orthographic stimulus and rested quietly instead of uttering a word. FMRI data were acquired using a sparse sampling protocol (see Belin et al., 1999; Gracco et al., 2005; Perrachione & Ghosh, 2013) that allowed participants to pro- duce the target syllables during silent intervals between volume acquisitions. A single volume was acquired approximately 4 s after speech onset on each trial which aligns with the 4–6 s delay in peak BOLD response onset (Belin et al., 1999). By scanning between speech productions, this protocol avoids the influence of scanner noise on speaker performance and brain activity responses and image artifacts resulting from speech-induced motion of the head. The cued sequence paradigm was identical to that used during training (outside of the scanner) except with an additional pause after the production of each syllable to temporally align the image acquisition to the expected peak of the hemodynamic response. As previously described, the test phase consisted of eight blocks of trials which corresponded to eight functional runs. A single volume was recorded on each trial and the delay between volumes was approximately 10 s. Each functional run contained five or six productions of each stimulus item, for a total of 40 trials per run (each run lasted approximately 6–7 min). This resulted in a total of 320 test trials (80 trials per condition, including baseline). Conditions were shuffled and pseudorandomly distributed across the eight functional runs with at least eight instances of each condition appearing in each run. Image Acquisition All neuro-imaging data were acquired using a 3-Tesla Siemens TIM Trio scanner, equipped with a 32-channel head coil. T2*-weighted gradient echo-planar fMRI images were collected to assess BOLD responses during the test phase. Forty-one horizontal slices were collected in each functional volume (in-plane resolution = 3.1 mm2, slice thickness = 3 mm, gap = 25%, acquisition time = 2.5 s, echo time [TE] = 20 ms); volumes were automatically aligned to the anterior commissure–posterior commissure line. Prior to collecting functional data, a gradient- echo field map sequence was collected; the resulting magnitude and phase images enabled offline correction of magnetic field distortions in functional images during data preprocessing (see below; Jezzard & Balaban, 1995). Structural images were collected using a T1-weighted Neurobiology of Language 114 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 / / / / 2 1 1 0 6 1 8 9 9 6 7 7 n o _ a _ 0 0 0 2 7 p d . / l f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Neural correlates of speech motor sequence learning multi-echo MPRAGE pulse sequence (MEMPRAGE, voxel size = 1 mm3, 176 axial slices, 256 × 256 field of view, repetition time [TR] = 2,530 ms, TE = 3.44 ms, flip angle = 7°). FMRI Data Analysis Image preprocessing Functional data were processed using tools from the following software packages that were inte- grated into a processing stream within SPM12 (Statistical Parametric Mapping, v12; www.fil.ion .ucl.ac.uk/spm/): FreeSurfer (Dale et al., 1999; Fischl, Salat, et al., 2002; Fischl, Sereno, et al., 1999; www.freesurfer.net), Artifact Detection Tools (ART; www.nitrc.org/projects/artifact _detect/), and the CONN toolbox (Whitfield-Gabrieli & Nieto-Castañón, 2012). Freesurfer was used to remove non-brain components of the T1 structural volumes; segment the brain into gray matter, white matter, and cerebral spinal fluid components; generate a reconstruction of the cortical surfaces of each hemisphere; and identify cortical and subcortical regions of interest (ROIs) (see below). Functional data were preprocessed through two pipelines: a surface/vertex- based pipeline for analysis of cortical responses and a volume/voxel-based pipeline for analysis of subcortical basal ganglia and cerebellar responses. Prior to pre-processing, the first volume of each functional series was removed because it served only as a trigger for the initial experimental trial. A surface/vertex-based analysis pipeline was used to assess BOLD response differences for each experimental group and for each contrast in the cerebral cortex. Functional images from each subject were simultaneously realigned to the mean subject image and unwarped (motion- by-inhomogeneity interactions) using the SPM12 realign and unwarp procedure (Andersson et al., 2001). Outlier scans were detected using ART based on framewise displacement (scan- to-scan motion threshold of 0.9 mm) and mean signal change (scan-to-scan signal change threshold of 5 standard deviations above the mean (see Nieto-Castañón, 2020, for details). Framewise displacement was computed at each timepoint by considering a 140 × 180 × 115 mm bounding box around the brain and estimating the largest displacement among six control points placed at the center of the bounding-box faces. Global BOLD signal change was computed at each timepoint as the change in average BOLD signal within SPM’s global-mean mask scaled to standard deviation units. Functional volumes from each subject were then coregistered with their high-resolution T1 structural images and resliced using SPM12’s inter-modal registration proce- dure with a normalized mutual information objective function. The functional data were then resampled at the location of the FreeSurfer fsaverage level-8 tessellation (163,842 vertices and 327,680 faces) projected on each subject-specific cortical surface, averaged across 10 intervals along the normal between the white matter and pial surfaces, and smoothed using iterative diffusion smoothing with a series of 40 discrete steps, approximately equivalent to a 8 mm full- width half-maximum (FWHM) two-dimensional Gaussian smoothing kernel (Hagler et al., 2006; Nieto-Castañón, 2020). A volume/voxel-based analysis pipeline was used to identify differences in subcortical BOLD responses. Following the realignment, unwarping steps described above, functional volumes, and the original T1 structural volumes, were simultaneously segmented and normalized directly to Montreal Neurological Institute (MNI) space using SPM12’s combined normalization and segmentation procedure (Ashburner & Friston, 2005). Prior to MNI-normalization, both the func- tional and anatomical volume origins were centered to coordinates [0, 0, 0] in order to improve the quality of the iterative procedure initial starting estimates. Functional volumes were then spatially smoothed using an 8 mm FWHM Gaussian kernel in order to increase BOLD signal- to-noise ratio and reduce the influence of residual variability in functional and gyral anatomy across subjects (Nieto-Castañón, 2020). Neurobiology of Language 115 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 / / / / 2 1 1 0 6 1 8 9 9 6 7 7 n o _ a _ 0 0 0 2 7 p d . / l f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Neural correlates of speech motor sequence learning Subject-level BOLD contrast analyses After preprocessing, BOLD responses were estimated using a general linear model in SPM12. Because functional volumes were acquired in a sparse sequence, the hemodynamic response function for each stimulus event was modeled as a finite impulse response. The model included four condition-specific variables (practiced native, practiced non-native, novel non-native, and baseline). Trials with productions that were perceptually rated as containing either a sequencing error (e.g., phoneme deletions, insertions or substitutions) or a stuttering-like disfluency (e.g., pho- neme/syllable repetitions, prolongations, or blocks), or were found to be outliers by ART were modeled as separate conditions (one individual regressor per trial) thereby removing variability resulting from these trials from the effects- and contrast-of-interest estimates. For each individual run, regressors were added to the model to remove linear effects of time (e.g., signal drift, ad- aptation) and the six motion covariates (taken from the realignment step) and a constant term. The model was estimated at each vertex (surface-based) analysis or voxel (volume-based) anal- ysis for each participant, resulting in two surface maps (one for each hemisphere) and a volume map of the model regressor coefficients for each condition. These condition estimates were then contrasted to yield effect-size maps of the following contrasts of interest: differences in the response during novel non-native and practiced non-native conditions (novel non-native− practiced non-native), and differences in the response during the practiced non-native and practiced native conditions ( practiced non-native−practiced native). Group-level analyses Group-level t statistics were calculated separately for each contrast map. Cluster-level inferences were based on Threshold Free Cluster Enhancement (TFCE; Smith & Nichols, 2009) with default parameters extent = 0.5 and height = 2. This method assigns TFCE scores to each voxel characterizing the amount of cluster-like local spatial support at each location. These scores are then compared to their expected distribution under the null hypothesis, estimated using 1,000 randomization/permutation iterations, and the results are thresholded at family-wise error (FWE) false positive probability ( pFWE) < 0.025. This threshold was used as a conservative means to ensure a cluster-level pFWE < 0.05 across the cortical surface and subcortical volume analyses. Region-of-Interest-Based Analyses To increase statistical power and sensitivity, we supplemented the surface/vertex-based and volume/voxel-based analyses with ROI analyses (Nieto-Castañón et al., 2003) based on a priori hypotheses derived from the results of Segawa, Tourville, et al. (2015). Specifically, we used anatomically defined ROIs in the speech production network that overlapped areas of significant BOLD activity in Segawa, Tourville, et al. (2015) for each contrast of interest. The predefined cortical and subcortical (basal ganglia and cerebellar) ROIs are listed in Table 2; a detailed description of the anatomical landmarks used for cortical parcellation are provided in Tourville and Guenther (2003). For the novel non-native−practiced non-native contrast, we tested whether brain activity was significantly different in globus pallidus internal (GPi) and external (GPe) segment ROIs; these ROIs were derived from the probabilistic atlas of the basal ganglia described by Keuken et al. (2014). For the practiced non-native−practiced native con- trast, we tested whether brain activity was significantly different in right cerebellum lobule VI, which was derived from the SUIT probabilistic atlas of the cerebellum (Diedrichsen et al., 2009). Affirmation of differences in these ROIs would replicate the Segawa, Tourville, et al. (2015) findings with neurotypical speakers. Neurobiology of Language 116 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 / / / / 2 1 1 0 6 1 8 9 9 6 7 7 n o _ a _ 0 0 0 2 7 p d . / l f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Neural correlates of speech motor sequence learning Table 2. in the current analyses List of the predefined cortical, subcortical, and cerebellar regions-of-interest (ROIs) used Novel non-native−practiced non-native contrast Subcortical ROIs L GPi L GPe Practiced non-native−practiced native contrast Subcortical ROIs R cerebellum lobule VI Brain-behavior motor learning correlation analyses Cortical ROIs L aINS R aINS L PT R preSMA L vPMC L pSTS L midPMC L aFO R aFO L pIFS L STG L STS Subcortical ROIs L GPi L GPe Brain-behavior stuttering severity correlation analyses (AWS only) Cortical ROIs L vPMC + midPMC + vMC + midMC L preSMA + SMA Neurobiology of Language 117 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 / / / / 2 1 1 0 6 1 8 9 9 6 7 7 n o _ a _ 0 0 0 2 7 p d . / l f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Neural correlates of speech motor sequence learning Table 2. (continued ) Subcortical ROIs L GPi L GPe L putamen L caudate L VA L VL Brain-Behavior Correlation Analyses Two types of hypothesis-driven analyses were conducted to identify potential relationships between behavioral measures and brain activity. In the first type of analysis, we conducted multiple correlation tests to identify relationships between behavioral measures of motor sequence learning success and BOLD activity. Specifically, we tested for correlations between (i) the mean BOLD response in the cortical and subcortical ROIs that showed task-activated clusters in Segawa, Tourville, et al. (2015; Table 2) in both the novel non-native−practice non-native and the practiced non-native−practiced native contrasts and (ii) each of the three motor learning indices (error rate, utterance duration, and RT). No correction was applied for the number of ROIs in this ROI list. FreeSurfer was used to define the ROIs on each individual cortical surface using the labeling system described in Cai et al. (2014). These ROIs included the left-lateralized aINS, vPMC, anterior frontal operculum (aFO), preSMA, IFS, posterior superior temporal sulcus (pSTS), and planum temporale (PT), as well as the left GPi and GPe. For each participant, we computed each motor learning index by subtracting the mean performance difference between the novel non-native and practiced non-native productions. We normalized for differences between participants by dividing these differences by the prac- ticed non-native syllable measure, on a per-participant basis. For example, the utterance dura- tion learning measure was the mean duration difference between the novel non-native and practiced non-native productions divided by mean duration of the practiced non-native pro- ductions. Each motor learning index was then correlated with the mean beta coefficient within each significant cluster from each contrast. In the second type of analysis, SSI measures (Riley, 2009) of stuttering frequency and overall severity (described above) were used, in addition to the number of disfluencies produced during the test phase (inside the scanner), to identify potential relations between stuttering- related behaviors and brain activity. Specifically, we tested for correlations in the AWS group only with the mean composite SSI scores and mean disfluency rates during test (inside the scanner) and the mean BOLD responses in eight hypothesis-based ROIs within the cortico- BG loops (listed in Table 2) in the novel non-native−practice non-native contrast (based on fluent trials only). These ROIs included left premotor and primary motor areas (ventral and mid premotor cortices [vPMC, midPMC] + ventral and mid motor cortices [vMC, midMC]), left medial prefrontal areas (left SMA + preSMA), as well as several left-lateralized subcortical basal ganglia sites (GPi, GPe, caudate, putamen, ventral anterior nucleus [VA], ventral lateral nucleus [VL]). Ventral premotor cortex (vPMC): Brain region that is critically important in programming speech movements. Neurobiology of Language 118 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 / / / / 2 1 1 0 6 1 8 9 9 6 7 7 n o _ a _ 0 0 0 2 7 p d . / l f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Neural correlates of speech motor sequence learning RESULTS Behavioral Measures of Speech Motor Sequence Learning Training phase Table 3 shows a summary of the behavioral results for each phase of the experiment and for each group (ANS vs. AWS). Our first set of analyses examined, at the behavioral level, whether partic- ipants showed evidence of incremental speech motor sequence learning over the course of the training phase by examining the time course of improvement in each of the three performance measures—mean error rates, utterance durations, and RTs—over the two days of speech motor practice. Each performance measure was averaged within each group, training block, condition, and participant. Again, duration and RT measures were only analyzed for utterances that were perceptually coded as having been properly executed (see above) on each day. To visualize the running estimate of the learning trajectories across participants, Figure 1 plots the time course of improvement of each performance measure as a function of group, training block, and condition. Separate analyses of variance (ANOVAs) were performed on the mean sequencing error rates, durations, and RTs with training block (1–8) and condition (native vs. non-native) as within- subjects factors and group (ANS vs. AWS) as a between-subjects factor. In these and all subse- quent ANOVAs, Greenhouse-Geisser corrections were applied when appropriate and partial eta-squared effect sizes were calculated for all main effects and interactions. Post-hoc pairwise comparisons were reported as significant at the 0.05 level. 2 = 0.878] and block [F(7, 203) = 15.985, p < 0.001, (cid:1) p The ANOVA performed on mean error rates (shown in Figure 1A) revealed a significant main effect of group [F(1, 29) = 5.036, p = 0.033, (cid:1) 2 = 0.148], such that AWS [mean (M ) = 32.7, stan- p dard deviation (SD) = 13.6] produced more sequencing errors than ANS [M = 23.0, SD = 10.3] in general. There were also highly significant main effects of condition [F(1, 29) = 208.822, p < 0.001, (cid:1) 2 = 0.355]. There was one sig- p nificant interaction, the stimulus condition × block interaction [F(7, 203) = 8.863, p < 0.001, (cid:1) 2 = p 0.234]. All other interactions were not significant [p > 0.05, in all cases]. Simple effects tests on
the condition × block interaction revealed that there was a significant effect of block on both the
native sequences [F(7, 210) = 2.958, p = 0.045, (cid:1)
2 = 0.090] and the non-native sequences [F(7,
p
210) = 20.495, p < 0.001, (cid:1) 2 = 0.406]. However, the effect size was considerably larger in the p non-native condition, indicating that sequence learning was greater for novel words with unfa- miliar consonant clusters than novel words with familiar consonant clusters during training. This is in keeping with the behavioral results reported by Segawa and colleagues (Segawa, Tourville, et al., 2015; Segawa, Masapollo, et al., 2019) which found that relatively less learning occurred for native compared to non-native sequences. The ANOVA performed on mean utterance duration (shown in Figure 1B) revealed a main effect of group [F(1, 29) = 8.963, p = 0.006, (cid:1)p 2 = 0.236], such that ANS produced shorter utter- ances [M = 0.59, SD = 0.05] than AWS [M = 0.69, SD = 0.10]. There was also a significant effect of stimulus condition [F(1, 7) = 34.318, p < 0.001, (cid:1) 2 = 0.542], such that the native sequences p [M = 0.61, SD = 0.09] were uttered faster than the non-native sequences [M = 0.67, SD = 0.10], as well as a condition × block interaction [F(7, 203) = 2.066, p = 0.049, (cid:1) 2 = 0.066]. All other p interactions were not significant [p > 0.05, in all cases]. Consistent with our expectations based
on our prior studies, simple effects tests on the condition × block interaction revealed that there
was a significant effect of block on the non-native sequences [F(7, 210) = 3.340, p = 0.002, (cid:1)
2 =
p
0.100] but not on the native sequences [F(7, 210) = 1.038, p = 0.382, (cid:1)
2 = 0.033]. Ainsi,
p
participants only got reliably faster at executing novel words during training if they contained
unfamiliar consonant clusters, likely because they were near ceiling performance for novel
words using native clusters.

