(cc) 2024 Ab aham Sánchez Sánchez (cc by-nc-nd 4.0)
Concep s a e buil om pe cep s
Neu al mechanisms suppo ing sub-lexical o lexico-seman ic
p ocessing
Ab aham Sánchez
Supe ised by
Manuel Ca ei as & Ped o M. Paz-Alonso
2024
Concep s a e buil om pe cep s
Neu al mechanisms suppo ing sub-lexical o lexico-seman ic
p ocessing
Ab aham Sánchez
Supe ised by
Manuel Ca ei as & Ped o M. Paz-Alonso
2024
I
ACKNOWLEDGEMENTS
Mos ques ions and p oposi ions o he philosophe s esul om
he ac ha we do no unde s and he logic o ou language.
Ludwig Wi gens ein
Fi s o all, many hanks o my supe iso s, Manolo and Kepa, o suppo ing me along
he way, o p o iding aluable eedback and counsel, o eaching me all I needed o know
o my PhD, and o gi ing me all he esou ces o ha e a s ong p o ile in he u u e.
I ha e el e y lucky o some yea s now. I was lucky o be gi en he chance o do my
PhD a he BCBL, and o lea n om inc edible p o essionals. I was also lucky o be su ounded
by amazing people ha g ew o be like a second amily. I ha e had hei p o essional and
pe sonal suppo all hese ou yea s, no ma e wha he p oblem was, o whe he hey knew
how o sol e i o no . And I eel lucky, abo e all hings, o ha e enjoyed as I ha e du ing he
pa h, bo h de eloping he wo ks o his hesis, bu also in li ing e e yday in such a un,
cons uc i e and ich en i onmen .
I emembe my i s day a he BCBL e y well. I did no join in easy imes, as I s a ed
igh in he middle o he COVID-19 pandemic. I emembe Chia a coming o my desk (wi h
he mask on, o cou se) o ell me ha he ew o us ha we e wo king om he cen e would
go o ha e lunch a 1pm. I emembe slowly mee ing he o he p edocs. Ch is o o os’ puns,
he loud Vicen e, and my un o una e commen s abou Valladolid o Lau a. I emembe going
on hose un wo kou s wi h Gio gio, Pie ma eo and Jose, and also o mee a Gio gio’s wi h
all o hem and I ene, Inés, Albe o, Asie … o laugh and ha e as much un as we could un il
he cu ew pushed us back o ou homes. And he “ amily” co ees a Bizi, o cou se, including
he ich (and some imes biza e) con e sa ions wi h Pa xi. So hank you, o all hese
momen s, o all o you, Chia a, Albe o, Jose, Pie ma eo, Gio gio, Vicen e, Lau a Fe nández,
Ch is o o os, Inés, I ene, Hana, Pa xi, Maca, Edi h, Ma a, Lau a de F u os, Coco, Dani López,
Asie , Da id Ca cedo, Lucía, Eneko, Giulia, Ma co, Ihin za… and he es o he BCBL
communi y I had he luck o sha e a co ee wi h. I ca y hese momen s wi h me.
Deep hanks o my housema es, Hana and (la e ) Tomas, o being like sis e and
b o he du ing he las yea s, and o making i possible o li e as i a home. And special
hanks o Tomas, also o always lis ening o my consul s and gi ing he g ea es ad ice when
I needed i .
Mos special hanks o Ma ía. Simply hank you wi h all my hea . You ha e been my
pilla , and he deepes well o lo e, un and inspi a ion. Te quie o con co du a.
II
My deepes g a i ude o my pa en s, my b o he s and sis e . Wi hou hem, his hesis
would no ha e been possible. You ga e me he li le you had, and always belie ed in my
abili y.
And inally, I would like o make wha , in my humble opinion, is a necessa y concep ual
exe cise. I ha e pu my e e y hing in o scien i ic esea ch. Bu no wi h passion, o de o ion.
These hea enly wo ds pe ain o he wo lds o eligion and belie . I ha e wo ked wi h
dedica ion and p o essionalism. As all p o essionals a he BCBL do. The wo ds we use
ma e . Le s y o unde s and he logic o ou language.
