B ain & Language 240 (2023) 105264
A ailable online 21 Ap il 2023
0093-934X/© 2023 The Au ho s. Published by Else ie Inc. This is an open access a icle unde he CC BY-NC-ND license (h p://c ea i ecommons.o g/licenses/by-
nc-nd/4.0/).
Using TMS o e alua e a causal ole o igh pos e io empo al co ex in
alke -speci ic phone ic p ocessing
☆
Sahil Lu h a
a
,
*
, Hannah Mech enbe g
a
, C is al Gio io
a
, Rachel M. Theodo e
a
,
James S. Magnuson
a
,
b
,
c
, Emily B. Mye s
a
a
Uni e si y o Connec icu , Uni ed S a es
b
BCBL. Basque Cen e on Cogni ion B ain and Language, Donos ia-San Sebas i´
an, Spain
c
Ike basque, Basque Founda ion o Science, Bilbao, Spain
ARTICLE INFO
Keywo ds:
Speech pe cep ion
Talke a iabili y
Phone ic p ocessing
Righ hemisphe e
TMS
ABSTRACT
Theo ies sugges ha speech pe cep ion is in o med by lis ene s’ belie s o wha phone ic a ia ion is ypical o a
alke . A p e ious MRI s udy ound igh middle empo al gy us (RMTG) sensi i i y o whe he a phone ic
a ian was ypical o a alke , consis en wi h li e a u e sugges ing ha he igh hemisphe e may play a key ole
in condi ioning phone ic iden i y on alke in o ma ion. The cu en wo k used ansc anial magne ic s imula ion
(TMS) o es whe he he RMTG plays a causal ole in p ocessing alke -speci ic phone ic a ia ion. Lis ene s
we e exposed o alke s who di e ed in how hey p oduced oiceless s op consonan s while TMS was applied o
RMTG, le MTG, o scalp e ex. Lis ene s subsequen ly showed nea -ceiling pe o mance in indica ing which o
wo a ian s was ypical o a ained alke , ega dless o p e ious s imula ion si e. Thus, e en hough he RMTG
is ec ui ed o alke -speci ic phone ic p ocessing, modula ion o i s unc ion may ha e only modes
consequences.
1. In oduc ion
The speech signal simul aneously con eys linguis ic in o ma ion,
including phone ic in o ma ion abou which pa icula consonan s and
owels a e being p oduced and alke in o ma ion abou he pe son
p oducing hose speech sounds (e.g., Abe c ombie, 1967). Classic neu-
opsychological da a sugges a leas pa ial sepa abili y be ween pho-
ne ic p ocessing and alke p ocessing, as pa ien s wi h le hemisphe e
damage o en exhibi selec i e impai men s in speech pe cep ion bu no
ocal iden i y p ocessing, whe eas pa ien s wi h igh hemisphe e
damage o en exhibi impai men s in impai men s in ocal iden i y
p ocessing bu no speech pe cep ion (Van Lancke & Can e , 1982;
We nicke, 1874). Mo e gene ally, con empo a y neu obiological ac-
coun s hold ha phone ic p ocessing is la gely suppo ed by he le
hemisphe e while ocal iden i y in o ma ion is la gely p ocessed by he
igh hemisphe e, hough ea ly acous ic–phone ic analysis has been
shown o ec ui he empo al co ex bila e ally (Hickok & Poeppel,
2000; Maguinness, Roswandowi z, & on K iegs ein, 2018).
While he dis inc ion be ween phone ic p ocessing and alke
p ocessing can be use ul, i is also an o e simpli ica ion. Indi idual
alke s di e in how hey p oduce hei speech sounds (Allen, Mille , &
DeS eno, 2003; Hillenb and, Ge y, Cla k, & Wheele , 1995; Libe man,
Coope , Shankweile , & S udde -Kennedy, 1967; Newman, Clouse, &
Bu nham, 2001; Pe e son & Ba ney, 1952), and lis ene s appea o
capi alize on he s uc u e in his a iabili y, o ming alke -speci ic
gene a i e models – ha is, se s o belie s o how di e en alke s end
o p oduce hei speech sounds (Kleinschmid , 2019). A la ge body o
e idence indica es ha lis ene s can op imally capi alize on hei
knowledge o a alke ’s idiosync asies o guide speech pe cep ion (Cla-
ya ds, Tanenhaus, Aslin, & Jacobs, 2008; Theodo e & Mon o, 2019), and
amilia i y wi h a alke ’s idiolec can acili a e speech pe cep ion (e.g.,
ecognizing speech in noise; Nygaa d e al., 1994; Souza, Gehani,
W igh , & McCloy, 2013) as well as ocal iden i y ecogni ion (Ganu-
gapa i & Theodo e, 2019).
To illus a e his phenomenon, i is use ul o conside a se o s udies
on lis ene sensi i i y o alke -speci ic di e ences in oice-onse - ime
(VOT). VOT is an acous ic–phone ic p ope y de ined as he amoun o
ime be ween he elease o a s op consonan and he onse o ocal old
☆
This pape is based on he Disse a ion Awa d a he 13 h Socie y o he Neu obiology o Language (SNL) sponso ed by B ain and Language.
* Co esponding au ho a : Depa men o Psychology, Ca negie Mellon Uni e si y, Pi sbu gh, PA 15213-3815, Uni ed S a es.
E-mail add ess: [email p o ec ed] (S. Lu h a).
Con en s lis s a ailable a ScienceDi ec
B ain and Language
jou nal homepage: www.else ie .com/loca e/b&l
h ps://doi.o g/10.1016/j.bandl.2023.105264
Recei ed 2 Decembe 2021; Recei ed in e ised o m 6 Ap il 2023; Accep ed 8 Ap il 2023
B ain and Language 240 (2023) 105264
2
ib a ion, and i is a p ima y cue o dis inguishing oiced s op conso-
nan s (/b/, /d/ and /g/) om hei oiceless coun e pa s (/p/, / / and
/k/, espec i ely). Talke s can di e subs an ially in he p ecise VOTs
hey use o cue oiceless s op consonan s (Allen e al., 2003), e en a e
accoun ing o o he ac o s ha can a ec VOT, such as speaking a e
(Kessinge & Blums ein, 1997; Mille , 1989). Allen and Mille (2004)
demons a ed ha lis ene s a e sensi i e o hese alke -speci ic di e -
ences. In hei s udy, lis ene s we e exposed o wo alke s, one o whom
p oduced he sound / / wi h a ela i ely long VOT and one o whom
p oduced i wi h a ela i ely sho VOT; no ably, bo h a ian s we e s ill
unambiguously iden i ied as / /. A e exposu e o hese wo alke s,
lis ene s we e able o explici ly indica e which o wo a ian s (long- o
sho -VOT) was ypical o each alke . Addi ional wo k in his domain
has shown ha hese judgmen s can gene alize ac oss place o a icu-
la ion (i.e., ha a alke who p oduces /k/ wi h a long VOT is likely o
p oduce o he oiceless s ops wi h a long VOT; Theodo e & Mille ,
2010). Thus, exposu e o a alke ’s idiosync a ic s yle o speaking leads
lis ene s o make adjus men s o a alke -speci ic gene a i e model,
allowing hem o make explici judgmen s abou whe he a p oduc ion is
ypical o a ypical o a pa icula alke .