Neurobiology of Language

119

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
n
o

je
/

un
r
t
je
c
e
–
p
d

F
/

2
1
1
0
6
1
8
9
9
6
7
7
n
o
_
un
_
0
0
0
2
7
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

Neural correlates of speech motor sequence learning

Dependent measure: Error rates

Training phase

Tableau 3.

ANOVA results for behavioral analyses

Test phase

Effect

Block

Condition

Groupe

Block × Condition

Block × Group

Condition × Group

Block × Condition × Group

2
p

Effect

η
2
p

15.985

208.822

5.036

8.863

1.554

1.137

1.312

<0.001 0.355 Block <0.001 0.878 Condition 0.033 0.148 Group 5.291 91.164 1.371 7 2 1 <0.001 0.159 <0.001 0.765 0.251 0.047 <0.001 0.234 Block × Condition 1.504 14 0.106 0.051 0.183 0.051 Block × Group 0.295 0.038 Condition × Group 1.164 0.838 7 2 0.325 0.04 0.438 0.029 0.268 0.043 Block × Condition × Group 0.647 14 0.825 0.023 Dependent measure: Utterance duration Effect Block Condition Group Block × Condition Block × Group Condition × Group Block × Condition × Group F df p η 2 p Effect F df p η 2 p 2.278 34.318 8.963 2.066 0.132 0.506 1.358 7 1 1 7 7 1 7 0.09 0.073 Block <0.001 0.542 Condition 0.006 0.236 Group 2.638 40.733 0.112 7 2 1 0.033 0.083 <0.001 0.584 0.74 0.004 0.049 0.066 Block × Condition 1.118 14 0.354 0.037 0.931 0.005 Block × Group 0.483 0.017 Condition × Group 0.517 0.949 7 2 0.738 0.018 0.357 0.032 0.25 0.045 Block × Condition × Group 1.059 14 0.388 0.035 Dependent measure: Reaction time Effect Block Condition Group Block × Condition Block × Group Condition × Group Block × Condition × Group F df p ηp 2 Effect F df p ηp 2 16.742 0.487 1.934 1.938 0.36 3.732 1.099 7 1 1 7 7 1 7 <0.001 0.366 Block 0.491 0.017 Condition 0.175 0.063 Group 3.114 1.688 5.926 7 2 1 0.039 0.097 0.2 0.054 0.021 0.17 0.094 0.063 Block × Condition 1.012 14 0.376 0.034 0.807 0.012 Block × Group 0.063 0.114 Condition × Group 0.851 4.973 7 2 0.455 0.029 0.012 0.146 0.363 0.037 Block × Condition × Group 0.805 14 0.664 0.027 Neurobiology of Language 120 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 / / / / 2 1 1 0 6 1 8 9 9 6 7 7 n o _ a _ 0 0 0 2 7 p d / . l f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Neural correlates of speech motor sequence learning Table 3. (continued ) Dependent measure: Disfluency rates (AWS only) Training phase Test phase Effect Block Condition F 2.618 1.379 df 7 2 p η 2 p Effect 0.084 0.158 Block 0.26 0.09 Condition F 1.095 9.959 df 7 2 p η 2 p 0.372 0.068 0.001 0.399 Block × Condition 2.038 14 0.108 0.127 Block × Condition 0.923 14 0.482 0.058 Note. Shown are the F value, the degrees of freedom, p value, and partial-eta-squared value for each effect. The ANOVA performed on mean RT scores (shown in Figure 1C) revealed a highly significant effect for block [F(7, 203) = 16.742, p < 0.001, (cid:1) 2 = 0.366], such that participants got faster at p initiating their utterances during the training sessions. The effects of group, condition, and all other interactions were not significant [p > 0.05 in all cases].

In an additional analysis, we examined the mean disfluency rates for the AWS group only.
These are shown in Figure 2 averaged across each sequence type (native vs. non-native) et
block (1–8). An ANOVA on these scores—sequence type × block—showed no significant main
effects or interaction [p > 0.05 in all cases], indicating that stuttering speakers were equally
disfluent for both sequence types throughout training.

Test phase

Our second set of analyses examined the same set of behavioral measures of sequence generation
while participants were inside the MRI scanner during the test phase. Each performance measure
was averaged within each group, testing block, condition, and participant. Duration and RT
measures were again only analyzed for utterances coded as properly sequenced productions
(see above) on each day. Tableau 4 shows the mean error rates during the test phase as a function

Chiffre 1. Time course of improvement of behavioral performance measures during training (outside the scanner) for the practiced native and
practiced non-native sequences as a function of training block (1–8) and group (ANS vs. AWS). (UN) Mean percentage of errors of each se-
quence type; (B) Mean durations of the properly executed utterances of each sequence type; (C) Mean reaction times of the properly executed
utterances of each sequence type. The shaded area represents the standard error of the mean.