CONTENTS
ACKNOWLEDGEMENTS ...................................................................................................... I
LIST OF ABBREVIATIONS ................................................................................................. VI
ABSTRACT .......................................................................................................................... 1
RESUMEN EN CASTELLANO ............................................................................................. 2
GENERAL INTRODUCTION ................................................................................................ 5
Backg ound and Mo i a ion ....................................................................................... 5
Objec i es and Thesis S uc u e ................................................................................ 7
CHAPTER I. INTEGRATION OF LINGUISTIC PERCEPTUAL INFORMATION ................... 9
1.1. LANGUAGE VISUAL PERCEPTION ........................................................................ 9
1.1.1. Visual Pa hway ............................................................................................... 9
1.1.2. Beyond V1: Encoding o Complex Visual Fea u es ........................................ 13
1.1.3. Ven al Occipi o empo al Co ex and he Pu a i e Visual Wo d Fo m A ea .... 16
1.2. INTEGRATION OF AUDITORY LANGUAGE PROCESSING ................................. 19
1.3. LANGUAGE PROCESSING NETWORKS ............................................................. 23
1.3.1. Do sal Rou e.................................................................................................. 27
1.3.2. Ven al Rou e ................................................................................................. 27
CHAPTER II. THE NEUROBIOLOGY OF SEMANTIC REPRESENTATIONS ................... 29
2.1. MODELS FOCUSED ON THE PROCESS ............................................................. 29
2.1.1. Memo y, Uni ica ion and Con ol (Hagoo , 2005, 2013) ................................ 29
2.1.2. The Cogni i e Con ol o Seman ic Memo y (Bad e & Wagne , 2007) ........... 32
2.2. MODELS FOR SEMANTIC REPRESENTATIONS ................................................. 34
2.2.1. Dis ibu ed e sus Dis ibu ed-Plus-Hub Pe spec i e (Pa e son e al, 2007) . 35
2.2.2. Embodied Abs ac ion (Binde e al, 2009; Binde & Desai 2011) .................. 37
2.2.3. Con olled Seman ic Cogni ion (CSC) (Ralph e al., 2017) ............................. 39
CHAPTER III. THE STUDY OF NEURAL LEXICAL REPRESENTATIONS ....................... 42
3.1. Psycholinguis ic P ope ies as a Window o Lexical Rep esen a ions ..................... 43
3.1.1. Wo d Conc e eness and Imageabili y ............................................................ 43
3.1.2. Wo d F equency and Familia i y .................................................................... 45
3.1.3. Phonological P ope ies ................................................................................. 48
3.2. No el Measu es: Na u alis ic Language P ocessing and Wo d Vec o s .................. 49
3.3. No el MRI App oaches: Rep esen a ional Simila i y Analysis (RSA) ..................... 51
CHAPTER IV. THE ROLE OF READING DEMANDS AND WORD FREQUENCY IN THE
ACCESS TO LEXICAL UNITS ........................................................................................... 55
4.1. RATIONALE ........................................................................................................... 55
4.2. METHODS ............................................................................................................. 57
4.2.1. Pa icipan s .................................................................................................... 57
4.2.2 Ma e ials and P ocedu e ................................................................................. 57
4.2.3. MRI Da a Acquisi ion .................................................................................... 58
4.2.4. MRI Da a Analyses ....................................................................................... 58
4.2.5. Func ional Connec i i y Analyses .................................................................. 60
4.3. RESULTS ............................................................................................................... 60
4.3.1. Beha iou al Pe o mance .............................................................................. 60
4.3.2. Whole-b ain esul s ........................................................................................ 61
4.3.3. ROI analysis .................................................................................................. 62
Ven al Ne wo k ................................................................................................. 63
Do sal Ne wo k .................................................................................................. 64
4.3.4. Func ional Connec i i y Analysis .................................................................... 65
4.4. DISCUSSION ......................................................................................................... 66
4.4.1. Wo d equency ............................................................................................. 67
4.4.2. Reading demands ......................................................................................... 67
4.4.3. The WFE is modula ed by eading demands in an e io IFG .......................... 68
4.4.4. Task- ela ed unc ional connec i i y ............................................................... 69
4.4.5. Limi a ions ..................................................................................................... 69
4.5. CONCLUSIONS ..................................................................................................... 70
CHAPTER V. NEURAL REPRESENTATIONS OF LEXICO-SEMANTIC KNOWLEDGE:
SIMILARITY OF SUB-LEXICAL AND LEXICAL MODELS WITH MULTIVARIATE BRAIN
RESPONSES ..................................................................................................................... 71
5.1. RATIONALE ........................................................................................................... 71
5.2. METHODS ............................................................................................................. 72
5.2.1. Pa icipan s .................................................................................................... 72
5.2.2. S imuli and Ma e ials ..................................................................................... 72
5.2.3. P ocedu e ...................................................................................................... 74
5.2.4. MRI Da a Acquisi ion and P ep ocessing ....................................................... 75
5.2.5. Uni a ia e analyses ....................................................................................... 76
5.2.6. RSA sea chligh ............................................................................................. 77
5.2.7. ROI-based RSA ............................................................................................. 79
5.3. RESULTS ............................................................................................................... 79
5.3.1. Beha iou al Pe o mance .............................................................................. 79
5.3.2. Uni a ia e esul s ........................................................................................... 81
5.3.3. RSA esul s ................................................................................................... 83
RSA Sea chligh Resul s.................................................................................... 83