The neu al sys ems ha suppo his alke -speci ic phone ic p o-
cessing emain ela i ely unde speci ied; howe e , he igh pos e io
empo al co ex is a p omising candida e egion ha may suppo a
lis ene ’s abili y o con ac alke -speci ic gene a i e models (Lu h a,
2021). Func ional neu oimaging s udies ha e implica ed he igh pos-
e io empo al co ex in bo h phone ic p ocessing and alke p ocessing
(Belin, Za o e, La aille, Ahad, & Pike, 2000; Kennedy-Higgins, De lin,
Nu all, & Adank, 2020; Tu kel aub & B anch Cosle , 2010; on
K iegs ein & Gi aud, 2004), and s ikingly, neu al decoding s udies
indica e ha po ions o he igh pos e io empo al co ex suppo he
classi ica ion o speech s imuli along bo h phone ic and alke di-
mensions (Fo misano, De Ma ino, Bon e, & Goebel, 2008; Lu h a,
Magnuson, & Mye s, 2023).
Fu he e idence o a po en ial ole o igh empo al co ex in
alke -speci ic phone ic p ocessing comes om an MRI s udy by Mye s
and Theodo e (2017), who in es iga ed he neu al mechanisms h ough
which amilia i y wi h a alke ’s idiolec can in luence speech pe cep-
ion. P io o scanning, lis ene s we e exposed o wo alke s who p o-
duced he wo ds gain and cane. Following he Allen and Mille (2004)
s udy desc ibed abo e, one alke p oduced he sound /k/ in cane
([keɪn]) wi h a sho VOT and one p oduced i wi h a long VOT, and
lis ene s showed high accu acy when asked o explici ly indica e which
o wo a ian s was ypical o each alke . In he scanne , lis ene s
comple ed a phone ic ca ego iza ion ask wi h he gain and cane s imuli.
C i ically, lis ene s hea d bo h long-VOT and sho -VOT a ian s o cane
o each alke , meaning ha hey hea d bo h ypical and a ypical a -
ian s. Mye s and Theodo e ound ha he esponse o he igh pos e io
empo opa ie al co ex – speci ically, a clus e in he igh pos e io
middle empo al gy us (MTG) ex ending in o he igh supe io empo al
gy us (STG) and igh angula gy us (AG) – depended on whe he he
a ian was ypical o he alke , e en hough he scanne ask did no
equi e making alke ypicali y judgmen s.
An open ques ion, howe e , is whe he ec ui men o he igh
pos e io empo al co ex is necessa y o con ac ing a lis ene ’s belie s
abou how a alke ypically p oduces hei speech sounds. In he cu en
s udy, we le e aged ansc anial magne ic s imula ion (TMS) o es his
ques ion di ec ly. Speci ically, we es ed whe he magne ic s imula ion
applied du ing exposu e o a alke ’s oice impac ed lis ene s’ abili y o
judge phone ic a ian s as ypical o a ypical o he alke du ing a
subsequen es phase. O in e es was how pe o mance would be
impac ed by s imula ion o igh pos e io empo al co ex as compa ed
o s imula ion o he co esponding egion in he le hemisphe e and
s imula ion o a con ol si e ( he e ex o he scalp). Impo an ly, ou
goal is no o TMS o dis up he encoding o phone ic in o ma ion o
lis ene s’ abili y o iden i y a alke — ins ead, we hypo hesize ha
s imula ion o he igh pos e io empo al co ex may impac a lis-
ene ’s abili y o link phone ic in o ma ion and alke in o ma ion, as
measu ed h ough a alke ypicali y judgmen pos es (collec ed a e
he exposu e phase).
The cu en s udy comp ises wo expe imen s. In Expe imen 1, we
sough o alida e ou beha io al pa adigm o he TMS expe imen ,
speci ically es ing whe he lis ene s we e able o show alke -speci ic
lea ning o h ee pai s o alke s. In Expe imen 2, we applied TMS o
h ee di e en si es (RMTG, LMTG, e ex) o e he cou se o he
expe imen , using a di e en pai o alke s o each s imula ion si e,
and assessed he consequences o s imula ion o de e mining which
phone ic a ian s we e ypical o a alke . S imuli, da a, and analysis
code o all expe imen s a e a ailable a h ps://os .io/c 9 8/.
2. Expe imen 1
P io o conduc ing an expe imen wi h TMS, we conduc ed an on-
line expe imen o e i y ha lis ene s could show alke -speci ic
lea ning wi h a ask design closely based on p e ious s udies (Allen &
Mille , 2004; Mye s & Theodo e, 2017; Theodo e & Mille , 2010). In
Expe imen 1, lis ene s we e exposed o h ee pai s o alke s (i.e., six
di e en alke s), wi h each pai consis ing o a male alke and a emale
alke . Wi hin each pai , one alke p oduced hei oiceless s op con-
sonan s (/p/, / / o /k/) wi h a ela i ely sho VOT and he o he alke
p oduced he same consonan wi h a ela i ely long VOT. Because we
would ul ima ely adminis e TMS a h ee s imula ion si es o each
pa icipan (wi h lis ene s hea ing a di e en pai o alke s o each
s imula ion si e), we speci ically aimed o es ablish ha lis ene s could
show alke -speci ic lea ning o each pai o alke s, wi h minimal
gene aliza ion om one pai o alke s o he nex . Du ing aining,
lis ene s would hea h ee pai s o alke s, each pai consis ing o a male
and emale alke . A es , lis ene s would indica e o he emale alke
only whe he okens wi h long s sho VOT we e mo e ypical o ha
alke . This modi ica ion om p io designs (which ha e used only wo
same-sex alke s) would enable us o ensu e ha lea ning e ec s in he
TMS expe imen (Expe imen 2) e lec ed only he in luence o he si e
being s imula ed, a he han an a e e ec om s imula ion a a p e-
ious si e.
2.1. Me hods
2.1.1. S imulus cons uc ion
We i s selec ed h ee minimal pai con inua di e ing in VOT ha
had been used in p e ious s udies. Speci ically, we selec ed (1) a bowl-
pole con inuum o iginally cons uc ed o Theodo e and Mille (2010),
(2) a dime- ime con inuum om Allen and Mille (2004), and (3) a gain-
cane con inuum om Theodo e and Mille (2010). The alke s o hese
con inua we e all women and o he sake o his s udy a e e e ed o as
Ca ol, Joanne, and Sheila, espec i ely. No e ha he con inua di e in
he place o a icula ion o he ini ial consonan (labial, al eola , ela ,
espec i ely) as well as in he ollowing owel; by choosing con inua
wi h phonologically dissimila wo ds, we aimed o discou age gene al-
iza ion om alke o alke .
The oiced endpoin o each con inuum (i.e., bowl, dime, gain) was
syn hesized h ough an LPC analysis o na u al okens p oduced by a
Table 1
Voice-onse - ime (VOT) alues o he s imuli used in his s udy.