Neurobiology of Language

121

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
n
o

je
/

un
r
t
je
c
e
–
p
d

F
/

2
1
1
0
6
1
8
9
9
6
7
7
n
o
_
un
_
0
0
0
2
7
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

Neural correlates of speech motor sequence learning

Chiffre 2. Mean percentage of disfluency errors for the practiced native and practiced non-native
sequences during training (outside the scanner) as a function of training block (1–8) for AWS only.
The shaded area represents the standard error of the mean.

of error subtype and sequence type. Chiffre 3 shows the mean error rates, utterance duration, et
RT scores as a function of group, test block, and condition. Separate ANOVAs were performed on
the mean error rates, utterance durations, and RTs with group (AWS vs. ANS) as a between-
subjects factor, and test block (1–8) and experimental condition ( practiced native vs. practiced
non-native vs. novel non-native) as within-subjects factors.

ANOVA performed on mean error rates (shown in Figure 3A) revealed significant main effects
of block [F(7, 196) = 5.291, p < 0.001, (cid:1)2 = 0.159] and condition [F(2, 56) = 91.164, p < 0.001, (cid:1)2 = 0.765]. There was no significant main effect of group [F(1, 28) = 1.371, p = 0.251, (cid:1)2 = 0.092] or interaction effects [p > 0.05 in all cases]. Post-hoc t-test comparisons performed on the main
effect of condition indicated that, regardless of group, the mean error rates for the practiced native
syllables [M = 4.0. SD = 7.2] were significantly lower than the practiced non-native syllables [M =
34.8, SD = 20.2, t(30) = −9.357, p < 0.001, Cohen’s d = 2.03] and novel non-native syllables [M = 44.7, SD = 21.0, t(30) = −10.978, p < 0.001, d = 2.59], and the mean error rates for the practiced Table 4. Mean error rates in the test blocks (inside the scanner) by error subtype and sequence type Sequencing errors Approximation of target Unrecognizable from target Unfinished utterance Silence Practiced native 14.6 0.0 0.0 0.0 Neurobiology of Language ANS Sequence type Practiced non-native 30.7 Novel non-native 39.4 Practiced native 32.5 AWS Sequence type Practiced non-native 33.3 0.0 0.0 0.0 0.0 0.0 0.0 0.3 0.2 0.6 0.1 2.8 1.8 Novel non-native 37.6 0.3 2.3 2.7 122 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 / / / / 2 1 1 0 6 1 8 9 9 6 7 7 n o _ a _ 0 0 0 2 7 p d . / l f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Neural correlates of speech motor sequence learning Figure 3. Behavioral performance measures during test (inside the scanner) for the practiced native, practiced non-native, and novel non- native sequences as a function of test block (1–8) and group (ANS vs. AWS). (A) Mean percentage of errors of each sequence type; (B) Mean durations of the properly executed utterances of each sequence type; (C) Mean reaction times of the properly executed utterances of each sequence type. The shaded area represents the standard error of the mean. non-native syllables were lower than the novel non-native syllables [t(30) = −4.653, p < 0.001, d = 0.48]. Measures of mean utterance duration (shown in Figure 3B) patterned similarly. ANOVA performed on mean utterance durations indicated that there were also significant main effects of block [F(7, 203) = 2.638, p = 0.033, (cid:1)2 = 0.083] and condition [F(2, 58) = 40.733, p < 0.001, (cid:1)2 = 0.584]. Again, there was no significant main effect of group [F(1, 29) = 0.112, p = 0.740, (cid:1)2 = 0.004] or interaction effects [p > 0.05 in all cases]. Post-hoc t-test comparisons
performed on the main effect of condition indicated that, regardless of group, participants were
faster at executing the practiced native syllables [M = 0.67, SD = 0.15] than the practiced
non-native syllables [M = 0.74, SD = 0.16, t(30) = −6.797, p < 0.001, d = 0.45] and novel non-native syllables [M = 0.77, SD = 0.19, t(30) = −7.454, p < 0.001, d = 0.58]. In addition, participants were faster at executing the practiced non-native syllables than the novel non- native syllables [t(30) = −3.044, p = 0.005, d = 0.17]. ANOVA performed on mean RT scores (shown in Figure 3C) showed no main effect of con- dition [F(2, 58) = 1.688, p = 0.200, (cid:1) 2 = 0.054]. There were, however, significant main effects of p block [F(7, 203) = 3.114, p = 0.039, (cid:1) 2 = 0.097] and group [F(1, 29) = 5.926, p = 0.021, (cid:1) 2 = p p 0.170], such that AWS took longer to initiate their utterances [M = 0.71, SD = 0.36] than ANS [M = 0.48, SD = 0.16]. The condition × group interaction effect also reached statistical signif- icance [F(2, 58) = 4.973, p = 0.012, (cid:1) 2 = 0.146]. Simple effects tests on the condition × group p interaction indicated that the effect of condition approached significance in AWS [F(2, 30) = 3.555, p = 0.053, (cid:1) 2 = 0.037]. All p other interactions were not significant [p > 0.05 in all cases].

2 = 0.369], but not in ANS [F(2, 28) = 0.540, p = 0.589, (cid:1)
p

Enfin, we again examined mean disfluency rates for the AWS group while in the scanner.
These scores are shown in Figure 4. ANOVA performed on these scores—sequence type ×
block—revealed a highly significant main effect of condition [F(2, 30) = 9.959, p = < 0.001, (cid:1)p 2 = 0.399]. There was no significant main effect of block or interaction effect [p >
0.05 in both cases]. Post-hoc t-test comparisons performed on the main effect of condition
indicated that AWS were less disfluent when executing the practiced native syllables [M =
4.6, SD = 11.8] compared to the practiced non-native syllables [M = 7.5, SD = 15.6, t(15) =
−2.442, p = 0.027, d = 0.32] and novel non-native syllables [M = 9.8, SD = 19.4, t(15) =
−3.476, p = 0.003, d = 0.49]. As well, AWS were less disfluent when producing the practiced
non-native than the novel non-native syllables [t(15) = −2.307, p = 0.036, d = 0.19].

Neurobiology of Language

123

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
n
o

je
/

un
r
t
je
c
e
–
p
d

F
/

2
1
1
0
6
1
8
9
9
6
7
7
n
o
_
un
_
0
0
0
2
7
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

Neural correlates of speech motor sequence learning

Chiffre 4. Mean percentage of disfluency errors (inside the scanner) for the practiced native, prac-
ticed non-native, and novel non-native sequences during test (inside the scanner) as a function of
test block (1−8) for AWS only. The shaded area represents the standard error of the mean.

En résumé, we obtained similar results to those in our prior studies on speech motor sequence
learning with neurotypical speakers (Segawa, Tourville, et coll., 2015; Segawa, Masapollo, et coll.,
2019): Participants—ANS and AWS alike—showed incremental improvements in performance
speed and accuracy, especially for the non-native sequences, with repetition and practice during
training and these gains were maintained overnight between experimental sessions. De plus,
the data indicated that, after two days of extended speech motor practice, speakers from both
groups showed further performance gains in accuracy (reduced error rate) for the practiced
non-native sequences throughout the test phase (Figure 3A). Although both groups showed com-
parable gains, the AWS produced the sequences slower and/or less accurately prior to and after
entraînement. Dans l'ensemble, these findings provide evidence at the behavioral level that AWS do not show
deficits in the acquisition or retention of new speech motor sequences. Enfin, RT scores were
notably higher during test (inside the scanner) than during training (outside of the scanner), surtout-
cially for the AWS group. Given that the experimental paradigm during the test phase was iden-
tical to that used during the training phase, this difference in initiation speed likely reflects
differences in performance anxiety outside versus inside the scanner.