ROI-based RSA Resul s .......................................................................................... 85
5.4. DISCUSSION ......................................................................................................... 88
5.4.1. Associa ions Be ween Wo d P ope ies ......................................................... 88
5.4.2. An e io - o-Pos e io dissocia ion in he le IFG ............................................. 89
5.4.3. In ol emen o he le an e io OTC in lexico-seman ic p ocessing ............ 90
5.4.4. Linguis ic P ope ies s Wo d Vec o s in Seman ic Hubs ............................... 91
5.5. CONCLUSIONS........................................................................................................... 92
CHAPTER VI. BEHAVIOURAL CORRELATES OF LEXICO-SEMANTIC
REPRESENTATIONS ......................................................................................................... 94
6.1. RATIONALE ........................................................................................................... 94
6.2. METHODS ............................................................................................................. 96
6.2.1. Pa icipan s .................................................................................................... 96
6.2.2. Tasks and Ma e ials ....................................................................................... 97
6.2.3. Task Pe o mance Compa ison ..................................................................... 98
6.2.4. D i -Di usion Models and Analyses .............................................................. 99
6.2.5. Associa ion wi h B ain Rep esen a ions ....................................................... 100
6.3. RESULTS ............................................................................................................. 101
6.3.1. Pe o mance in Bo h Tasks .......................................................................... 101
6.3.2. D i -Di usion Resul s .................................................................................. 103
6.3.3. Co ela ions be ween DDMs and B ain Rep esen a ions ............................. 105
6.4. DISCUSSION ....................................................................................................... 107
6.4.1. Psycholinguis ic P ope ies and Decision-Making ........................................ 107
6.4.2. Associa ion be ween D i Ra e and B ain Rep esen a ions ......................... 108
6.5. CONCLUSIONS ................................................................................................... 111
GENERAL DISCUSSION ................................................................................................. 112
Func ional Dissocia ions in he IFG and OTC and hei Dynamic Na u e ............. 112
Seman ic Rep esen a ions in Seman ic Hubs: Psycholinguis ics and Na u al
Language P ocessing (NLP) Combined ................................................................. 114
Limi a ions and Fu u e Di ec ions .......................................................................... 115
Conclusions ........................................................................................................... 116
REFERENCES ................................................................................................................. 118
VI
LIST OF ABBREVIATIONS
ACC: an e io cingula e co ex
AF: a cua e asciculus
AG: angula gy us
AIC: Akaike in o ma ion c i e ion
ATL: an e io empo al lobe
BA: B odmann a ea
BF: Bayes Fac o
CSC: con olled seman ic cogni ion model
DCT: dual-coding heo y
DDM: d i -di usion model
dlPFC: do sola e al p e on al co ex
dmPFC: do somedial p e on al co ex
DRC: Dual- ou e cascaded model
EEG: elec oencephalog aphy
FFG: usi o m gy us
FDR: alse disco e y a e
MRI: unc ional magne ic esonance imaging
FWE: amily-wise e o
FWHM: ull wid h a hal -maximum
GLM: gene al linea model
HRF: hemodynamic esponse unc ion
IC: in e io colliculus
IFG: in e io on al gy us
IFOF: in e io on o-occipi al asciculus
ILF: in e io longi udinal asciculus
IOG: in e io occipi al gy us
IPL: in e io pa ie al lobule
IPS: in apa ie al sulcus
ITG: in e io empo al gy us
ITI: in e - ial in e al
LGN: la e al genicula e nucleus
LMM: linea mixed model
LO/hOclp4: la e al occipi al co ex
LRT: Likelihood Ra io es
M: magnocellula
6 OF 150
Mo e speci ically, I chose his agmen because i ouches some o he cen al aspec s
and ideas ha a e he objec o s udy o he cu en hesis. This wo k is mainly conce ned wi h
he cogni i e and neu al sys ems ha allow us o o m and sus ain concep s ( he chunks o
in o ma ion abou he wo ld a ound us), and o access hem du ing eading. I pe ains o he
whole con inuum o mechanisms ha make i possible o make sense o he abs ac symbols
and sounds ha cons i u e language, o ul ima ely o m meaning ul ideas and connec hem o
ou en i onmen . As illus a ed in he F ankens ein’s agmen , such concep s a e
encapsula ed in lexical uni s, o wo ds. As concep s e lec en i ies in ou wo ld ha a y in a
wide ange o ea u es, so do wo ds, as linguis ic e e ences o such concep s. F ankens ein’s
c ea u e i s acqui es hose wo ds ha ha e a clea connec ion o he pe cep ual wo ld
immedia ely a ound him, like he milk he co age s d ink, he b ead hey ea , o he wood hey
use o make a i e. O he wo ds a e s ill puzzling o him, since he canno pe cei e he en i y
hey e e o. One can ha dly see goodness o unhappiness, i no linked o e y speci ic and,
o a g ea ex en , aci indica o s o such emo ional s a es. I is h ough epe i ion, he i e a ed
connec ion be ween he pe cep ual wo ld and he abs ac e e ences o i , ha F ankens ein’s
c ea u e can a leas pa ially g asp such igu a i e ideas.
This highligh s a c i ical poin o he cu en hesis: he impo ance o he connec ion
be ween he pe cep ual inpu and he abs ac ions ha can be c ea ed om i . While he
obse a ion o he pe cep ual wo ld a ound us allows humans o ela i ely easily and na u ally
acqui e language, eading equi es yea s o ac i e ins uc ion, and e en a e ha , many
s uggle o consolida e such a complica ed abili y (S ein, 2022; Yea man & Whi e, 2021).
Language acquisi ion en ails he link o i s sounds o he en i ies in ou en i onmen , while
eading equi es o link abs ac symbols o disc e e sounds, o whole-wo d sounds, and hen
o link hose sounds o meaning. The e a e many possible pieces ha can b eak in his chain
o complex p ocesses. The e o e, i is no su p ising ha he cogni i e and b ain mechanisms
o language and eading ha e ecei ed a g ea deal o a en ion bo h in he neu obiology o
language in gene al, and in he neu obiology o eading and he associa ed eading di icul ies.
Reade s mus “deal” wi h wo “pe cep ual wo lds”, ou na u al en i onmen , o med by
he objec s and pe cep ual en i ies a ound us, and he abs ac wo ld o language, o med by
i s isual symbols and audi o y componen s. In e es ingly, he neu al mechanisms ha ac as
an in e ace be ween hese wo wo lds a e s ill poo ly unde s ood. How does ou b ain
ansi ion om he linguis ic pe cep ual wo ld o he concep ual wo ld o ideas and e e ences
o ou en i onmen ? Wha a e he neu al unde pinnings ha sus ain he o ma ion o
abs ac ions and seman ic knowledge associa ed wi h wo ds and concep s? These a e
ques ions ha a e s ill wide open.
7 OF 150
Objec i es and Thesis S uc u e
The gene al objec i e o he cu en hesis is o o e a comp ehensi e pe spec i e o
he neu al mechanisms ha suppo he access and use o concep ual in o ma ion du ing
eading. On he one hand, his encompassess hose p ocesses ela ed o deciphe ing wo ds
e en be o e hey a e p ocessed as a whole. This, in u n, includes he segmen a ion o he
basic uni s ha cons i u e a wo d (syllables), and linking hem o hei sound basic uni s
(phonemes). This is wha I will e e o when I use he exp ession sub-lexical p ocessing.