Talke s Con inuum Voice-Onse -Time (ms)
Voiced Sho -VOT Long-VOT
Al in/Ca ol bowl/pole 20 T ain: 60, 70
Tes : 65
T ain: 150, 160
Tes : 155
Don/Joanne dime/ ime 15 T ain: 70, 80
Tes : 75
T ain: 160, 170
Tes : 165
Pe e /Sheila gain/cane 20 T ain: 80, 90
Tes : 85
T ain: 170, 180
Tes : 175
S. Lu h a e al.
B ain and Language 240 (2023) 105264
3
di e en emale na i e speake o English. Each successi e con inuum
s ep was c ea ed by i e a i ely modi ying pa ame e s o he LPC analysis
o u n oiced ames in o oiceless ames so as o inc ease VOT ac oss
successi e s eps. Fo addi ional de ails on he cons uc ion o hese
s imuli, he eade is e e ed o he s udies o which hey we e o igi-
nally cons uc ed (Allen & Mille , 2004; Theodo e & Mille , 2010).
We selec ed se e al okens om each con inuum o use in he cu -
en s udy, choosing a oiced oken, h ee oiceless okens wi h ela-
i ely sho VOTs, and h ee oiceless okens wi h ela i ely long VOTs.
The speci ic VOT alues o he s imuli (Table 1) we e chosen based on
VOT alues used in he s udies om which he s imuli we e selec ed;
no e ha he mo e pos e io he place o a icula ion, he longe he VOT
o he oiceless s imuli we chose, consis en wi h how hese s imuli a e
p oduced na u ally (Liske & Ab amson, 1964). Following p e ious
s udies using his pa adigm (e.g., Allen & Mille , 2004), we selec ed wo
sho -VOT a ian s (e.g., o he pole con inuum, we selec ed one 60 ms
VOT a ian and one 70 ms a ian ) and wo long-VOT s eps o use
du ing aining; he VOTs o aining a ian s di e ed by 10 ms and
allowed us o simula e wi hin- alke a iabili y. A es , lis ene s did no
hea he same okens as had been p esen ed a aining bu ins ead hea d
an in e media e one (e.g., pole wi h a 65 ms VOT).
We hen made se e al modi ica ions o he selec ed con inuum s eps
using P aa (Boe sma & Weenik, 2017). Fi s , o dec ease he pe cep ual
simila i y be ween he emale alke s, we shi ed he pi ch con ou o
Ca ol’s s imuli down by 15 Hz and he pi ch con ou o Sheila’s s imuli
up by 30 Hz. Subsequen ly, we c ea ed h ee male alke s by applying
he “Change ocal ac size, pi ch, and du a ion” unc ion in he P aa
Vocal Toolki . A male alke named Al in was syn hesized by ans-
o ming Ca ol’s s imuli; speci ically, we applied a o man shi a io o
0.85, se a new pi ch median o 100 Hz, and se he pi ch a ia ion o
Al in’s oice o be 80 % o Ca ol’s. A male alke named Don was
de i ed om Joanne’s oice by applying a o man shi a io o 0.80
and a median pi ch alue o 85 Hz; no change was made o he pi ch
a ia ion. Finally, a male alke named Pe e was c ea ed by applying a
o man shi a io o 0.70 o Sheila’s speech and se ing a median pi ch
alue o 126 Hz. No e ha no changes in s imulus du a ion we e
in oduced du ing his s ep. The pi ch and o man shi a io manipu-
la ions esul ed in a plausibly male oice o each ans o ma ion.
Owing o he pa icula way in which he VOT con inua we e con-
s uc ed, s imuli wi h sho e VOTs (and he e o e longe owels) we e
associa ed wi h highe o e all ampli ude han s imuli wi h longe VOTs
(see Allen & Mille , 2004). As a esul , sho -VOT okens had a mean
oo -mean-squa e (RMS) ampli ude o 0.070 Pa, whe eas long-VOT
s imuli had a mean RMS ampli ude o 0.055 Pa. We ollowed he same
app oach as Allen and Mille o ensu e ha VOT was no con ounded
wi h ampli ude; speci ically, we c ea ed bo h a high-ampli ude e sion
(RMS ampli ude se o 0.070 Pa) and low-ampli ude e sion (RMS
ampli ude was se o 0.055 Pa) o each oken, and bo h ampli ude
a ian s we e p esen ed h oughou he expe imen .
2.1.2. S imulus p e es
P io o conduc ing Expe imen 1, we p e es ed ou s imuli o ensu e
ha ou six alke s had pe cep ually dis inc oices. Fo he p e es , we
ec ui ed 15 English-speaking monolinguals ia he online pa icipan
ec ui men pla o m P oli ic (h ps://www.p oli ic.co/); all pa ici-
pan s sel - epo ed ha hey we e cu en ly esiding in he Uni ed
S a es, had no mal o co ec ed- o-no mal ision, and had no hea ing
di icul ies and no language- ela ed diso de s. Pa icipan s comple ed a
sho sc eening es o ensu e ha hey we e wea ing headphones
(Woods, Siegel, T ae , & McDe mo , 2017). In his es , pa icipan s
mus decide which o h ee ones is quie es ; c i ically, one one is p e-
sen ed 180 deg ees ou o phase ac oss s e eo channels, such ha i is
judged o be ela i ely quie when p esen ed o e loudspeake s bu no
when p esen ed ia headphones. Thus, pe o mance on his ask di e s
depending on whe he lis ene s a e wea ing headphones o lis ening
o e hei compu e speake s. Th ee pa icipan s ailed he headphone
sc eene wice and so we e excluded om analyses, yielding a inal
sample o 12 (9 emale, 3 male; mean age: 29 yea s, age ange: 19–44
yea s). The expe imen was p og ammed using he Go illa expe imen
builde (Anwyl-I ine, Massonni´
e, Fli on, Ki kham, & E e shed, 2020),
which is well-sui ed o online expe imen s. All p ocedu es we e
app o ed by he Uni e si y o Connec icu Ins i u ional Re iew Boa d.
Each pa icipan p o ided in o med consen p io o pa icipa ing and
ecei ed mone a y compensa ion o hei ime.
Du ing he p e es , lis ene s we e i s amilia ized wi h each alke ,
hen ained o associa e each alke ’s oice o hei name, and inally
es ed on hei abili y o iden i y each alke om hei oice. The p e es
was blocked by alke sex, wi h hal he pa icipan s comple ing all h ee
phases wi h he male alke s be o e hea ing he emale alke s, and hal
he pa icipan s comple ing he p e es wi h he emale alke s i s .
Du ing he ini ial amilia iza ion pe iod, lis ene s hea d ou p o-
duc ions om each alke , wi h each alke saying a di e en wo d. Fo
hal he lis ene s, each male alke p oduced an i em wi h wo d-ini ial
oicing (i.e., Al in said bowl, Don said dime, and Pe e said gain), and
each emale alke p oduced an i em ha began wi h a oiceless con-
sonan (i.e., Ca ol said pole, Joanne said ime, and Sheila said cane). Fo
he o he hal o he lis ene s, he male alke s p oduced he i ems
beginning wi h oiceless consonan s and he emale alke s p oduced
he i ems wi h wo d-ini ial oiced consonan s. Familia iza ion was
blocked by alke and he o de o he alke s was ixed. The amilia -
iza ion pe iod had an in e -s imulus in e al (ISI) o 1000 ms.