Neural Correlates of Speech Motor Sequence Learning

FMRI analysis

Before performing direct group comparisons, we first assessed differences between the condition-
specific brain activations using the pooled results (ANS and AWS combined). For each of the two
speech contrast conditions, we report the results from one-sided (positive only) tests. Figure 5A and
Tableau 5 show the brain regions that were significantly more active during the execution of novel
non-native than practiced non-native syllables (TFCE cluster-level pFWE < 0.05) aggregated across both experimental groups. The surface/vertex-based analyses revealed that the production of novel non-native syllables resulted in greater BOLD response in preSMA, aINS, aFO, pFO, and IPS bilat- erally. In the left hemisphere, additional cortical clusters were noted with peaks in IFS, vPMC, pos- terior inferior frontal gyrus (pIFG), midPMC, vMC, midMC, superior temporal gyrus (STG), and inferior temporal occipital (ITO) region. The volume/voxel-based analysis found no statistically Neurobiology of Language 124 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 / / / / 2 1 1 0 6 1 8 9 9 6 7 7 n o _ a _ 0 0 0 2 7 p d / . l f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Neural correlates of speech motor sequence learning 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 / / / / 2 1 1 0 6 1 8 9 9 6 7 7 n o _ a _ 0 0 0 2 7 p d / . l f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Figure 5. Brain areas showing greater BOLD activation for novel non-native than practiced non-native trials. (A) Averaged across all 15 ANS and 16 AWS; (B) 15 ANS (left) and 16 AWS (right). Activation is displayed on a canonical inflated cortical surface. Colors indicate the relative significance level at each voxel/vertex for the comparison of novel non-native and practiced non-native trials. significant differences in subcortical activity. We supplemented this voxel-based analysis with a hypothesis-based subcortical ROI analysis based on the results reported in Segawa, Tourville, et al. (2015) for this contrast. Specifically, we tested for differences in activation in the left GPi and GPe. Results from this analysis demonstrated that the left GPi was more active during produc- tion of the novel non-native than the practiced non-native words [t(30) = 2.22, punc = 0.034, r = 0.375], in effect replicating the finding of Segawa, Tourville, et al. (2015) for the same contrast. No significant groups effects (ANS vs. AWS) were found for any cortical or subcortical region in either hemisphere for the novel non-native–practiced non-native contrast; nevertheless, we present the results (albeit non-significant) for the two groups separately in Figure 5B. Figure 6A and Table 6 show the brain regions that were significantly more active during the execution of practiced non-native than practiced native syllables (TFCE cluster-level pFWE < 0.05). The surface/vertex-based analyses revealed that the production of practiced non-native words resulted in greater BOLD response in pSTG and occipital cortex (OC) bilaterally. The in- creased activity in OC likely reflects greater attention and/or depth of processing while reading Neurobiology of Language 125 Neural correlates of speech motor sequence learning Table 5. Summary of significant cortical and subcortical activation peaks for the novel non-native−practiced non-native contrast Novel non-native–practiced non-native contrast* Anatomical region** Left hemisphere L pIFS, L aINS, L vPMC, L pFO, L midPMC, L aFO, L pMFG, L aIFS, L midMC, L dIFO, L vIFO, L IFR, L vMC, L aMFG, L FOC No. peaks MNI coordinates of peak vertex y z x Size TFCE pFWE 33 −29 26 7 10170 5292.04 <0.0001 L SPL, L AG, L pSMG, L aSMG, L OC, L midSC 27 L ITO, L pITG, L TOF L preSMA, L SFG, L SMA L pdSTS, L pSTG, L PT, L pvSTS L pITG Right hemisphere R aINS, R pFO, R aFO, R IFR, R FOC R SPL, R AG, R OC R SFG, R preSMA R preSMA 7 3 3 5 9 1 3 1 −27 −47 −7 −63 −47 36 29 9 7 −65 −59 12 −35 −43 30 −62 21 16 37 −9 55 4 −21 −1 35 38 55 7665 4210.05 <0.0001 713 2009.43 0.001 1150 1663.39 0.002 1662 1262.41 0.005 23 936.82 0.024 2202 2562.18 <0.0001 1066 1923.23 0.002 180 267 1003.95 0.019 953.0 0.024 Note. From left to right, the columns show the anatomical region name(s), number of peaks, MNI stereotactic coordinates, cluster size, TFCE value, and pFWE value. * For the one-sided (positive-only) contrast. ** Cluster regions are listed in descending order of number of significant vertices. 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 / / / / 2 1 1 0 6 1 8 9 9 6 7 7 n o _ a _ 0 0 0 2 7 p d / . l f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 the non-native text stimuli, rather than differences in speech production. In the left hemisphere, additional cortical clusters were noted with peaks in preSMA, aINS, FO, pIFG, vPMC, midPMC, aSTG, ventral division of somatosensory cortex (vSC), and IPS. The volume/voxel-based analysis found increased activation in the right cerebellum (lobule XIII) during the practiced non-native than the practiced native condition. A hypothesis-based subcortical ROI analysis based on the results reported in Segawa, Tourville, et al. (2015) for this contrast was then performed. Specifically, we tested for differences in activation in right cerebellum (lobule VI). Results from this analysis demonstrated that this region was more active during production of the practiced non-native than the practiced native words [t(30) = 2.61, pFDR = 0.014, r = 0.430], again repli- cating findings from Segawa, Tourville, et al. (2015). As in the other contrast, no region in either hemisphere was found to be significantly more active for this contrast in either group (ANS vs. AWS) at the cortical or subcortical levels, further bolstering the view that speech motor learning mechanisms are unimpaired in stuttering; again, we present the results (albeit non-significant) for the two groups separately in Figures 6B. Although no significant groups effects were found, several interesting trends emerged that are worth noting and speculating about. First, brain areas linked to speech premotor planning (vPMC, midPMC, FO, aINS) showed a non-significant trend toward greater activation in ANS during novel sequence production, suggesting that neurotypical speakers may show greater learning Neurobiology of Language 126 Neural correlates of speech motor sequence learning 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 / / / / 2 1 1 0 6 1 8 9 9 6 7 7 n o _ a _ 0 0 0 2 7 p d / . l Figure 6. Brain areas showing greater BOLD activation for practiced non-native than practiced native trials. (A) Averaged across all 15 ANS and 16 AWS; (B) 15 ANS (left) and 16 AWS (right). Activation is displayed on a canonical inflated cortical surface. The significant subcortical cluster is shown on a slice through the cerebellum at z = −54 in the MNI template (panel A, bottom right); left and right hemispheres are indicated by L and R, respectively. Colors indicate the relative significance level at each voxel/vertex for the comparison of novel non-native and practiced non-native trials. f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 of the motor chunks for the trained sequences than AWS. Second, STG, an auditory processing area, showed a non-significant trend toward greater activation in AWS during novel sequence production, suggesting that, following motor practice, stuttering speakers may still be relying more on auditory feedback to guide them toward the intended production targets compared to ANS. Critically, however, there were no brain areas activated in AWS that fell outside the network of areas identified in ANS and in the previous study (Segawa, Tourville, et al., 2015). Brain-Behavior Correlation Analysis The correlation tests between behavioral measures of learning success and mean activation in the 12 hypothesis-based ROIs from Segawa, Tourville, et al. (2015) (listed in Table 2) for the novel non-native−practiced non-native contrast revealed no group differences at either the cortical or subcortical levels. In the analyses performed on the pooled data (ANS and AWS combined), we found that learning success, as measured by the participant-normalized difference in mean error Neurobiology of Language 127 Neural correlates of speech motor sequence learning Table 6. Summary of significant cortical and subcortical activation peaks for the practiced non-native−practiced native syllable contrast Practiced non-native–practiced native contrast* Anatomical region** Left hemisphere L SPL, L PT, L aSMG, L vSC, L PO, L pSTG, L H, L PP, L pSMG,L AG, L midSC, L pdSTS, L aSTG, L pCO, L OC, L adSTS, L pINS L vPMC, L midPMC, L aINS, L pFO, L aFO, L vIFO, L vMC, L dIFO, L pIFS, L IFR, L aCO, L pMFG L preSMA, L SMA, L dCMA L OC L pMFG, L pdPMC, L mdPMC, L midPMC Right hemisphere R PP, R H, R pINS R OC Cerebellum lobule VIII Cerebellum lobule VII b No. peaks 47 17 1 1 1 5 1 11 1 MNI coordinates of peak vertex y z x Size TFCE pFWE −48 −51 −9 −15 −30 46 22 18 32 −38 1 12 −101 −8 −18 −100 −70 −74 13 13952 4603.4 <0.0001 45 50 −7 46 2 6 −46 −56 6766 3863.43 <0.0001 1246 1502.18 0.001 2089 1246.82 0.009 726 977.68 0.019 407 1668.33 0.001 1358 1583.67 0.001 304 846.54 0.009 1 645.83 0.024 Note. From left to right, the columns show the anatomical region name(s), number of peaks, MNI stereotactic coordinates, cluster size, TFCE value, and pFWE value. * For the one-sided (positive-only) contrast. ** Cluster regions are listed in descending order of number of significant vertices. rates between the novel non-native and practiced non-native words, was positively correlated with the mean activation in both the left [t(29) = 2.22, p = 0.034, r = 0.381] and right aINS [t(29) = 2.38, p = 0.024, r = 0.404], such that processing in these areas increased most with learning in participants who showed the greatest performance improvements (shown in Figure 7). These results provide support for the hypothesis that these premotor areas are involved in speech motor sequence learning. No significant correlations emerged between the participant-normalized differ- ences in mean utterance duration (between the novel non-native and practiced non-native words) and the mean activation in any of the ROIs (in all cases, punc > 0.05).

Enfin, within the AWS group, we assessed whether stuttering severity (as indexed by SSI
scores and mean disfluency rates inside the scanner during the test phase) were correlated with
activation in eight hypothesis-based ROIs within the cortico-BG loops (listed in Table 2) for the
novel non-native−practiced non-native contrast. Correlation tests revealed that mean disfluency
rates inside the scanner were negatively correlated with mean activation in the left GPe [t(14) =
−2.88, pFDR = 0.047, r = −0.609] and the left caudate [t(14) = −2.74, pFDR = 0.047, r = −0.591]
(c'est à dire., the more disfluent participants were during the test phase, the less BOLD activity change was
observed in these basal ganglia sites for this contrast; shown in Figure 8). Although this result is
correlational and therefore a causal relationship cannot be firmly established, it is nonetheless

Neurobiology of Language

128

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
n
o

je
/

un
r
t
je
c
e
–
p
d

F
/

2
1
1
0
6
1
8
9
9
6
7
7
n
o
_
un
_
0
0
0
2
7
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

Neural correlates of speech motor sequence learning

Significant correlation between participant-normalized difference in mean error rates between the novel non-native and practiced
Chiffre 7.
non-native words and the mean activation in the left aINS [r = 0.381, p = 0.034] (left panel) and right aINS [r = 0.404, p = 0.024] (right panel).

compatible with the idea that speakers are more likely to stutter when the left GPe and the left
caudate are more hypoactive. In contrast, no statistically significant correlations were found
between SSI scores and activation in any of the eight ROIs. The latter result is perhaps unsurprising
given that the SSI does not measure disfluency during non-native sequence production and its
affiliated scores were not based on speech uttered inside the scanner.

En résumé, we found that, across both ANS and AWS, greater activity was observed during the
production of novel sequences in brain regions previously associated with learning and maintaining
speech motor programs, including lateral premotor cortex, FO, aINS, posterior superior temporal
cortex, and right cerebellum (Guenther, 2016; Segawa, Tourville, et coll., 2015). Measures of learning
success correlated positively with activity in the left and right aINS, suggesting that this brain area
plays an important role in speech motor learning. In AWS, measures of in-scanner disfluency rates
were negatively correlated with activity in left-lateralized basal ganglia structures. Dans l'ensemble, ces

Chiffre 8. Significant correlation between mean disfluency rates during the test phase (inside the scanner) and mean activation in the left GPe
[r = −0.609, pFDR = 0.047] (left panel) and the left caudate [r = −0.591, pFDR = 0.047] (right panel).

Neurobiology of Language

129

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
n
o

je
/

un
r
t
je
c
e
–
p
d

F
/

2
1
1
0
6
1
8
9
9
6
7
7
n
o
_
un
_
0
0
0
2
7
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

Neural correlates of speech motor sequence learning

findings provide evidence that the neurobiological correlates of speech motor sequence learning do
not differ across ANS and AWS.

DISCUSSION

The goal of the current study was to investigate the behavioral and neural correlates of speech
motor sequence learning in both stuttering and neurotypical adult speakers. Using a cued
sequence production task, we examined changes in the performance of pseudowords containing
non-native consonant clusters over time as a function of speech motor practice. At the behavioral
level, we found that, although AWS were less accurate and/or slower compared to ANS at
producing words involving non-native consonant clusters regardless of whether the words were
practiced, they showed improvements in these measures with practice that were statistically
comparable to those seen in ANS. At the neural level, we found that, across both groups and both
speech contrasts (novel non-native−practiced non-native and practiced non-native−practiced
native), practice producing words with non-native clusters led to significant decreases in BOLD
activity in brain areas implicated in verbal working memory and speech motor planning (dans-
cluding the left preSMA, FO, aINS, and bilateral IPS), suggesting that processing load and artic-
ulatory effort decrease as the nervous system forms regularized motor programs (or chunks) pour
coordinated sequences of vocal tract gestures. Critique, there were no reliable group differences
in either contrast map in any areas involved in the cortico-BG loops (or otherwise). En outre,
the degree of motor performance gains in accuracy were correlated positively with activity in the
left and right aINS. Collectively, these findings suggest that cortico-BG-mediated mechanisms
involved in learning novel phoneme sequences are not impaired in AWS; instead, the impairment
is primarily in the motor execution of speech sequences, novel and learned alike.

As in our prior study (Segawa, Tourville, et coll., 2015), we also observed pre- to post-training
BOLD activation changes in cortical areas that extend beyond those implicated in verbal working
memory and speech motor planning processes. Spécifiquement, we found greater activity in areas
related to auditory processes (the left PT, aSTG, and pSTG) during production of the novel non-
native sequences relative to the practiced non-native sequences (Chiffre 5). These auditory
regions are thought to be involved in guiding speech movements based on self-generated audi-
tory feedback (Guenther, 2016; Hickok, 2012). During speech production, activity in this area
has been reported to be greater when there is a mismatch between predicted and actual auditory
feedback (Tourville et al., 2008). Several existing theoretical models of speech production (par exemple.,
Guenther, 2016; Hickok, 2012) propose that error signals arising from these regions are used to
fine-tune speech motor programs over the course of repeated production attempts. Ainsi, learning
is thought to rely on the transmission of these auditory error signals to frontal regions involved in
motor planning and execution.