These sublexical p ocesses a e mos ly (al hough no exclusi ely) de eloped om b ain a eas
a he bo om o he hie a chy, in pe cep ual egions, and hence a e mos ly co e ed by wha
a e e med bo om-up mechanisms. On he o he hand, I will co e hose p ocesses ela ed o
he pe cep ion o wo ds as a whole, and will o en e e o hese as lexical uni s (as wo ds a e
he en ies o ou lexicon). And going a s ep u he , I will o en allude o lexico-seman ic
p ocesses o e e o he cogni i e mechanisms ha allow o he connec ion be ween he
whole-wo d and he concep i ep esen s (i.e., i s meaning). P e ious knowledge and
expec a ions in ol e connec ions om b ain a eas a he op o he hie a chy (o high le el
a eas) o pe cep ual and associa ion egions, and hence hey a e o en e e ed o as op-
down mechanisms. Al hough his hesis is mos ly conce ned wi h he second g oup o
p ocesses, hey canno be unde s ood wi hou aking sublexical p ocessing in o accoun .
The speci ic objec i es o he hesis a e: 1) o in es iga e he in luence o p e ious
knowledge and expec a ions (he e e e ed o as op-down in luences) in how wo ds a e
accessed a he neu al le el du ing eading; 2) o explo e he in luence o wo d ea u es ha
e lec c i ical psycholinguis ic p ope ies on he neu al ep esen a ions o concep ual
knowledge; and 3) o assess di e en models ha e lec a a ie y o lexical p ope ies, om
sublexical o lexico-seman ic, and analyse hei ep esen a ion a he neu al le el.
The hesis will con ain six chap e s. The i s h ee will be heo e ical chap e s ha aim
o espond o he gene al objec i e, by o e ing a c i ical e iew o he a ailable knowledge on
lexical ep esen a ions and he neu al mechanisms ha suppo hem. This will go om how
he b ain deciphe s isual linguis ic inpu , o how i ep esen s complex abs ac in o ma ion,
and how we can explo e his h ough he use o unc ional magne ic esonance imaging ( MRI).
Chap e 1 will co e he neu al unde pinnings o he isual in eg a ion o linguis ic in o ma ion.
This will include how isual inpu s a e decoded by he b ain, and how he b ain makes i
possible o ul ima ely ecognise wo ds. A he end o he chap e , I will co e how hese neu al
p ocesses a e in eg a ed in o la ge b ain language ne wo ks. Chap e 2 will be dedica ed o
e iew he mos in luen ial models ha ied o explain how seman ic in o ma ion is
ep esen ed in he b ain, and wha a e he neu al mechanisms ha suppo he access and
manipula ion o his in o ma ion. And Chap e 3 will ackle he me hods we ha e o s udy neu al
8 OF 150
seman ic ep esen a ions. I will almos exclusi ely e e o MRI in es iga ions, and he
manipula ions ha can be made o explo e lexico-seman ic p ocessing a he neu al le el.
The las h ee chap e s a e empi ical, and y o add ess he speci ic objec i es o he
hesis, while also con ibu ing o he gene al amewo k. In Chap e 4, I p esen a MRI s udy
in which I explo ed he in luence o op-down eading demand (pe cep ual demand e sus
seman ic demand) and wo d equency as c i ical a iables ha a ec he neu al esponses o
wo ds as hey a e ead. Chap e 5 desc ibes an in es iga ion ha used no el MRI analy ical
app oaches o explo e neu al ep esen a ions associa ed wi h sublexical o lexico-seman ic
p ocessing. And inally, Chap e 6 will y o complemen he s udy desc ibed in Chap e 5 by
c i ically analysing he beha iou al co ela es associa ed wi h he neu al ep esen a ions
p esen ed in he p e ious chap e .
9 OF 150
CHAPTER I. INTEGRATION OF LINGUISTIC PERCEPTUAL INFORMATION
1.1. LANGUAGE VISUAL PERCEPTION
1.1.1. Visual Pa hway
F om a s ic neu ophysiological poin o iew, he e y i s s ep o he human b ain
o deciphe isual linguis ic in o ma ion is o cap u e he physical s imuli ha cons i u e he
linguis ic inpu (S ein, 2022). These physical s imuli a e hen ans o med in o elec ical
impulses ha ou b ain in e p e s in o de o o m he pe cep ual expe ience ha allows
eading. This i s -le el s ep is possible hanks o he e ina and he isual pa hway (Kiley &
Us ey, 2016; S ein, 2022).
The e ina con ains pho o ecep o s ha cap u e ligh pho ons, and elay his s imula ion
o he e inal ganglion cells in he o m o elec ochemical signals (Sa ucci & Po cia i, 2024)
ha a e ca ied by he op ic ne e (Celesia, 2005). A e pa ially c ossing i s ib es o he
con ala e al pa a he op ic chiasm, he op ic ne e o ms he op ic ac . Some o i s ib es
each he supe io colliculi o con ol ocula e lexes, and he pul ina nucleus o he halamus,
whe e minimal isual p ocessing akes place (Celesia, 2005; Kahle e al., 2022). The majo i y
o he op ic ac ib es, howe e , un la e ally, con eying he low-le el isual in o ma ion o
he la e al genicula e nucleus (LGN) in he halamus. F om he e, he ib es cons i u e he op ic
adia ion, which eaches he s ia e co ex in he p ima y isual co ex, also called V1 a ea
(Celesia, 2005; Kiley & Us ey, 2016), as well as he seconda y isual co ex o V2 o a much
lesse ex en (Al a ez e al., 2015; A igo e al., 2016)
V1 ollows a e ino opic o ganisa ion, whe e each quad an o he con ala e al isual
hemi ield is ep esen ed in a speci ic loca ion o he s ia e co ex. This o ganisa ion s a s in
he e ina and is kep along he isual pa hway, h ough he op ic ne e, ac and adia ion
(Kahle e al., 2022). In his e ino opic o ganisa ion, he o ea, a e inal a ea wi h he highes
isual sha pness (gi en o i s issue composi ion), is o e ep esen ed in V1 as compa ed o
he pe iphe y o he isual ield (Celesia, 2005). Fu he mo e, he neighbou ing cells in V1
ep esen adjacen isual ecep i e ields o one ano he (Kiley & Us ey, 2016). Figu e 1.1
shows he a angemen o op ic ib es, he o ganisa ion o he isual quad an s along he isual
pa hway and hei ep esen a ion in V1.