The aining ask was a 3-al e na i e o ced choice ask in which
lis ene s hea d a single wo d on each ial and we e asked o iden i y he
alke om he se o alke s o ha sex. Lis ene s we e old i hey we e
co ec o inco ec , and i hey we e inco ec , hey we e old wha he
co ec answe was. To ensu e ha lis ene s did no lea n o dis inguish
he alke s based solely on which wo d hey we e p oducing, we also
included some s imuli ha we e no used in Expe imen 1. Speci ically,
we also included p oduc ions o gain and cane spoken by Al in and Ca ol
as well as p oduc ions o bowl and pole p oduced by Pe e and Sheila;
hese s imuli we e cons uc ed ollowing he same app oach desc ibed
abo e, and o each alke , lis ene s hea d an equal numbe o p o-
duc ions o he possible wo ds (e.g., hey hea d Al in saying gain jus as
o en as hey hea d him saying bowl). Lis ene s comple ed a o al o 96
ials (16 pe alke ), spli be ween he wo aining blocks (wi h only
alke s o he same sex p esen ed in a gi en aining block). Each alke
p oduced an equal numbe o oiced and oiceless wo ds, and o he
oiceless okens, hal had a sho VOT and hal had a long VOT. The e
was an ISI o 1000 ms.
The es phase was iden ical o he aining phase, excep lis ene s
did no ecei e eedback on he accu acy o hei esponses. As shown in
Fig. 1A, pa icipan s had high accu acy du ing he es phase, wi h mean
accu acy g ea e han 70 % o all alke s; no e ha chance-le el accu-
acy is 33 %. Con usion ma ices, shown in Fig. 1B, also indica e ha
pa icipan s we e gene ally accu a e in iden i ying he alke s, hough
he e we e some asymme ies in hei e o s. Fo ins ance, Ca ol was
some imes (23 %) misiden i ied as Joanne, bu Joanne was a ely (1 %)
iden i ied as Ca ol.
2.1.3. Pa icipan s
Fo Expe imen 1, we ec ui ed 36 pa icipan s ia he online sys em
P oli ic. These pa icipan s did no pa icipa e in he s imulus p e es
desc ibed in Sec ion 2.1.2. All Expe imen 1 pa icipan s sel - epo ed
being English-speaking monolinguals esiding in he Uni ed S a es
S. Lu h a e al.
B ain and Language 240 (2023) 105264
4
wi h no mal o co ec ed- o-no mal ision. Pa icipan s epo ed ha
hey did no ha e any hea ing di icul ies o language- ela ed diso de s.
Fi e pa icipan s ailed he headphone sc eening es wice and so we e
excluded. Da a om one addi ional pa icipan we e excluded o equa e
he numbe o pa icipan s in each coun e balancing condi ion. Thus, 30
pa icipan s (12 emale, 18 male; mean age: 32, age ange: 20–64)
1
we e
included in he analysis; his sample size was based on p e ious s udies
using his pa adigm (Allen & Mille , 2004; Mye s & Theodo e, 2017;
Theodo e & Mille , 2010), which obse ed he beha io al e ec o in-
e es wi h smalle samples ( ange: 17–20). In selec ing his sample size,
we we e also guided by a se o p e ious speech pe cep ion s udies
(Bes elmeye , Belin, & G osb as, 2011; Heim a h, Sp ¨
oggel, Repplinge ,
Heinze, & Zaehle, 2019; Kennedy-Higgins e al., 2020; Meye , Elsne ,
Tu ke , Kuhnke, & Ha wigsen, 2018; Nixon, Laza o a, Hodino -Hill,
Gough, & Passingham, 2004; Rome o, Walsh, & Papagno, 2006;
Smalle, Roge s, & M¨
o ¨
onen, 2015) ha obse ed TMS e ec s wi h a
mean sample size o 18 ( ange: 6–48).
All p ocedu es we e app o ed by Uni e si y o Connec icu Ins i u-
ional Re iew Boa d. Pa icipan s p o ided in o med consen p io o
beginning he expe imen and ecei ed mone a y compensa ion o hei
ime.
2.1.4. P ocedu e
Expe imen 1 consis ed o h ee blocks, and lis ene s hea d a
di e en pai o alke s (and hus a di e en con inuum) in each block.
Block o de (Al in/Ca ol, Don/Joanne, Pe e /Sheila) was coun e -
balanced using a La in squa e. C i ically, he alke s in a single block had
a di e en cha ac e is ic VOT o hei oiceless s op consonan s; he
speci ic VOTs a e p o ided in Table 1. To discou age gene aliza ion
ac oss blocks, he cha ac e is ic VOT o alke s o he same sex
Fig. 1. We conduc ed a p e es o ensu e ha ou six alke s had pe cep ually dis inc oices. On each es ial, lis ene s had o iden i y who was speaking om
among he se o same-sex alke s. (A) Accu acy on alke iden i ica ion p e es . Talke names a e shown on he x-axis and pe cen accu acy on he y-axis. Do s
ep esen indi idual subjec da a. E o ba s indica e s anda d e o o he mean. (B) Con usion ma ices o he alke iden i ica ion p e es . Rows indica e which
alke was speaking, and columns indica e pa icipan s’ esponses. P opo ions in a ow may no sum o 1 due o ounding.
1
Expe imen 1 used a la ge age ange (20–64) compa ed o Expe imen 2
(19–35). To ensu e ha di e ences in age ange did no d i e he esul s seen in
Expe imen 1, we also conduc ed an analysis o he Expe imen 1 da a ha only
included pa icipan s aged 35 and younge . This analysis (N=22) yielded he
same pa e ns o signi icance as compa ed o he ull sample and is epo ed in
Supplemen a y Ma e ials.
S. Lu h a e al.
B ain and Language 240 (2023) 105264
5
al e na ed ac oss blocks (always sho -long-sho o he emale alke s
and long-sho -long o he male alke s).
A he s a o each block, lis ene s we e old ha hey would be
exposed o wo alke s who di e ed in how hey p oduced a pa icula
speech sound (e.g., he /p/ sound in pole). They we e old ha hei job
was o lea n he unique way ha each alke p oduced his sound.
Du ing an ini ial amilia iza ion pe iod, lis ene s hea d eigh okens
om each alke ( ou oiced and ou oiceless). Fo he oiceless o-
kens, lis ene s only hea d he a ian s ha we e ypical o he alke .
The alke ’s name was shown on sc een as each s imulus played, and
he e was an ISI o 1000 ms. S imuli we e blocked by alke , wi h lis-
ene s always hea ing he male alke i s , and he o de o i ems p o-
duced by each alke was andomized.