We also noted greater activity in cortical areas related to orthographic processing (the left ITO)
during production of the novel non-native sequences relative to the practiced non-native
sequences (Chiffre 5). Prior neuroimaging studies suggest that the ITO is a higher-level visual
processing area involved in identifying letters and words from lower-level shape images (voir,
par exemple., Pugh, Mencl, et coll., 2001) and is therefore highly likely to be related to viewing and decoding
the pseudowords that our participants were instructed to produce, rather than speech motor
control. In contrast, production of the practiced non-native sequences did not produce signifi-
cantly greater activity compared with the practiced native sequences in the ITO (Chiffre 6).
This difference between the condition-specific brain activations suggests that pseudowords
containing unfamiliar phonotactic sequences are more difficult to process prior to reading and
repetition practice (cf. Pugh, Frost, et coll., 2008).

Neurobiology of Language

130

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
n
o

je
/

un
r
t
je
c
e
–
p
d

F
/

2
1
1
0
6
1
8
9
9
6
7
7
n
o
_
un
_
0
0
0
2
7
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

Neural correlates of speech motor sequence learning

The finding that the right cerebellum (which is structurally and functionally connected with left
cortical areas) was recruited to a greater extent across both groups during the production of the
practiced non-native than the practiced native sequences (shown in Figure 6) was also observed
by Segawa, Tourville, et autres. (2015) in ANS, and corroborates other reports with neurotypical
participants that the cerebellum plays an important role in motor sequence learning and fine
motor control (par exemple., Ackermann, 2008; Doyon, Song, et coll., 2002; Guenther, 2016, Chapter 2).
Par exemple, Bohland and Guenther (2006) found that different regions of the cerebellum
showed differential sensitivity to syllable complexity (par exemple., “stra” vs. “ta”) and serial complexity
(par exemple., “da-da-da” vs. “da-ru-ti”) in ANS, suggesting involvement in speech sequencing. That said,
it is somewhat surprising that we did not find significant differences in cerebellar activity in the
novel non-native−practiced non-native contrast. Cependant, this comparison possessed less
statistical power because of more removed error trials.

An alternative, but not mutually exclusive, interpretation to the cortical activity differences we
found for both speech contrasts is that these differences result from the fact that the novel non-
native and practiced non-native sequences are more difficult to vocally imitate than the practiced
native sequences. Two sources of evidence support this conjecture. D'abord, prior neuroimaging
études (par exemple., Irwin et al., 2011) have reported that imitation of auditory speech produces addi-
tional significant activations in the left IFG and aINS when compared to a passive listening con-
dition. Deuxième, current computational models, such as the Directions Into Velocities of the
Articulators (DIVA) model (Guenther, 2016), propose that speech motor learning is driven by
initial mismatches between newly acquired sound targets and one’s own production attempts
as represented in the auditory cortex. Auditory error signals are then transformed into corrective
motor commands, and these corrective commands alter the feedforward commands for the next
production attempt. As the feedforward commands improve, fewer error signals are generated
and thus the contribution of the feedback control system gradually diminishes. The DIVA model
thus predicts decreases in BOLD activation in both motor planning and auditory cortical areas as
a consequence of imitation-based learning. In this view, the nervous system begins to form
speech motor programs via an imitation-based learning mechanism, which may account for
the increased activity in motor and auditory areas for the novel non-native compared with the
practiced non-native sequences, and also for the practiced non-native compared with the
practiced native sequences.

A number of other neuroimaging studies, compiled and discussed in a meta-analysis by Belyk
et coll., (2015), have reported that AWS tend to display higher activity in a number of right-
hemisphere regions during fluently produced native speech when compared to ANS. Some
researchers have proposed that this right-hemisphere cortical hyperactivity arises from impaired
left-hemisphere function (voir, par exemple., Belyk et al., 2015; Fox et al., 1996; Guenther, 2016; Neef et al.,
2017; cf. Connally et al., 2018). In view of these results, it is perhaps surprising that no clusters
emerged in the right hemisphere for AWS in the current study. Cependant, it is possible that right-
hemisphere hyperactivity in AWS may not occur in all speaking situations. En particulier, le
current speaking task likely required more attention and articulatory effort than those in prior
études, since the participants were required to produce non-native sequences of segments, lequel
is known to be difficult (par exemple., Davidson, 2006, 2010). En outre, there may be methodological
issues that concern the difference between the experimental procedures we used to test for a
functional activation difference between ANS and AWS and the ones used by other research
groupes, as the existing literature on stuttering encompasses a diverse set of sampling protocols
and motor tasks.

In the current study, it was further demonstrated that, in AWS, the in-scanner mean disfluency
covaried with BOLD responses in left-lateralized basal ganglia sites (the GPe and left caudate),

Neurobiology of Language

131

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
n
o

je
/

un
r
t
je
c
e
–
p
d

F
/

2
1
1
0
6
1
8
9
9
6
7
7
n
o
_
un
_
0
0
0
2
7
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

Neural correlates of speech motor sequence learning

such that the more disfluent participants were during the test phase, the less BOLD activity change
was observed in these basal ganglia sites for the novel non-native−practiced non-native contrast.
This negative correlation does not establish a causal relationship; rather it establishes an asso-
ciative link. Although this relationship is broadly consistent with the long-standing view that
stuttering reflects a malfunction within the cortico-BG loops (par exemple., Alm, 2004; Chang & Guenther,
2020; Connally et al., 2018; Fox et al., 1996; Giraud et al., 2008; Kell et al., 2009; Lu et al., 2010;
Watkins et al., 2008), it would appear to be at odds with other neuroimaging studies (albeit that
did not use a sparse-sampling method) which have reported positive relationships between
basal ganglia activity and stuttering severity (par exemple., Giraud et al., 2008; see also Metzger et al.,
2018, for a similar finding based on manual motor responses in AWS). Differences between the
findings of the current study and these prior studies suggest that the brain mechanisms underlying
general stuttering traits and transient disfluent states are dissociable (for further discussion, voir
Connally et al., 2018).

At first glance, the lack of a robust learning difference between the ANS and AWS groups may
appear to be at odds with other behavioral studies reporting motor learning deficits in AWS
(Ludlow et al., 1997; Namasivayam & van Lieshout, 2008; Smits-Bandstra et al., 2006). Several
important differences between those studies and ours might account for the divergent results.
D'abord, our study measured sequence and cluster learning using perception-based segmental
transcription and simple acoustic measures. Cependant, this approach required transcribers to make
categorical decisions regarding the segments that speakers produced, and thus did not permit
quantitative analysis of the presence or magnitude of various gestures, or of how speakers
continually updated their implementation of novel gestural scores. In contrast, other motor
learning studies (par exemple., Namasivayam & van Lieshout, 2008) used kinematic measures to com-
pare more subtle aspects of speech articulation between ANS and AWS. Ainsi, subtle differences
in the articulatory correlates of speech motor learning may exist between ANS and AWS.
Ongoing studies based on the current design that use electromagnetic articulography will
determine whether AWS and ANS differ in their ability to learn to execute and coordinate inter-
articulator movements for novel speech sequences.

Deuxième, our study focused on the learning of new phoneme sequences within a single syllable
with which the participants had no prior experience, since they violated English phonotactic
constraints. This task required the learning of new articulatory sequences at a motoric level. Dans
contraste, other studies in the literature on speech motor sequence learning in AWS have involved
production of more extended, multi-syllabic sequences that conform to English phonotactics
(par exemple., Namasivayam & van Lieshout, 2008). Cependant, there is evidence that multi-syllabic
sequences place higher demands on phonological working memory mechanisms than on motor
programming mechanisms. Par exemple, in another fMRI study, McGettigan et al. (2011) trouvé
that activity in the PT, a region implicated in auditory working memory, increased as the number
of syllables in either a non-word repetition or a passive listening task increased, whereas activity
in the left SMA, a region implicated in speech motor programming, increased as the number of
consonant clusters increased. Dans cette lumière, the increased error rates reported in other studies with
AWS might be interpreted instead as reflecting differences in phonological working memory
mechanisms rather than motor learning or implementation mechanisms (see also Yang et al.,
2019). Yet others have reported some differences between ANS and AWS in the degree of motor
adaptation following an experimental perturbation to the online sensory feedback (par exemple., Daliri
et coll., 2018), suggesting that mechanisms involved in updating existing motor programs may still
be impaired in stuttering.

It is also important to note that the present fMRI measures focused on the outcome of speech
motor sequence learning, not the online process of learning. Cependant, previous research spanning

Neurobiology of Language

132

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
n
o

je
/

un
r
t
je
c
e
–
p
d

F
/

2
1
1
0
6
1
8
9
9
6
7
7
n
o
_
un
_
0
0
0
2
7
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

Neural correlates of speech motor sequence learning

a wide range of learning tasks and paradigms has shown that participants may have similar learning
outcomes despite very different learning trajectories (see Karuza et al., 2014, for a review). Ainsi,
despite showing comparable BOLD activity patterns at test, ANS and AWS may not show evidence
of the same qualitative neural changes during the course of speech motor sequence training. Il
would therefore be of interest to conduct further fMRI studies, based on the current design, que
perform BOLD contrast analyses throughout both the training and the test phases.

Enfin, the current study focused primarily on the mechanisms of speech motor learning and
control, from the phonetic encoding stage down to the motor commands to the speech articula-
tors. Spécifiquement, we examined how ANS and AWS learn to transform discrete phonological
chunks (speech sounds that can be phonemes, syllables, or words) into a set of articulator move-
ments that achieves the intended auditory “target.” We employed monosyllabic pseudoword
stimuli to provide a “purer” measure of the speech motor system, as this would limit the recruit-
ment of higher-level cognitive and linguistic processing strategies to assist task performance (voir
also McGettigan et al., 2011). Cependant, any viable theory of speech production learning will
ultimately have to explicate how these phonological chunks are integrated with prosodic struc-
photos, as well as their relation to syntactic/semantic planning processes. Future studies will aim to
develop predictions from the highly controlled laboratory experiments presented here and else-
où (e.g.,Segawa, Masapollo, et coll., 2019; Segawa, Tourville, et coll., 2015) that may be tested
when speakers are instructed to produce novel speech sequences in a more natural context.

REMERCIEMENTS

We are grateful to Barbara Holland and Diane Constantino for assistance with participant recruit-
ment and/or data collection. We also thank Riccardo Falsini, Farwa Faheem, Abigail Cragin, Ariel
Gordon, Angelise Bulit, and Jessica Smith for help with data analysis and visualization. Enfin,
this work benefited from helpful discussions with, or comments from, Elaine Kearney, Megan
Thompson, Elizabeth Heller Murray, Jason Bohland, Cara Stepp, Kenneth Logan, and Tyler
Perrachionne. Research reported in this publication was supported by the National Institute on
Deafness and other Communication Disorders of the National Institutes of Health under award
number R01DC007683 (F. H. Guenther, PI). The content is solely the responsibility of the authors
and does not necessarily represent the official views of the National Institutes of Health.