Figu e 1.1. A) A angemen o he op ic ib es; B) Posi ion o e inal quad an s along he isual
10 OF 150
pa hway, including V1 (bo om igh ). Taken om Kahle e al (2022).
The main inpu o V1 comes om he LGN, which is o med by six laye s o simple
cells ha al eady p ocess some basic isual ea u es. The wo p edominan ypes o cells in
he LGN a e he magnocellula (M) and pa ocellula (P) cells (Celesia, 2005). While P cells
espond o low empo al equency and high spa ial equency, and ha e hinne axons and
smalle ecep i e ields, M cells espond o high empo al equency and low spa ial equency,
and ha e la ge ecep i e ields and hicke axons. This is why P cells a e sensi i e o
ch oma ic in o ma ion and isual acui y, whe eas M cells a e sensi i e o as mo ion
in o ma ion (Celesia, 2005; Kiley & Us ey, 2016). The ecep i e ields o all LGN cells a e
o ganised in wha is called a cen e-su ound (o ON and OFF) s uc u e ha consis s in a
cen e ha esponds o inc emen o dec emen o ligh , and an annula po ion ha su ounds
i , which esponds in he opposi e way o he cen al po ion (Celesia, 2005; Kiley & Us ey,
2016). The V1 a ea is o med by simple cells and complex cells. Simple cells in V1 also ha e
On and O ecep i e ields, bu a e elonga ed and cap u e o ien a ion o isual s imuli, wi h a
p e e ence o a speci ic o ien a ion (Celesia, 2005; Kiley & Us ey, 2016). Simple cells in V1
ecei e hei inpu om mul iple LGN simple cells. Complex cells in V1 ecei e hei inpu om
mul iple simple cells ha sha e he same o ien a ion p e e ence, bu di e in hei spa ial
a angemen (Kiley & Us ey, 2016). Figu e 1.2 ep esen s he s uc u e o LGN and V1 simple
cells, and he connec ions be ween hem.
11 OF 150
Figu e 1.2. Rep esen a ion o LGN and V1 simple cells. The igu e also ep esen s he eed o wa d
model in which V1 simple cells a e ed by he inpu om mul iple LGN cells. Taken om Kiley &
Us e y (2016).
Ho izon ally, he V1 a ea is o ganised in six main laye s. The laye 4 con ains he
majo i y o he simple cells. M and P cells e mina e in di e en sublaye s o laye 4. Laye s 5
and 6 send eedback o he LGN and o he subco ical egions. The laye 4 p ojec s i s ib es
o laye s 2 and 3, which send hei p ojec ions o ex as ia e a eas. Ve ically, he V1 a ea
ollows a columna o ganisa ion, in which he complex cells a e a anged in bundles ha
con ain he o ien a ion in o ma ion om one o he espec i e hemi ields (le o igh ), which
is why hey a e also called ocula dominance columns (Kahle e al., 2022; Kiley & Us ey,
2016). The combina ions o wo adjacen columns om he ipsila e al and con ala e al eyes
a e called hype columns, and hey ep esen he ecep i e ield o he co esponding eye
(Kahle e al., 2022). This esul s in wo inpu s ha ep esen he same in o ma ion in wo
di e en ways (le columns and igh columns). This phenomenon is called binocula dispa i y,
and is c i ical o dep h pe cep ion (Kiley & Us ey, 2016). Be ween he ocula dominance
columns lay he blobs, columns ha a e no sensi i e o o ien a ion, bu ha ha e high
cy och ome oxidase con en , and a e he e o e sensi i e o colou (Kahle e al., 2022). Blobs
a e loca ed in laye 2 o V1, and ecei e hei inpu di ec ly om LGN cells. Figu e 1.3 shows
he ho izon al laye s and e ical columna o ganisa ion o V1.
12 OF 150
Figu e 3. A) Ho izon al o ganisa ion o V1 in o 6 di e en laye s. B) Ve ical o ganisa ion o
V1 in o ocula dominance columns. 1-6. Ho izon al laye s; 7. ; 8. ; 9. A hype column; 10.
Blobs. Taken om Kahle e al. (2022).
Thanks o he p ope ies o he isual pa hway desc ibed hus a , he p ima y isual
co ex and i s inpu s allow o encode basic isual ea u es, such as o ien a ion, ocula
dominance, edges and shape, and s imulus size (Kiley & Us ey, 2016). I is in V1 whe e he
pe cep ion o all hese ea u es akes place (Celesia, 2005), and only minimal p ocessing o
speci ic ea u es o he isual s imuli occu ea lie in he pa hway (Celesia, 2005; Kiley & Us ey,
2016). Howe e , mo e complex cha ac e is ics o he isual s imuli a e p ocessed la e in he
isual pa hway, in ex as ia e a eas (Kiley & Us ey, 2016).