Following he amilia iza ion phase, lis ene s comple ed a aining
phase and es phase (schema ized in Fig. 2A). Du ing he aining phase
o each block, lis ene s pe o med a 4-al e na i e o ced choice ask. Fo
each s imulus, lis ene s made a keyboa d esponse o indica e bo h who
was alking and wha wo d hey said. Feedback (a g een check o co -
ec esponses, a ed x o inco ec esponses) was shown on sc een o
500 ms a e lis ene s made hei esponse, and he e was a 1000 ms
in e al be ween ials. In o al, aining consis ed o 96 ials (48 pe
alke ), wi h ials p esen ed in andom o de . Lis ene s hea d an equal
numbe o oiced and oiceless okens om each alke , and hey hea d
an equal numbe o high-ampli ude and low-ampli ude e sions o each
oken.
Du ing he es po ion o he block, lis ene s hea d only he emale
alke . We op ed o es on only one alke ’s oice pe block o wo
easons, bo h ela ed o po en ial e ec s o TMS being in es iga ed in
Expe imen 2. Fi s , a p o ac ed es phase would gi e pa icipan s
addi ional exposu e o alke -a ypical a ian s, po en ially a enua ing
lea ning e ec s and possibly also encou aging gene aliza ion ac oss se s
o alke s; while his issue could in heo y be amelio a ed by a longe
aining phase, p ac ical conside a ions ela ed o he numbe o TMS
pulses ha can sa ely be deli e ed in a single session made his
imp ac ical (Rossi e al., 2009). Secondly, es ing on only one oice
allowed us o educe he numbe o a iables ha would need o be
coun e balanced (e.g., he o de in which oices we e es ed), he eby
emo ing a po en ial sou ce o be ween-subjec a iabili y and
imp o ing ou abili y o obse e po en ial e ec s o s imula ion si e
(which we manipula ed wi hin pa icipan s, as desc ibed below) wi h a
ela i ely small numbe o ials, as necessi a ed by sa e y conside -
a ions. On each ial, lis ene s hea d a sho -VOT and a long-VOT a ian
(wi h he o de o a ian s coun e balanced) and we e asked o indica e
which was mo e ypical o he alke . As no ed in Table 1, he VOT hea d
du ing es was no exac ly he same as he ones hea d du ing aining;
o ins ance, i lis ene s had hea d Sheila p oducing 80 ms and 90 ms
a ian s o /k/ du ing aining, he es phase would in ol e deciding
whe he an 85 ms o 175 ms a ian was mo e ypical o Sheila. The
ampli ude o he okens was held cons an wi hin each ial. Lis ene s
comple ed 32 ials du ing each es phase.
2.2. Resul s
Pe o mance on he aining ask is isualized in Fig. 2B. In analyzing
he aining da a, we sepa a ely assessed lis ene s’ abili y o iden i y he
alke ( ega dless o whe he hey we e co ec in iden i ying which
wo d was said) as well as hei abili y o de e mine which wo d was said
( ega dless o whe he hey we e co ec in iden i ying he alke ). Lis-
ene s we e highly accu a e in he alke decision and he phone ic de-
cision, ega dless o he alke o hei ypical VOT (mean accu acy >91
% in all cases).
Because a sho -VOT a ian migh be mo e easily con used wi h a
oiced s imulus (compa ed o a long-VOT a ian ), we s a is ically
assessed he in luence o he alke ’s ypical VOT (i.e., whe he he
alke p oduced oiceless s ops wi h a sho o long VOT) on he aining
ask; sepa a e models we e conduc ed o alke iden i ica ion
pe o mance and phone ic iden i ica ion pe o mance. These models
we e implemen ed in R (R Co e Team, 2019) using he “mixed” unc ion
in he a ex package (Singmann, Bolke , Wes all, & Aus , 2018). This
unc ion i s a mixed-e ec s model o he da a (using he lme4 package;
Ba es, Maechle , Bolke , & Walke , 2015) and e alua es he signi icance
o each ixed e ec by compa ing he ull model o a educed model
wi hou ha ixed e ec . He e, each model included a ixed ac o o
Typical VOT (long/sho , sum-coded) and andom in e cep s o each
subjec and o each alke . We speci ied a binomial amily wi h a logi
link, and we used likelihood a io es s o e alua e signi icance. The
alke ’s ypical VOT did no in luence pe o mance on he alke iden-
i ica ion componen o he ask,
χ
2
(1) =0.00, p =0.97, bu did ha e a
signi ican e ec on phone ic iden i ica ion,
χ
2
(1) =22.51, p <0.0001;
his la e e ec was d i en by sligh ly less accu a e esponses when he
alke had a sho VOT (mean: 0.95, SD: 0.22) compa ed o when he
alke had a long VOT (mean: 0.97, SD: 0.16). Tha is, sho -VOT a i-
an s we e mo e likely o be con used wi h oiced okens, bu long-VOT
a ian s we e mislabeled ela i ely less o en.
Mean o e all accu acy in he es phase was 68.9 % (SD: 16.0 %), and
esul s om he es phase a e plo ed in Fig. 2C. Visually, i is clea ha
pa icipan s we e mo e likely o selec he long-VOT a ian as he mo e
ypical one when he alke had p e iously p oduced long-VOT a ian s
du ing aining. To e alua e his s a is ically, es da a we e submi ed o
a linea mixed e ec s eg ession ha assessed how ixed ac o s o Talke
(Ca ol, Joanne, Sheila; sum-coded) and Typical VOT (long/sho ; sum-
coded) in luenced whe he pa icipan s selec ed he long-VOT a ian .
To selec ou andom e ec s uc u e, we began wi h he maximal
andom e ec s uc u e ha con e ged (Ba , Le y, Scheepe s, & Tily,
2013) and used a backwa d-s epping p ocedu e o iden i y whe he we
could use a simple model s uc u e wi hou signi ican ly comp omising
model i (Ma uschek, Kliegl, Vasish h, Baayen, & Ba es, 2017). In his
way, we selec ed a andom e ec s s uc u e wi h andom by-subjec
slopes o Typical VOT as well as andom by-subjec in e cep s. We
obse ed a signi ican e ec o Typical VOT,
χ
2
(1) =15.12, p =0.0001,
d i en by mo e long-VOT esponses i he alke ’s cha ac e is ic VOT
was long (mean: 0.56, SD: 0.50) han i i was sho (mean: 0.24, SD:
0.43). No o he e ec s we e signi ican (p >0.19).
2.3. Discussion
In Expe imen 1, lis ene s we e exposed o pai s o alke s ha
di e ed in how hey p oduced hei oiceless s op consonan s. Du ing
aining, pa icipan s demons a ed nea -ceiling pe o mance in hei
abili y o iden i y who was alking and wha wo d hey we e saying.