INFORMATIONS SUR LE FINANCEMENT

Frank H. Guenther, National Institute on Deafness and Other Communication Disorders (http://
dx.doi.org/10.13039/100000055), Award ID: R01DC007683.

CONTRIBUTIONS DES AUTEURS
Matthew Masapollo: Visualisation; Analyse formelle; Rédaction – ébauche originale. Jennifer Segawa: Données
curation; Analyse formelle. Deryk S. Beal: Conservation des données; Analyse formelle. Jason A. Tourville:
Conceptualisation; Conservation des données; Analyse formelle, Rédaction – révision & édition. Alfonso Nieto-
Castañón: Analyse formelle; Rédaction – révision & édition. Matthias Heyne: Analyse formelle. Saul
UN. Frankford: Conservation des données; Analyse formelle. Frank H. Guenther: Conceptualisation; En écrivant -
revoir & édition.

RÉFÉRENCES

Ackermann, H. (2008). Cerebellar contributions to speech produc-
tion and speech perception: Psycholinguistic and neurobiological
perspectives. Trends in Neurosciences, 31(6), 265–272. EST CE QUE JE:
https://doi.org/10.1016/j.tins.2008.02.011, PMID: 18471906

Alexander, G. E., DeLong, M.. R., & Strick, P.. L. (1986). Parallel orga-
nization of functionally segregated circuits linking basal ganglia and
cortex. Revue annuelle des neurosciences, 9, 357–381. EST CE QUE JE: https://
doi.org/10.1146/annurev.ne.09.030186.002041, PMID: 3085570

Neurobiology of Language

133

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
n
o

je
/

un
r
t
je
c
e
–
p
d

F
/

2
1
1
0
6
1
8
9
9
6
7
7
n
o
_
un
_
0
0
0
2
7
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

Neural correlates of speech motor sequence learning

Alm, P.. UN. (2004). Stuttering and the basal ganglia circuits: A critical
review of possible relations. Journal of Communication Disorders,
37(4), 325–369. EST CE QUE JE: https://doi.org/10.1016/j.jcomdis.2004
.03.001, PMID: 15159193

Andersson, J.. L., Hutton, C., Ashburner, J., Tourneur, R., & Friston, K.
(2001). Modeling geometric deformations in EPI time series.
NeuroImage, 13(5), 903–919. EST CE QUE JE: https://doi.org/10.1006/nimg
.2001.0746, PMID: 11304086

Ashburner, J., & Friston, K. J.. (2005). Unified segmentation.
NeuroImage, 26(3), 839–851. EST CE QUE JE: https://est ce que je.org/10.1016
/j.neuroimage.2005.02.018, PMID: 15955494

Belin, P., Zatorre, R.. J., Hoge, R., Evans, UN. C., & Pike, B. (1999).
Event-related fMRI of the auditory cortex. NeuroImage, 10(4), 417–429.
EST CE QUE JE: https://doi.org/10.1006/nimg.1999.0480, PMID: 10493900
Belyk, M., Kraft, S. J., & Brun, S. (2015). Stuttering as a trait or
state—an ALE meta-analysis of neuroimaging studies. européen
Journal des neurosciences, 41(2), 275–284. EST CE QUE JE: https://est ce que je.org/10
.1111/ejn.12765, PMID: 25350867

Berent, JE., Steriade, D., Lennertz, T., & Vaknin, V. (2007). Ce que nous
know about what we have never heard: Evidence from perceptual
illusions. Cognition, 104(3), 591–630. EST CE QUE JE: https://est ce que je.org/10
.1016/j.cognition.2006.05.015, PMID: 16934244

Bloodstein, O., & Ratner, N. B. (2008). A handbook on stuttering.

Thomson/Delmar Learning.

Bohland, J.. W., Bullock, D., & Guenther, F. H. (2010). Neural repre-
sentations and mechanisms for the performance of simple speech
sequences. Journal des neurosciences cognitives, 22(7). EST CE QUE JE: https://
doi.org/10.1162/jocn.2009.21306, PMID: 19583476, PMCID:
PMC2937837

Bohland, J.. W., & Guenther, F. H. (2006). An fMRI investigation of
syllable sequence production. NeuroImage, 32, 821–841. EST CE QUE JE:
https://doi.org/10.1016/j.neuroimage.2006.04.173, PMID:
16730195

Boersma, P., & Weenink, D. (2020). Praat: Doing phonetics by
computer ( Version 6.1.09) [Computer program]. Retrieved 26
Janvier 2020 from https://www.praat.org/.

Boutsen, F. (1993). A comparative study of stress timing of stutterers
and nonstutterers. Journal of Fluency Disorders, 20(2), 145–155.
EST CE QUE JE: https://doi.org/10.1016/0094-730X(94)00023-M.

Brawn, T. P., Fenn, K. M., Nusbaum, H. C., & Margoliash, D.
(2010). Consolidating the effects of waking and sleep on motor-
sequence learning. Journal des neurosciences, 30(42), 13977–13982.
EST CE QUE JE: https://doi.org/10.1523/JNEUROSCI.3295-10.2010, PMID:
20962219, PMCID: PMC2978076

Buchwald, UN., Calhoun, H., Rimikis, S., Steinberg Lowe, M.,
Wellner, R., & Edwards, D. (2019). Using tDCS to facilitate motor
learning in speech production: The role of timing. Cortex, 111,
274–285. EST CE QUE JE: https://doi.org/10.1016/j.cortex.2018.11.014,
PMID: 30551048, PMCID: PMC6358507

Cai, S., Tourville, J.. UN., Beal, D. S., Perkell, J.. S., Guenther, F. H., &
Ghosh, S. S. (2014). Diffusion imaging of cerebral white matter in
persons who stutter: Evidence for network-level anomalies.
Frontiers in Human Neuroscience, 8(54). EST CE QUE JE: https://doi.org
/10.3389/fnhum.2014.00054, PMID: 24611042, PMCID:
PMC3920071

Chang, S.-E., & Guenther, F. H. (2020). Involvement of the cortico-
basal-ganglia-thalamocortical loop in developmental stuttering.
Frontiers in Psychology, 10, 1–15. EST CE QUE JE: https://doi.org/10.3389
/fpsyg.2019.03088, PMID: 32047456, PMCID: PMC6997432
Chang, S.-E., Ohde, R.. N., & Conture, E. G. (2002). Coarticulation
and formant transition rate in young children who stutter. Journal
of Speech, Language, and Hearing Research, 45(4), 676–688.
EST CE QUE JE: https://doi.org/10.1044/1092-4388(2002/054)

Cholin, J., Levelt, W. J.. M., & Schiller, N. Ô. (2006). Effects of syllable
frequency in speech production. Cognition, 99(2), 205–235. EST CE QUE JE:
https://doi.org/10.1016/j.cognition.2005.01.009, PMID:
15939415

Civier, O., Bullock, D., Max, L., & Guenther, F. H. (2013).
Computational modeling of stuttering caused by impairments
in a basal ganglia thalamocortical circuit involved in syllable
selection and initiation. Brain and Language, 126(3), 263–278.
EST CE QUE JE: https://doi.org/10.1016/j.bandl.2013.05.016, PMID:
23872286, PMCID: PMC3775364

Connally, E. L., Ward, D., Pliatsikas, C., Finnegan, S., Jenkinson,
M., Boyles, R., & Watkins, K. E. (2018). Separation of trait and
state in stuttering. Human Brain Mapping, 39(8), 3109–3126.
EST CE QUE JE: https://doi.org/10.1002/ hbm.24063, PMID: 29624772,
PMCID: PMC6055715

Coltheart, M.. (1981). The MRC Psycholinguistic Database. Le
Quarterly Journal of Experimental Psychology Section A, 33(4),
497–505. EST CE QUE JE: https://doi.org/10.1080/14640748108400805
Craig-McQuaide, UN., Akram, H., Zrinzo, L., & Tripoliti, E. (2014). UN
review of brain circuitries involved in stuttering. Frontiers in
Human Neuroscience. EST CE QUE JE: https://doi.org/10.3389/fnhum
.2014.00884, PMID: 25452719, PMCID: PMC4233907

Dale, UN. M., Fischl, B., & Séréno, M.. je. (1999). Cortical surface-based
analysis—I. Segmentation and surface reconstruction. NeuroImage,
9(2), 179–194. EST CE QUE JE: https://doi.org/10.1006/nimg.1998.0395,
PMID: 9931268

Daliri, UN., Wieland, E. UN., Cai, S., Guenther, F. H., & Chang, S. E.
(2018). Auditory-motor adaptation is reduced in adults who stutter
but not in children who stutter. Developmental Science, 21(2),
e12521. EST CE QUE JE: https://doi.org/10.1111/desc.12521, PMID:
28256029, PMCID: PMC5581739

Davidson, L. (2006). Phonology, phonetics, or frequency:
Influences on the production of non-native sequences. Journal
of Phonetics, 34, 104–137. EST CE QUE JE: https://doi.org/10.1016/j.wocn
.2005.03.004

Davidson, L. (2010). Phonetic bases of similarities in cross-language
production: Evidence from English and Catalan. Journal de
Phonetics, 38(2), 272–288. EST CE QUE JE: https://doi.org/10.1016/j.wocn
.2010.01.001

Davidson, L., & Shaw, J.. (2012). Sources of illusion in consonant
cluster perception. Journal of Phonetics, 40(2), 234–248. EST CE QUE JE:
https://doi.org/10.1016/j.wocn.2011.11.005

Diedrichsen, J., Balsters, J.. H., Flavell, J., Cussans, E., & Ramnani, N.
(2009). A probabilistic MR atlas of the human cerebellum.
NeuroImage, 15(1), 39–46. EST CE QUE JE: https://doi.org/10.1016/j.neuroimage
.2009.01.045, PMID: 19457380

Doyon, J., Albouy, G., Vahdat, S., & King, B. (2015). Neural correlates
of motor skill acquisition and consolidation. In A. Toga (Ed.), Cerveau
mapping: An encyclopedic reference. Elsevier. EST CE QUE JE: https://est ce que je
.org/10.1016/B978-0-12-397025-1.00275-X

Doyon, J., Bellec, P., Amsel, R., Penhune, V., Monchi, O., Carrier,
J., Lehericy, S., & Benali, H. (2009). Contributions of the basal
ganglia and functionally related brain structures to motor learning.
Behavioral Brain Research, 199(1), 61–75. EST CE QUE JE: https://est ce que je.org/10
.1016/j.bbr.2008.11.012, PMID: 19061920