13 OF 150
1.1.2. Beyond V1: Encoding o Complex Visual Fea u es
So a , we ha e co e ed how he human isual pa hway, up o he p ima y isual
co ex, con e s physical s imuli in o elec ical impulses ha cons i u e he basis o isual
pe cep ion o low-le el ea u es. Howe e , we ha e no answe ed he ques ion o how isual
in o ma ion is ca ego ised and ecognised by he human b ain, a c i ical s ep in he isual
in eg a ion o linguis ic inpu and, he e o e, in eading. Al hough he p ima y isual co ex is
necessa y o he ecogni ion o isual inpu , his ask is ac ually ca ied ou by a se o a eas
ha eside ou side V1, especially in he en al occipi o empo al co ex ( OTC) (G ill-Spec o
& Weine , 2014; Weine e al., 2014). The s ia e co ex is connec ed en ally o he OTC,
bu be o e eaching he OTC, V1 sends pa allel p ojec ions o a eas V2-V5, also known as
he seconda y isual co ex. These a eas, as V1, a e e ino opically o ganised, and a e
especially ele an in p ocessing addi ional isual ea u es like angle, mo ion o ex u e (Fu lan
& Smi h, 2016; G ill-Spec o & Weine , 2014; Okazawa e al., 2016; Zhong & Wang, 2021).
A eas V2-V5 p ocess in pa allel se e al di e en ea u es, while exhibi ing each o hem a
ce ain p e e ence o a conc e e isual cha ac e is ic. Fo ins ance, i has been shown ha
V2 encodes con ou (edges and co ne de ec ion) o isual s imuli (R. Chen e al., 2017; Roe
& Ts’o, 2015; Zhong & Wang, 2021), while being necessa y o isual awa eness (Salminen-
Vapa an a e al., 2012). On he o he hand, V3 and V5/MT a e especially ele an o mo ion
p ocessing (Fu lan & Smi h, 2016; Jeschke e al., 2023; Sil an o e al., 2005), al hough a ea
V2 also con ibu es o ea ly mo ion encoding (Fu lan & Smi h, 2016). Finally, a ea V4 is o en
ecognised as he main seconda y isual egion o colou pe cep ion (Bou ie & Engel, 2006;
B ouwe & Heege , 2009; Desimone e al., 1985; Pasupa hy e al., 2020), al hough colou
encoding is no limi ed o V4 (Bou ie & Engel, 2006; B ouwe & Heege , 2009). Fu he mo e,
V4, along wi h V2, also encodes o he isual ea u es like ex u e (Okazawa e al., 2016) o
size (Tanaka & Fuji a, 2015), which is why V4 is belie ed o pa icipa e in ea ly isual objec
ecogni ion (Pasupa hy e al., 2020).
The ca ego isa ion and ecogni ion o isual objec s (which includes linguis ic uni s
such as wo ds), howe e , is possible hanks o he unc ion o he OTC. I has long been
epo ed ha lesions in he OTC in p ima es lead o di icul ies in disc imina ing isual objec s,
wi hou impai ing he abili y o pe cei e hose objec s’ spa ial con igu a ion (Pohl, 1973). These
indings led o he accep ed no ion ha he isual pa hway is di ided in o wo main s eams
(Goodale & Milne , 1992; Mishkin e al., 1983): a do sal s eam ha uns p edominan ly
h ough he occipi opa ie al co ex, and a en al s eam ha uns h ough he occipi o empo al
co ex. The do sal s eam is also called he whe e o how pa hway, since i suppo s spa ial
and mo o isual pe cep ion. The en al s eam allows he ecogni ion o objec iden i y by
es ablishing a b idge be ween isual inpu and in o ma ion s o ed in memo y (see Figu e 1.4),
14 OF 150
and hence has been named he wha pa hway (C. B. Ma in & Ba ense, 2023). I is impo an
o no e ha hese wo s eams a e no absolu ely independen , and ha hey dynamically
in e play wi h one ano he .
Figu e 1.4. Rep esen a ion o he en al isual pa hway and i s unc ion in he p ima e
b ain. Taken om Ma in & Ba ense (2023).
Al hough bo h isual s eams a e necessa y o language p ocessing, he en al
s eam is pi o al o eading, as i is p incipally esponsible o wo d ecogni ion and he
in eg a ion o isual and language p ocessing (Yea man & Whi e, 2021). Mo e speci ically, i
is he OTC whe e isual ea u es a e in eg a ed o ul ima ely cons uc he pe cep ion o isual
objec s (G ill-Spec o & Weine , 2014). The OTC displays a high le el o cy oa chi ec onic
and unc ional specialisa ion, and hence i s sub egions selec i ely ep esen di e en objec
ca ego ies (Downing e al., 2006; G ill-Spec o & Weine , 2014; Weine e al., 2014). Fo
ins ance, some ew ca ego ies ha a e conside ed “ecologically- alid” (i.e., hey ha e been
ele an o he e olu ion o he human species) a e commonly clus e ed oge he in OTC
sub egions: aces a e clus e ed a ound he mid- and pos e io usi o m gy us (FFG), and he
in e io occipi al gy us (IOG) (G ill-Spec o e al., 2017); places usually clus e a ound he
pa ahippocampal place a ea (PPA), in he medial po ion o he OTC (Eps ein & Bake , 2019;
Nas e al., 2011); body pa s end o be ep esen ed in occipi o empo al sulcus (OTS)
(Downing e al., 2006; G ill-Spec o & Weine , 2014; Peelen & Downing, 2005; Ri chie e al.,
15 OF 150
2021); objec s ep esen a ions a e usually also clus e ed a ound he OTS, pa ially
o e lapping wi h body clus e s, and ex ending pos e io ly (G ill-Spec o & Weine , 2014;
Ri chie e al., 2021); and wo ds and symbols a e commonly ep esen ed in he OTS, pa ially
o e lapping wi h objec and body clus e s, bu ex ending en ally (G ill-Spec o & Weine ,
2014), wi hin he so-called isual wo d o m a ea (VWFA) (Dehaene e al., 2002). Figu e 1.5
illus a es he unc ional specialisa ion o he OTC.