Recall ha o he TMS expe imen , we a e pa icula ly in e es ed in
whe he s imula ion a aining in luences he ex en o lea ning wha
a ia ion is ypical o a alke a es , a he han whe he s imula ion
in luences a lis ene ’s abili y o pe o m phone ic iden i ica ion and/o
alke iden i ica ion (as assessed du ing aining). As desc ibed in he
in oduc ion, we hypo hesized ha TMS o igh supe io empo al
co ex should impac lis ene s’ abili y o de e mine wha phone ic
a ia ion is ypical (o a ypical) o each alke (i.e., o link alke in o -
ma ion wi h phone ic de ail), no ha TMS should in luence lis ene s’
abili y o encode alke o phone ic de ail. No e ha i accu acy on he
Expe imen 1 aining ask had been below ceiling, hen any po en ial
e ec s o TMS obse ed in Expe imen 2 could be d i en (a leas in pa )
by dis up ions o he ea lie p ocesses o encoding alke in o ma ion o
phone ic de ail, a he han being d i en by speci ic dis up ions o he
p ocess o lea ning which phone ic a ian is ypical o a ypical o a
gi en alke ’s idiolec . Fo his eason, he nea -ceiling pe o mance on
he aining ask in Expe imen 1 is no a cause o conce n.
Fu he mo e, esul s om he es phase indica e ha pa icipan s
we e able o lea n he phone ic idiosync asies o mul iple alke s, as
measu ed by hei abili y o explici ly iden i y whe he a sho VOT o a
long VOT was ypical o each alke ’s p oduc ions o oiceless s op
consonan s. Speci ically, lis ene s we e signi ican ly mo e likely o selec
S. Lu h a e al.
B ain and Language 240 (2023) 105264
6
Fig. 2. Expe imen 1 design and esul s. (A) Pa icipan s comple ed a 4-AFC aining ask ha in ol ed simul aneous alke and phone ic iden i ica ion, ollowed by
a es phase whe e pa icipan s we e que ied on which phone ic a ian was ypical o a alke . (B) Pe o mance on he aining ask, sepa a ely conside ing whe he
lis ene s we e accu a e in iden i ying who was alking (“Talke Decision”) and which wo d hey said (“Phone ic Decision”). Accu acy alues a e shown on he y-axis.
Each ow shows pe o mance on a di e en block. In plo s on he le , he emale alke p oduced oiceless s op consonan s wi h a sho VOT, and in plo s on he
igh , she p oduced hese consonan s wi h a long VOT. Do s ep esen indi idual subjec da a. E o ba s indica e s anda d e o o he mean. (C) Resul s om he es
phase o Expe imen 1, showing he p obabili y ha a lis ene selec ed he long-VOT a ian (y-axis) as a unc ion o he alke (x-axis) and whe he he alke
p oduced oiceless s ops wi h long (blue ba s) o sho (o ange ba s) VOTs du ing aining. Do s ep esen indi idual subjec da a. E o ba s indica e s anda d e o
o he mean.
S. Lu h a e al.
B ain and Language 240 (2023) 105264
7
a long-VOT a ian as ypical o he alke i ha alke had p e iously
p oduced oiceless s ops wi h long VOTs han i she had p oduced hem
wi h sho VOTs. C i ically, pa icipan s did no exhibi ceiling- o loo -
le el pe o mance on he ypicali y judgmen ask du ing Expe imen 1,
wi h a mean o e all accu acy le el o 69 % on he ypicali y judgmen s;
his accu acy le el allows us o measu e bo h TMS- ela ed enhance-
men s o dis up ions in alke -speci ic phone ic lea ning. Finally, he
ac ha we obse ed obus alke -speci ic lea ning o all alke s we
es ed sugges s ha his is a alid pa adigm o ou TMS expe imen
(Expe imen 2).
Howe e , i is wo h no ing ha when he alke had p e iously
p oduced long-VOT oiceless s op consonan s, lis ene s appea ed o be
close o chance in hei endency o selec he long-VOT a ian , as
illus a ed in Fig. 2C. We e i ied his h ough a one-sample - es ( s
chance) conduc ed on he subjec -by-subjec p opo ions o long-VOT
esponses in he long-VOT condi ion, (29) =0.90, p =0.37. We sug-
ges ha his esul e lec s lis ene s’ gene al p e e ence o sho VOTs,
as hese a e mo e ypical o oiceless s op consonan s in gene al. Tha is,
sho -VOT a ian s a e a be e i o he English /p/, / / and /k/ pho-
ne ic ca ego ies han a e long-VOT a ian s. None heless, he ac ha
Fig. 3. (A) Mye s and Theodo e (2017) ound ha he ac i a ion o igh pos e io empo al co ex du ing an in-scanne phone ic ca ego iza ion ask was modula ed
by whe he a phone ic a ian was ypical o a ypical o he alke . (B) The cu en TMS expe imen in ol ed h ee s imula ion si es: he igh empo al egion
iden i ied by Mye s and Theodo e, he homologous le hemisphe e egion, and scalp e ex (con ol si e; no isualized he e). (C) Fo each s imula ion si e, pa -
icipan s comple ed a aining (4AFC) ask and a es (2AFC) ask. Sample sc een displays a e shown, illus a ing how esponses mapped on o he se en-bu on
esponse box used in he expe imen . (Pa icipan s esponded by p essing he bu ons wi h oam ci cles.) S imula ion was adminis e ed p io o each aining
ial, wi h no TMS adminis e ed du ing he es phase.
S. Lu h a e al.
B ain and Language 240 (2023) 105264
8
lis ene s made mo e long-VOT esponses in he long-VOT condi ion han
in he sho -VOT condi ion sugges s ha hey did indeed lea n wha was
ypical o he alke (e en i hey we e hesi an o make long-VOT e-
sponses in gene al because o he low base a e o long-VOT oiceless
s ops in English). Based on his esul , howe e , we decided ha in
Expe imen 2, we would o ally emphasize o pa icipan s ha hey
should make hei esponse wi h ega d o which es a ian was mo e
ypical o he alke a he han mo e ypical o he phone ic ca ego y.
This would be easie o do in Expe imen 2 as i was conduc ed in pe son,
unlike Expe imen 1.
Addi ionally, we no e ha he e was no signi ican in e ac ion be-
ween Talke and Typical VOT in Expe imen 1, sugges ing ha he
deg ee o lea ning was compa able ac oss all h ee pai s o alke s.
Visually, howe e , Fig. 2C sugges s ha he deg ee o lea ning may ha e
been la ge o some alke s han o o he s; when lis ene s hea d
Joanne, o ins ance, hey made long-VOT esponses 70 % o he ime o
long-VOT a ian s and 20 % o he ime o sho -VOT a ian s, bu when
lis ene s hea d Ca ol, hey made long-VOT esponses 43 % o he ime o
long-VOT a ian s and 23 % o he ime o sho -VOT a ian s. Despi e
he lack o a signi ican s a is ical in e ac ion be ween hese ac o s, we
decided o e on he side o cau ion and op ed o coun e balance which
se o alke s (Al in/Ca ol, Don/Joanne, Pe e /Sheila) was associa ed
wi h which pa icula s imula ion si e (RMTG, LMTG, e ex) in
Expe imen 2.