Doyon, J., Song, UN. W., Karni, UN., Lalonde, F., Adams, M.. M., &
Ungerleider, L. G. (2002). Experience-dependent changes in
cerebellar contributions to motor sequence learning. Procédure
of the National Academy of Sciences, 99(2), 1017–1022. EST CE QUE JE:
https://doi.org/10.1073/pnas.022615199, PMID: 11805340,
PMCID: PMC117423

Dupoux, E., Kakehi, K., Pallier, Y., Hirose, C., & Mehler, J.. (1999).
Epenthetic vowels in Japanese: A perceptual illusion? Journal de

Neurobiology of Language

134

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
n
o

je
/

un
r
t
je
c
e
–
p
d

F
/

2
1
1
0
6
1
8
9
9
6
7
7
n
o
_
un
_
0
0
0
2
7
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

Neural correlates of speech motor sequence learning

Experimental Psychology: Perception humaine et performance,
25(6), 1568–1578. EST CE QUE JE: https://doi.org/10.1037/0096-1523.25
.6.1568

Dupoux, E., Parlato, E., Frota, S., Hirose, Y., & Peperkamp, S. (2011).
Where do illusory vowels come from? Journal of Memory and
Language, 64(3), 199–210. EST CE QUE JE: https://doi.org/10.1016/j.jml
.2010.12.004

Fee, M.. S., & Goldberg, J.. H. (2011). A hypothesis for basal ganglia
dependent reinforcement learning in the songbird. Neurosciences,
198, 152–170. EST CE QUE JE: https://doi.org/10.1016/j.neuroscience
.2011.09.069, PMID: 22015923, PMCID: PMC3221789

Fenn, K. M., Nusbaum, H. C., & Margoliash, D. (2003).
Consolidation during sleep of perceptual learning of spoken
langue. Nature, 425, 614–616. EST CE QUE JE: https://est ce que je.org/10.1038
/nature01951, PMID: 14534586

Ferraro, F. R., Balota, D. UN., & Connor, L. T. (1993). Implicit memory
and the formation of new associations in nondemented Parkinson’s
disease individuals and individuals with senile dementia of the
Alzheimer type: A serial reaction time (SRT) enquête. Cerveau
et cognitif, 21(2), 163–180. EST CE QUE JE: https://doi.org/10.1006/brcg
.1993.1013, PMID: 8442933

Fischl, B., Salat, D. H., Busa, E., Albert, M., Dieterich, M.,
Haselgrove, C., van der Kouwe, UN., Killiany, R., Kennedy, D.,
Klaveness, S., Montillo, UN., Makris, N., Rosen, B., & Dale, UN. M..
(2002). Whole brain segmentation: Automated labeling of neuro-
anatomical structures in the human brain. Neurone, 33(3), 341–355.
EST CE QUE JE: https://doi.org/10.1016/S0896-6273(02)00569-X

Fischl, B., Séréno, M.. JE., & Dale, UN. M.. (1999). Cortical surface-based
analysis—II: Inflation, flattening, and a surface-based coordinate
système. NeuroImage, 9(2), 195–207. EST CE QUE JE: https://est ce que je.org/10
.1006/nimg.1998.0396, PMID: 9931269

Fox, P.. T., Ingham, R.. J., Ingham, J.. C., Hirsch, T. B., Downs, J.. H.,
Martine, C., Jerabek, P., Verre, T., & Lancaster, J.. L. (1996). A PET study
of the neural systems of stuttering. Nature, 382(6587), 158–161.
EST CE QUE JE: https://doi.org/10.1038/382158a0, PMID: 8700204

Giraud, A.-L., Neumann, K., Bachoud-Levi, A.-C., von Gudenberg,
UN. W., Euler, H. UN., Lanfermann, H., & Preibisch, C. (2008).
Severity of dysfluency correlates with basal ganglia activity in
persistent developmental stuttering. Brain and Language, 104(2),
190–199. EST CE QUE JE: https://doi.org/10.1016/j.bandl.2007.04.005,
PMID: 17531310

Gracco, V. L., Tremblay, P., & Pike, G. B. (2005). Imaging speech
production using fMRI. NeuroImage, 26, 294–301. EST CE QUE JE: https://
doi.org/10.1016/j.neuroimage.2005.01.033, PMID: 15862230
Guenther, F. H. (2016). Neural control of speech. AVEC Presse. EST CE QUE JE:

https://doi.org/10.7551/mitpress/10471.001.0001

Hagler, D. J., Jr., Saygin, UN. P., & Séréno, M.. je. (2006). Smoothing
and cluster thresholding for cortical surface-based group analysis
of fMRI data. NeuroImage, 33(4), 1093–1103. EST CE QUE JE: https://doi.org
/10.1016/j.neuroimage.2006.07.036, PMID: 17011792, PMCID:
PMC1785301

Helmuth, L., Mayr, U., & Daum, je. (2000). Sequence learning in
Parkinson’s disease: A comparison of spatial-attention and number-
response sequences. Neuropsychologie, 38, 1443–1451. EST CE QUE JE:
https://doi.org/10.1016/S0028-3932(00)00059-2

Hickok, G. (2012). Computational neuroanatomy of speech pro-
duction. Nature Revues Neurosciences, 13(2), 135–145. EST CE QUE JE:
https://doi.org/10.1038/nrn3158, PMID: 22218206, PMCID:
PMC5367153

Ingham, R.. J., Grafton, S. T., Bothe, UN. K., & Ingham, J.. C. (2012).
Brain activity in adults who stutter: Similarities across speaking
tasks and correlations with stuttering frequency and speaking
rate. Brain and Language, 122(1), 11–24. EST CE QUE JE: https://doi.org

/10.1016/j.bandl.2012.04.002, PMID: 22564749, PMCID:
PMC3372660

Irwin, J.. R., Frost, S. J., Mencl, E. W., Chen, H., & Fowler, C. UN.
(2011). Functional activation for imitation of seen and heard
speech. Journal of Neurolinguistics, 24(6), 611–618. EST CE QUE JE: https://
doi.org/10.1016/j.jneuroling.2011.05.001, PMID: 21966094,
PMCID: PMC3182484

Jackson, G. M., Jackson, S. R., Harrison, J., Henderson, L., & Kennard,
C. (1995). Serial reaction time learning in Parkinson’s disease:
Evidence for a procedural learning deficit. Neuropsychologie, 33(5),
577–593. EST CE QUE JE: https://doi.org/10.1016/0028-3932(95)00010-Z
Jezzard, P., & Balaban, R.. S. (1995). Correction for geometric distor-
tion in echo planar images from B0 field variations. Magnetic
Resonance in Medicine, 34(1), 65–73. EST CE QUE JE: https://est ce que je.org/10
.1002/mrm.1910340111, PMID: 7674900

Karuza, E. UN., Emberson, L. L., & Aslin, R.. N. (2014). Combining
fMRI and behavioral measures to examine the process of human
learning. Neurobiology of Learning and Memory, 109, 193–206.
EST CE QUE JE: https://doi.org/10.1016/j.nlm.2013.09.012, PMID: 24076012,
PMCID: PMC3963805

Kell, C. UN., Neumann, K., von Kriegstein, K., Posenenske, C., von
Gudenberg, UN. W., Euler, H., & Giraud, A.-L. L. (2009). How the
brain repairs stuttering. Cerveau, 132(10), 2747–2760. EST CE QUE JE: https://
doi.org/10.1093/brain/awp185, PMID: 19710179

Keuken, M.. C., Bazin, P.-L., Crown, L., Hootsmans, J., Laufer, UN.,
Müller-Axt, C., Sier, R., van der Putten, E. J., Schafer, UN., Tourneur, R.,
& Forstmann, B. U. (2014). Quantifying inter-individual anatomical
variability in the subcortex using 7T structural MRI. NeuroImage,
94, 40–46. EST CE QUE JE: https://doi.org/10.1016/j.neuroimage.2014
.03.032, PMID: 24650599

Kotz, S. UN., & Schwartze, M.. (2010). Cortical speech processing
unplugged: A timely subcortico-cortical framework. Trends in
Sciences cognitives, 14(9), 392–399. EST CE QUE JE: https://est ce que je.org/10.1016
/j.tics.2010.06.005, PMID: 20655802

Lu, C., Chen, C., Ning, N., Ding, G., Guo, T., Peng, D., Lequel, Y., Li,
K., & Lin, C. (2010). The neural substrates for atypical planning
and execution of word production in stuttering. Experimental
Neurologie, 221(1), 146–156. EST CE QUE JE: https://est ce que je.org/10.1016/j
.expneurol.2009.10.016, PMID: 19879262

Ludlow, C. L., Siren, K., & Zikria, M.. (1997). Speech production
learning in adults with chronic developmental stuttering. In W.
Hujstijn, H. F. M.. Peters, & P.. H. H. M.. van Lieshout (Éd.),
Speech production: Motor control, brain research and fluency
disorders (pp. 212–229). Elsevier Science Publishers.

Max, L., Guenther, F. H., Gracco, V. L., Ghosh, S. S., & Wallace,
M.. E. (2004). Unstable or insufficiently activated internal models
and feedback-biased motor control as sources of dysfluency: UN
theoretical model of stuttering. Contemporary Issues in
Communication Science and Disorders, 31, 105–122. EST CE QUE JE:
https://doi.org/10.1044/cicsd_31_S_105

McGettigan, C., Warren, J.. E., Eisner, F., Maréchal, C. R.,
Shanmugalingam, P., & Scott, S. K. (2011). Neural correlates of
sublexical processing in phonological working memory. Journal de
Neurosciences cognitives, 23(4), 961–977. EST CE QUE JE: https://doi.org
/10.1162/jocn.2010.21491, PMID: 20350182, PMCID: PMC3376447
Metzger, F. L., Auer, T., Helms, G., Paulus, W., Frahm, J., Sommer,
M., & Neef, N. E. (2018). Shifted dynamic interactions between
subcortical nuclei and inferior frontal gyri during response prep-
aration in persistent developmental stuttering. Brain Structure
and Function, 223, 165–182. EST CE QUE JE: https://doi.org/10.1007
/s00429-017-1476-1, PMID: 28741037, PMCID: PMC5772149
Mollaei, F., Shiller, D. M., & Gracco, V. L. (2013). Sensorimotor adap-
tation of speech in Parkinson’s disease. Journal of Movement

Neurobiology of Language

135

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
n
o

je
/

un
r
t
je
c
e
–
p
d

F
/

2
1
1
0
6
1
8
9
9
6
7
7
n
o
_
un
_
0
0
0
2
7
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

Neural correlates of speech motor sequence learning

Disorders, 28(12), 1668–1674. EST CE QUE JE: https://doi.org/10.1002/mds
.25588, PMID: 23861349, PMCID: PMC3812368

Nakamura, K., Sakai, K., & Hikosaka, Ô. (1998). Neuronal activity
in medial frontal cortex during learning of sequential procedures.
Journal de neurophysiologie, 80(5), 2671–2687. EST CE QUE JE: https://est ce que je
.org/10.1152/jn.1998.80.5.2671, PMID: 9819272