Figu e 1.5. Func ional clus e s o OTC sub egions and he ca ego ies ha a e ep esen ed wi hin
hem. 1) in e io occipi al gy us (IOG); 2) pos e io usi o m ace-selec i e egion (pFus), o ace
usi o m a ea-1 (FFA-1); 3); mid- usi o m ace-selec i e egion (mFus) o FFA-2; 4) occipi o empo al
sulcus (OTS); 5) pa ahippocampal place a ea (PPA); 6) isual wo d o m a ea (VWFA). MFS: mid-
usi o m sulcus. Taken om G ill-Spec o & Weine (2014).
The bo om-line message is ha he OTC is a undamen al b ain egion o he
ecogni ion o isual inpu , and ha i shows high speci ici y as o wha ype o s imuli a e
p e e en ially ep esen ed in i s subdi isions. The unc ion o he OTC allows us o dis inguish
one kind o s imulus om he o he s, so ha we can espond acco ding o ou needs in a
pa icula si ua ion. Taken o eading, his en ails being able o ecognise isual linguis ic inpu
( he bo om-up p ocess) such as wo ds. Bu in o de o be able o ecognise wo ds as whole,
22 OF 150
2020). The in luence o audi o y p ocessing o wo ds o e he OTC is bes illus a ed by a
numbe o s udies ha demons a ed au oma ic op-down ac i a ion o he OTC du ing
speech p ocessing (Cao e al., 2010; Conan e al., 2020; Dziȩgiel-Fi e e al., 2023;
Lude sdo e e al., 2016; Pa amadilok e al., 2019; Plan on e al., 2019; Yonche a e al.,
2010; o a e iew, see Dȩbska e al., 2023). Howe e , he e is no clea consensus abou he
na u e o his op-down in luence. The p oponen s o he VWFA link he OTC esponse o
audi o y s imuli o co-ac i a ions o he o hog aphic whole-wo d code (Dehaene & Cohen,
2011). Ano he possibili y is ha OTC ac i a ion e lec s phonological p ocessing o smalle
o hog aphic uni s (e.g., phonemes) (Pa amadilok e al., 2019). This iew is suppo ed by
s udies showing an e ec o phonological ac o s, like syllable s uc u e, o e OTC ac i a ion
(Conan e al., 2020). A hi d explana ion conside s he coexis ence o phonological,
o hog aphic and seman ic clus e s o neu ons wi hin he OTC (Dȩbska e al., 2023). This is
a p omising accoun , ha aligns wi h e idence sugges ing a di ision o labou wi hin he OTC
(Le ma-Usabiaga e al., 2018; Sebas ian e al., 2014; Whi e e al., 2019; Zemmou a e al.,
2015), and ha is being inc easingly explo ed wi h mul i a ia e app oaches ha can del e in o
hese ine dis inc ions (Fische -Baum e al., 2017; X. Wang e al., 2018). We will come back
o his ma e in Chap e 3.
We canno conclude he sec ion wi hou cla i ying a ac ha has been poin ed ou
abo e: ha o he a eas ou o he seconda y audi o y co ex also display igh connec i i y wi h
OTC egions (Le ma-Usabiaga e al., 2018). In ac , he unc ional coupling wi h he OTC
changes as a unc ion o li e acy, some hing ha goes in line wi h he in e ac i e accoun o
he OTC by P ice & De lin (2011). In e es ingly, he connec i i y be ween he STG and he
OTC dec eases wi h inc easing li e acy, while connec i i y wi h pa ie al a eas and he on o-
pa ie al ne wo k inc eases wi h inc easing li e acy (López-Ba oso e al., 2020; Moul on e al.,
2019). This eminds us ha language p ocessing, and speci ically, eading, equi es
coo dina ion be ween se e al di e en cogni i e p ocesses, and hus, connec i i y be ween
se e al di e en b ain a eas. The pa ie al co ex pa icipa es in he mapping o phonological
senso y ep esen a ions o language wi h i s mo o a icula ion, and i is he e o e expec ed o
inc ease i s pa icipa ion wi h inc easing li e acy (Hickok & Poeppel, 2007). Aspec s like his
illus a e how impo an i is o unde s and he b ain ne wo ks in ol ed in language p ocessing,
and no jus he egional ac i a ion associa ed wi h i . This is why he nex sec ion will be
dedica ed o language p ocessing ne wo ks.
1.3. LANGUAGE PROCESSING NETWORKS
Be o e and du ing he ea ly 2000s, compu a ional models coming om
neu opsychology and linguis ics we e a popula me hod o a emp ing o explain how di e en
23 OF 150
cogni i e p ocesses in e ac o p oduce a ce ain beha iou , especially in he con ex o b ain
damage (P o opapas e al., 2016). In sho , hese models a e composed o inpu uni s ( he
pe cep ual co ela es o any expe ience), hidden uni s ( he cogni i e p ocesses ha ans o m
he pe cep ual inpu ) and ou pu uni s ( he beha iou al p oduc ion esul ing om he in e ac ion
be ween he o me wo), componen s ha in e ac wi h each o he in ma hema ically
exp essed ela ions (Col hea e al., 2001). Taken o eading, se e al di e en compu a ional
models ied o ma hema ically exp ess how he di e en sub-lexical and lexical componen s
in e ac o p oduce isual wo d ecogni ion and eading aloud. The e a e mul iple examples o
compu a ional models o eading (Col hea e al., 2001; G ainge & Jacobs, 1996; McClelland
& Rumelha , 1981; Plau e al., 1996), bu , a om ocusing on hese models, I will b ie ly
desc ibe one o hem ha has ecei ed a g ea deal o a en ion, and ha se ed as he basis
o one o he mos popula amewo ks o he neu obiology o eading: he Dual- ou e
cascaded model (DRC) (Col hea e al., 2001). Fo cla i y, I p esen he e, in Figu e 1.7, a
g aphical ep esen a ion o he DRC model.