3. Expe imen 2
In Expe imen 2, pa icipan s we e asked wi h lea ning he phone ic
signa u es o h ee pai s o alke s, as in Expe imen 1. In p e ious wo k,
Mye s and Theodo e (2017) ound ha a pos e io RMTG clus e was
sensi i e o whe he he phone ic a ian hea d du ing a phone ic
ca ego iza ion ask was ypical o a ypical o a alke ; his clus e is
shown in Fig. 3A. In he cu en s udy, apid TMS was deli e ed o a
di e en s imula ion si e (RMTG, LMTG, e ex) du ing each aining
block; hese clus e s a e isualized in Fig. 3B. Each subjec ecei ed
s imula ion a all h ee si es o e he cou se o he expe imen , wi h si e
o de coun e balanced ac oss pa icipan s. O in e es was how TMS a
each si e would a ec pe o mance du ing he subsequen es phase; we
hypo hesized ha TMS o he igh MTG would in luence a pa icipan ’s
abili y o de e mine wha phone ic a ia ion was ypical o each alke ,
as measu ed du ing es . No e ha we op ed o compa e pe o mance o
a con ol ( e ex) si e ins ead o applying sham s imula ion; he appli-
ca ion o TMS o he speci ic empo al si es used in his s udy can esul
in pa icipan s expe iencing mild jaw wi ches due o di ec s imula ion
o acial muscles, making i di icul o apply a con incing sham
s imula ion.
3.1. Me hods
3.1.1. S imuli
We used he same s imuli as in Expe imen 1.
3.1.2. Pa icipan s
Thi y-one igh -handed na i e speake s o Ame ican English we e
ec ui ed om he Uni e si y o Connec icu communi y. Pa icipan s
epo ed ha ing no mal o co ec ed- o-no mal ision, no hea ing loss
and no his o y o neu ological impai men . Each pa icipan was
sc eened o MRI and TMS con aindica ions ollowing es ablished
sa e y p o ocols (Rossi e al., 2009). Da a om ou pa icipan s had o
be excluded due o a p og amming e o . Analyses he e o e ep esen
da a om 27 pa icipan s (20 emale, 7 male, mean age: 24, age ange:
19–35). All pa icipan s p o ided in o med consen p io o pa ici-
pa ing and ecei ed mone a y compensa ion o hei ime. No pa ici-
pan s who pa icipa ed in Expe imen 1 pa icipa ed in Expe imen 2.
All p ocedu es we e app o ed by he Uni e si y o Connec icu Ins i u-
ional Re iew Boa d.
3.1.3. P ocedu e
Expe imen 2 was conduc ed o e wo sessions, bo h o which ook
place a he B ain Imaging Resea ch Cen e a he Uni e si y o Con-
nec icu . Du ing he i s session, we i s acqui ed a T1-weigh ed
s uc u al magne ic esonance image (unless we al eady had such an
image on ile o he pa icipan om a p e ious s udy). Ana omical
images we e acqui ed on a 3-T Siemens P isma scanne wi h a 64-chan-
nel head coil using a T1-weigh ed magne isa ion-p epa ed apid acqui-
si ion g adien echo (MP-RAGE) sequence (TR =2400 ms, TE =2.15 ms,
FOV =256 mm, lip angle =8 deg ees, 1 mm sagi al slices). These
images we e used in conjunc ion wi h he Locali e TMS Na iga o
(Locali e, S . Augus in, Ge many) o moni o he TMS coil posi ion
ela i e o each s imula ion si e.
To iden i y he app op ia e le el o s imula ion o each pa icipan ,
we de e mined each pe son’s es ing mo o h eshold – ha is, he
minimal amoun o s imula ion ha mus be applied o he mo o hand
a ea o elici a eliable muscle esponse in he hand. The mo o hand
a ea in he le hemisphe e was iden i ied h ough isual inspec ion o
he pa icipan ’s b ain ana omy (Yous y e al., 1997). The muscle ac-
i i y o he con ala e al humb was eco ded while we s imula ed he
mo o hand a ea and nea by b ain egions; he loca ion a which we
elici ed he s onges esponse was iden i ied as he mo o ho spo
(Ahdab, Ayache, B ugi`
e es, Fa ha , & Le aucheu , 2016). The mo o
h eshold a he ho spo was hen de e mined using he Mo o Th eshold
Assessmen Tool (Awiszus & Bo cka d , 2011), which includes an
adap i e Pa ame e Es ima ion by Sequen ial Tes ing (PEST) p ocedu e
o de e mining mo o h esholds. Mo o -e oked po en ials we e
eco ded using a Biopac MP160 sys em (Biopac Sys ems Inc., Gole a,
CA), and s imula ion was deli e ed using a MagP o X100 TMS de ice
wi h a dynamically cooled bu e ly double coil in combined ac i e and
sham (Cool B-65 A/P) con igu a ion (MagVen u e, Inc., A lan a, GA).
Mo o h esholding was pe o med du ing he i s session o mos
pa icipan s, hough o h ee pa icipan s, i was pe o med a he s a
o he second session. These mo o h esholds ha e been shown o be
eliable wi hin a pa icipan as well as ac oss sessions (Va na a, S okes,
& Chambe s, 2011).
Pa icipan s comple ed he expe imen al ask a hei second session,
ollowing a simila p ocedu e o ha o Expe imen 1. The expe imen al
pa adigm is summa ized in Fig. 3C. On aining ials, pa icipan s made
a 4-al e na i e o ced-choice decision, indica ing bo h who was alking
(e.g. “Al in” o “Ca ol”) and wha hey we e saying (e.g. “bowl” o
“pole”); no e ha du ing aining, lis ene s hea d bo h he male and
emale alke s. On es ials, pa icipan s indica ed which o wo a i-
an s was mo e ypical o he alke ; no e ha du ing es , lis ene s only
hea d he emale alke s, as in Expe imen 1. On es ials, pa icipan s
esponded wi h hei le hand o indica e ha he i s a ian was mo e
ypical o he alke and wi h he igh hand i he second a ian was
mo e ypical; any esponse on he le side o he bu on box was coded
as a i s - a ian esponse, and any igh -side esponse was coded as a
second- a ian esponse.
TMS was adminis e ed online du ing he aining po ion o each
block using a s imula ion p o ocol ha was consis en wi h es ablished
sa e y ecommenda ions (Rossi e al., 2009). Speci ically, p io o each
aining ial, we adminis e ed i e biphasic bu s TMS pulses a a 10 Hz
equency; his s imula ion a e was based on p e ious s udies in which
10 Hz s imula ion o he empo al co ex led o impai men s in ocal
iden i y p ocessing (Bes elmeye e al., 2011) and speech pe cep ion
(Kennedy-Higgins e al., 2020). We used a 5000 ms ITI in be ween
aining ials, in con as o he 1000 ms ITI used in Expe imen 1.
S imula ion in ensi y was se o 90 % o he pa icipan ’s es ing mo o
h eshold; his co esponded o a mean o 49 % o he maximum s im-
ula ion ou pu (MSO), wi h a ange o 37–65 % MSO. Occasionally, TMS
o he empo al lobes esul ed in pa icipan s expe iencing jaw wi ches,
due o di ec s imula ion o acial muscles; hough conside ed a negli-
gible sa e y isk (Rossi e al., 2009), we checked in wi h any pa icipan
who expe ienced hese wi ches o ensu e hey we e no expe iencing
S. Lu h a e al.
B ain and Language 240 (2023) 105264
9
se e e discom o and s ill wished o con inue.