Namasivayam, UN. K., & Van Lieshout, P.. H. H. M.. (2008). Bite-
block perturbation in people who stutter: Immediate compensatory
and delayed adaptive processes. Journal of Communication
Disorders, 41(4), 372–294. EST CE QUE JE: https://doi.org/10.1016/j.jcomdis
.2008.02.004, PMID: 18405914

Neef, N. E., Anwander, UN., Butfering, C., Schmidt-Samoa, C.,
Friederici, UN. D., Paulus, W., & Sommer, M.. (2017). Structural
connectivity of right frontal hyperactive areas scales with stuttering
severity. Cerveau, 141(1), 191–204. EST CE QUE JE: https://est ce que je.org/10.1093
/brain/awx316, PMID: 29228195, PMCID: PMC5837552

Niethammer, M., Feigin, UN., & Eidelber, D. (2012). Functional neuro-
imaging in Parkinson’s disease. Cold Spring Harbor Perspectives
in Medicine, 2(5), a009274. EST CE QUE JE: https://doi.org/10.1101
/cshperspect.a009274, PMID: 22553499, PMCID: PMC3331691
Nieto-Castañón, UN. (2020). Handbook of functional connectivity
magnetic resonance imaging methods in CONN. Hilbert Press.
Nieto-Castañón, UN., Ghosh, S. S., Tourville, J.. UN., & Guenther, F. H.
(2003). Region of interest based analysis of functional imaging
data. Neuroimage, 19(4), 1303–1316. EST CE QUE JE: https://est ce que je.org/10
.1016/S1053-8119(03)00188-5

Oldfield, R.. C. (1971). The assessment and analysis of handedness:
The Edinburgh inventory. Neuropsychologie, 9(1), 97–113. EST CE QUE JE:
https://doi.org/10.1016/0028-3932(71)90067-4

Pan, S. C., & Rickard, T. C. (2015). Sleep and motor learning: Is
there room for consolidation? Psychological Bulletin, 141(4),
812–834. EST CE QUE JE: https://doi.org/10.1037/ bul0000009, PMID:
25822130

Perrachione, T. K., & Ghosh, S. S. (2013). Optimized design and anal-
ysis of sparse-sampling fMRI experiments. Frontiers in Neuroscience,
18(7). EST CE QUE JE: https://doi.org/10.3389/fnins.2013.00055, PMID:
23616742, PMCID: PMC3629333

Pitt, M.. UN. (1998). Phonological processes and the perception of pho-
notactically illegal consonant clusters. Perception & Psychophysics,
60(6), 941–951. EST CE QUE JE: https://doi.org/10.3758/BF03211930, PMID:
9718954

Pugh, K. R., Frost, S. F., Sandak, R., Landi, N., Rueckl, J.. G.,
Constable, R.. T., Seidenberg, M., Fulbright, R., Katz, L., &
Mencl, W. E. (2008). Effects of stimulus difficulty and repetition
on printed word identification: An fMRI comparison of non-
impaired and reading disabled adolescent cohorts. Journal de
Neurosciences cognitives, 20(7), 1146–1160. EST CE QUE JE: https://doi.org
/10.1162/jocn.2008.20079, PMID: 18284344, PMCID:
PMC3152957

Pugh, K. R., Mencl, W. E., Jenner, UN. R., Lee, J.. R., Katz, L., Frost, S. J.,
Shaywitz, S. E., & Shaywitz, B. UN. (2001). Neurobiological studies
of reading and reading disability. Journal of Communicable
Diseases, 34(6), 479–492. EST CE QUE JE: https://doi.org/10.1016/S0021
-9924(01)00060-0

Rauschecker, UN. M., Pringle, UN., & Watkins, K. E. (2008). Changes
in neural activity associated with learning to articulate novel au-
ditory pseudowords by covert repetition. Human Brain Mapping,
29(11), 1231–1242. EST CE QUE JE: https://doi.org/10.1002/ hbm.20460,
PMID: 17948887, PMCID: PMC6870739

Riley, G. D. (2009). Stuttering severity instrument for children and

adultes (SSI-4) (4th ed.). Pro-Ed.

Robb, M., & Blomgren, M.. (1997). Analysis of F2 transitions in the
speech of stutterers and nonstutterers. Journal of Fluency

Disorders, 22, 1–16. EST CE QUE JE: https://doi.org/10.1016/S0094-730X
(96)00016-2

Robb, M., Blomgren, M., & Chen, Oui. (1998). Formant frequency
fluctuation in stuttering and nonstuttering adults. Journal de
Fluency Disorders, 23, 73–84. EST CE QUE JE: https://est ce que je.org/10.1016
/S0094-730X(97)00029-6

Rottschy, C., Langner, R., Dogan, JE., Reetz, K., Laird, UN. R., Schulz,
J.. B., & Eickhoff, S. B. (2012). Modelling neural correlates of
working memory: A coordinate-based meta-analysis. NeuroImage,
60, 830–846. EST CE QUE JE: https://doi.org/10.1016/j.neuroimage
.2011.11.050, PMID: 22178808, PMCID: PMC3288533

Sapir, S. (2014). Multiple factors are involved in the dysarthria
associated with Parkinson’s disease: A review with implications
for clinical practice and research. Journal of Speech and Hearing
Research, 57(4), 1330–1343. EST CE QUE JE: https://doi.org/10.1044/2014
_JSLHR-S-13-0039, PMID: 24686571

Segawa, J., Masapollo, M., Tong, M., Forgeron, D.J., & Guenther, F. H.
(2019). Chunking of phonological units in speech sequencing.
Brain and Language, 195, 104636. EST CE QUE JE: https://est ce que je.org/10.1016
/j.bandl.2019.05.001, PMID: 31202179, PMCID: PMC6686190
Segawa, J., Tourville, J.. UN., Beal, D. S., & Guenther, F. H. (2015).
The neural correlates of speech motor sequence learning. Journal
of Cognitive Neuroscience, 27(4), 819–831. EST CE QUE JE: https://doi.org
/10.1162/jocn_a_00737, PMID: 25313656, PMCID: PMC4344924
Skodda, S. (2012). Effect of deep brain stimulation on speech perfor-
mance in Parkinson’s disease. Parkinson’s Disease, 2012, 850596.
EST CE QUE JE: https://doi.org/10.1155/2012/850596, PMID: 23227426,
PMCID: PMC3512320

Forgeron, S. M., & Nichols, T. E. (2009). Threshold-free cluster enhance-
ment: Addressing problems of smoothing, threshold dependence
and localisation in cluster inference. NeuroImage, 44(1), 83–98.
EST CE QUE JE: https://doi.org/10.1016/j.neuroimage.2008.03.061, PMID:
18501637

Smits-Bandstra, S., De Nil, L. F., & Saint-Cyr, J.. (2006). Speech and
nonspeech sequence skill learning in adults who stutter. Journal
of Fluency Disorders, 31(2), 116–136. EST CE QUE JE: https://est ce que je.org/10
.1016/j.jfludis.2006.04.003, PMID: 16735062

Smits-Bandstra, S., & Gracco, V. (2013). Verbal implicit sequence
learning in persons who stutter and persons with Parkinson’s
maladie. Journal of Motor Behavior, 45(5), 381–393. EST CE QUE JE: https://
doi.org/10.1080/00222895.2013.812058, PMID: 23844763,
PMCID: PMC4210384

Sternberg, S., Monsell, S., Knoll, R.. L., & Wright, C. E. (1978). The latency
and duration of rapid movement sequences: Comparisons of speech
and typing. In G. E. Stelmach (Ed.), Information processing in motor
control and learning (pp. 117–152). Academic Press. EST CE QUE JE: https://
doi.org/10.1016/B978-0-12-665960-3.50011-6

Tourville, J.. UN., & Guenther, F. H. (2003). A cortical and cerebellar
parcellation system for speech studies (Boston University Technical
Report CAS/CNS-03-022). Université de Boston.

Tourville, J.. UN., Reilly, K. J., & Guenther, F. H. (2008). Neural mecha-
nisms underlying auditory feedback control of speech. NeuroImage,
39(3), 1429–1443. EST CE QUE JE: https://doi.org/10.1016/j.neuroimage
.2007.09.054, PMID: 18035557, PMCID: PMC3658624

Toyomura, UN., Fujii, T., & Kuriki, S. (2015). Effect of an 8-week
practice of externally triggered speech on basal ganglia activity
of stuttering and fluent speakers. NeuroImage, 109, 458–468.
EST CE QUE JE: https://doi.org/10.1016/j.neuroimage.2015.01.024, PMID:
25595501

Vahdat, S., Fogel, S., Benali, H., & Doyon, J.. (2017). Network-wide
reorganization of procedural memory during NREM sleep revealed
by fMRI. eLife, 11, e24987. EST CE QUE JE: https://doi.org/10.7554/eLife
.24987

Neurobiology of Language

136

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
n
o

je
/

un
r
t
je
c
e
–
p
d

F
/

2
1
1
0
6
1
8
9
9
6
7
7
n
o
_
un
_
0
0
0
2
7
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

Neural correlates of speech motor sequence learning

Vakil, E., Kahan, S., Huberman, M., & Osimani, UN. (2000). Motor
and non-motor sequence learning in patients with basal ganglia
lesions: The case of serial reaction time (SRT). Neuropsychologie,
38, 1-dix. EST CE QUE JE: https://doi.org/10.1016/S0028-3932(99)00058-5
Watkins, K. E., Forgeron, S. M., Davis, S., & Howell, P.. (2008).
Structural and functional abnormalities of the motor system in
developmental stuttering. Cerveau, 131, 50–59. EST CE QUE JE: https://est ce que je
.org/10.1093/ brain/awm241, PMID: 17928317, PMCID:
PMC2492392

Whitfield-Gabrieli, S., & Nieto-Castañón, UN. (2012). CONN: UN
functional connectivity toolbox for correlated and anticorrelated
brain networks. Brain Connectivity, 2(3), 125–141. EST CE QUE JE: https://
doi.org/10.1089/brain.2012.0073, PMID: 22642651

Wiener, N. (1949). Extrapolation, interpolation, and smoothing of
stationary time series. AVEC Presse. EST CE QUE JE: https://doi.org/10.7551
/mitpress/2946.001.0001

Lequel, Y., Jia, F., Fox, P.. T., Siok, W., & Tan, L. H. (2019). Abnormal
neural response to phonological working memory demands in
persistent developmental stuttering. Human Brain Mapping, 40(1),
214–225. EST CE QUE JE: https://doi.org/10.1002/ hbm.24366, PMID:
30145850, PMCID: PMC6865627

Yaruss, J.. S., & Conture, E. G. (1993). F2 transitions during sound/
syllable repetitions of children who stutter and predictions of
stuttering chronicity. Journal of Speech, Language, and Hearing
Research, 36(5), 883–896. EST CE QUE JE: https://doi.org/10.1044/jshr
.3605.883, PMID: 8246477