Du ing each expe imen al block, s imula ion was applied o a
di e en si e (RMTG, LMTG, e ex), wi h s imula ion si e o de coun-
e balanced using a La in squa e. Fo each pa icipan , he RMTG
s imula ion si e was de ined by p ojec ing he unc ionally de ined
RMTG clus e om Mye s and Theodo e (2017) om Talai ach and
Tou noux (1988) space on o each subjec ’s indi idual ana omy using
he 3dF ac ionize command in AFNI (Cox, 1996); ecall ha Mye s and
Theodo e ound ha his RMTG clus e was sensi i e o whe he pho-
ne ic a ian s we e ypical o a ypical o a alke in hei phone ic
ca ego iza ion ask. No e ha we could no ask pa icipan s in ou s udy
o comple e a unc ional localize , as asking pa icipan s o do he Mye s
and Theodo e (2017) ask would ha e equi ed hem o lea n wha
phone ic a ia ion was ypical o each alke — exac ly he p ocess we
hoped o dis up wi h TMS. Howe e , o he s udies ha e success ully
obse ed modula o y e ec s o TMS a e localizing s imula ion si es
om ana omical MRI scans (Kennedy-Higgins e al., 2020; Meye e al.,
2018; Nixon e al., 2004; Rome o e al., 2006). The LMTG si e was
de ined by iden i ying he homologous si e in he le hemisphe e, and
he scalp e ex was iden i ied isually using he Locali e na iga ion
so wa e. To isualize he le and igh s imula ion si es (Fig. 3B), we
d ew a sphe e wi h an 8-mm adius a ound each subjec ’s s imula ion
si e and p ojec ed each sphe e in o Talai ach and Tou noux space; we
hen o e laid he di e en subjec -speci ic s imula ion si es.
Fo he expe imen al ask, s imuli we e deli e ed h ough a Focus i e
Sca le 2i2 digi al audio in e ace (High Wycombe, England) coupled o
a pai o ER-3C inse headphones wi h oam ea ips (E ymo ic Resea ch,
Elk G o e Village, IL). This se up allowed he pa icipan s o hea he
s imuli while also p o iding hea ing p o ec ion agains he acous ic
clicks o he TMS coil. The expe imen was p og ammed in OpenSesame
(Ma hˆ
o , Sch eij, & Theeuwes, 2012), and pa icipan s made hei e-
sponses ia a handheld bu on box.
3.2. Resul s
O e all pe o mance on he aining ask is isualized in Fig. 4. F om
Fig. 4A, i is clea ha pe o mance was high ac oss bo h he alke
iden i ica ion and phone ic iden i ica ion componen s o he ask, bu
s ikingly, phone ic iden i ica ion pe o mance appea s o ha e been
modes ly impai ed when pa icipan s ecei ed s imula ion o RMTG.
Speci ically, lis ene s we e sligh ly less accu a e in deciding which wo d
hey hea d a e RMTG s imula ion (mean: 0.96, SD: 0.19) compa ed o
LMTG (mean: 0.99, SD: 0.11) and con ol (mean: 0.98, SD: 0.14)
s imula ion.
To assess his s a is ically, ial-le el da a om he aining ask we e
submi ed o logis ic mixed e ec s eg ession analyses. As in Expe imen
1, sepa a e analyses conside ed he likelihood o co ec ly iden i ying
he alke e sus he likelihood o making he co ec phone ic decision.
Fo each analysis, we i s i he da a using a model ha included ixed
ac o s o S imula ion Si e (le / igh / e ex; sum-coded) and Typical
VOT (long/sho ). The i o his model was compa ed o ha o a
simple model, which jus es ed o a ixed e ec o S imula ion Si e; he
simple model was p e e ed only i i did no en ail a signi ican loss in
he goodness-o - i be ween he model and he da a (Ma uschek e al.,
2017). Bo h models included andom in e cep s o each subjec as well
as o each alke . This p ocedu e led us o selec he simple model o
he alke iden i ica ion analysis and he mo e complex model o he
phone ic iden i ica ion analysis. Fo all models, we speci ied a binomial
amily wi h a logi link, and we used likelihood a io es s o e alua e
signi icance.
Talke iden i ica ion abili y was no signi ican ly a ec ed by s imu-
la ion si e,
χ
2
(1) =2.30, p =0.32; o all s imula ion si es, mean accu-
acy was g ea e han 99 %.
While accu acy on he phone ic iden i ica ion componen o he ask
was also high, pe o mance was in luenced by ou ac o s o in e es .
Speci ically, a ma ginal e ec o S imula ion Si e,
χ
2
(1) =5.60, p =0.06,
as well as a signi ican e ec o Typical VOT (long o sho ),
χ
2
(1) =
32.35, p <0.0001, d i en by highe accu acy when he alke p oduced
hei oiceless s ops wi h a long VOT (mean: 0.98, SD: 0.12) compa ed
o when hey p oduced hem wi h a sho VOT (mean: 0.97, SD: 0.18).
We also obse ed a signi ican in e ac ion be ween he wo ac o s,
χ
2
(1) =9.30, p =0.01, isualized in Fig. 4B.
To u he p obe he ma ginal e ec o S imula ion Si e, we con-
duc ed ollow-up pai wise compa isons o each o ou s imula ion si es;
his was implemen ed using he emmeans package (Len h, 2021), and p
alues we e Tukey-adjus ed o co ec o mul iple compa isons. This
analysis sugges ed ha he e was a ma ginal di e ence be ween he
phone ic iden i ica ion accu acy o RMTG s imula ion compa ed o
LMTG s imula ion, p =0.05, bu nonsigni ican di e ences o he o he
wo pai s (LMTG s e ex: p =0.11; RMTG s e ex: p =0.93).
To ollow-up on he signi ican in e ac ion be ween S imula ion Si e
and Typical VOT, we used he emmeans package o e alua e he e ec o
S imula ion Si e o each le el o Typical VOT. We ound ha o alke s
wi h long-VOT oiceless s ops, pa icipan s we e mos accu a e when
ecei ing LMTG s imula ion (LMTG s RMTG: p =0.01; LMTG s e ex:
p =0.02; RMTG s e ex: p =0.96), bu no pai wise di e ences we e
obse ed o alke s wi h sho -VOT oiceless s ops (LMTG s RMTG: p
=1.00; LMTG s e ex: p =0.95; RMTG s e ex: p =0.93).
Fig. 4. Pe o mance on he aining ask (du ing which TMS was applied) in
Expe imen 2. In panel (A), we sepa a ely show pe o mance on he alke
decision and phone ic decision componen s o he ask. Panel (B) shows only
he phone ic iden i ica ion componen o he ask, sepa a ely conside ing
whe he he alke p oduced long-VOT oiceless s ops (le ba s) o sho -VOT
oiceless s ops ( igh ba s). Fo bo h panels, accu acy alues a e shown on he
y-axis. Ba colo indica es whe he s imula ion was applied o le MTG (o ange
ba s), igh MTG (g een ba s) o scalp e ex (blue ba s) o ha block. Do s
ep esen indi idual subjec da a. E o ba s indica e s anda d e o o
he mean.
S. Lu h a e al.