24 OF 150
Figu e 1.7. A g aphical ep esen a ion o he DRC model, aken om Col hea e al. (2001).
As he name indica es, he DRC model unde s ands ha he e a e wo main s eams
o p ocessing p in ed wo ds: 1) a g apheme-phoneme co espondence ou e in cha ge o
con e ing le e s in o phonemes (o a phonological ou e); and 2) a lexical ou e ha
p ocesses wo ds as a whole, bo h o hog aphically (i.e., he phonological ules ha cons i u e
he wo d en y in he lexicon) and seman ically (i.e., he meaning o ha wo d en y in he
lexicon). Impo an ly, Col hea e al. concei ed ha hese componen s in e ac in a cascaded
ashion. This is, ins ead o being ac i a ed sequen ially a e eaching a gi en h eshold, once
one p ocess (e.g., inpu le e uni s) s a s, i can send and ecei e exci a o y o inhibi o y
connec ions o/ om he ones connec ed o i .
These ea u es made he DRC model an a ac i e op ion o be es ed wi h
neu oimaging ools in o de o in es iga e he p ocessing ou es o language. In ac , he DRC,
and o he simila dual- ou e models, can be conside ed he compu a ional s a ing poin o one
o he mos popula amewo ks in he neu obiology o language: he di ision o labou
25 OF 150
be ween he do sal and en al ou es (Joba d e al., 2003). This concep ion was based on
he adi ional idea ha senso y p ocessing o language equi es he in e ace be ween 1) a
mo o -a icula o y sys em ha allows o he in eg a ion o audi o y inpu in o he mo o
a icula ion equi ed o ul ima ely p oduce speech; and 2) a concep ual sys em ha allows us
o comp ehend language (Hickok & Poeppel, 2004). D awing on e idence in he neu obiology
o ision, Hickok and Poeppel (2000, 2004) p oposed a simila di ision o labou o language
p ocessing: a do sal ou e, c i ical o audi o y-mo o in eg a ion o language; and a en al
ou e ha is in ol ed in mapping he language uni s on o meaning. O pu in o he wo ds, while
he do sal ou e akes ca e o phonological p ocessing, he en al ou e is mainly in cha ge o
lexico-seman ic p ocessing (Oli e e al., 2017; Sandak e al., 2004). The do sal ou e is mainly
composed o he middle and pos e io STG, he in e io pa ie al lobule (IPL), p emo o co ex
(PMC), and he pos e io pa o he in e io on al gy us (IFG), which includes he pa s
ope cula is (B odmann A ea (BA) 44); he en al ou e is p edominan ly o med by he in e io
empo al gy us (ITG), including he OTC, he an e io empo al lobe (ATL) and he an e io
pa o he IFG, which includes he pa s iangula is (BA 45) and pa s o bi alis (BA 47)
(F iede ici, 2011; Hickok & Poeppel, 2007; Oli e e al., 2017; Sandak e al., 2004; Sau e al.,
2008). I is usually assumed ha bo h ou es display a ce ain deg ee o le la e alisa ion
(F iede ici, 2011; F iede ici & Gie han, 2013). Howe e , i has been p oposed ha while he
do sal ou e is usually mo e clea ly le -la e alised, he en al ou e is mo e bila e ally
dis ibu ed (Hickok & Poeppel, 2007). In any case, bo h hemisphe es show a simila s uc u al
connec i i y pa e n be ween he abo e-men ioned a eas, al hough each o hem can se e
dis inc cogni i e unc ions du ing language p ocessing (Dick e al., 2014). Impo an ly, we
cu en ly know ha hese ou es a e no isola ed om each o he , bu hey can in e ac wi h
one ano he , and o m edundan connec ions be ween he a eas ha cons i u e hem (López-
Ba oso & De Diego-Balague , 2017)
In gene al, his iew se es us as a use ul amewo k o unde s and how eading
ec ui s se e al b ain mechanisms in unison, and hus ely on la ge-scale b ain ne wo ks.
Ne e heless, his adi ional iew can pe haps be minimally inged o ma ch he mos ecen
e idence, pa o which we al eady in oduced in p e ious sec ions. The i s nuance ha
ecen e idence can add is he knowledge ha o he la ge-scale b ain ne wo ks can con ibu e
o language p ocessing (Hagoo , 2019; Sie powska e al., 2022). Language usage ec ui s
bo h i s basic uni s (e.g. phonemes, wo ds, sen ences), and a numbe o ope a ions ha can
be made o e hese uni s (e.g., ecognising, e ie ing, o linking a wo d wi h s o ed
in o ma ion and i s con ex ) (Deniz e al., 2023; Hagoo , 2019). The consequence o his is
ha mul iple p ocesses can be in ol ed e en when p ocessing he single uni s o language
(Roge , Rod igues De Almeida, e al., 2022). In e ms o s uc u al connec i i y, i is belie ed
ha his ea u e acili a es s onge and mo e widesp ead whi e ma e ac s be ween he
26 OF 150
human empo al co ex and o he a eas ha ac as associa ion hubs, like he pa ie al o on al
co ices (B aunsdo e al., 2021; Sie powska e al., 2022). This connec s o he second
nuance o he do sal/ en al amewo k. G owing e idence om di usion weigh ed imaging
(DWI) s udies poin ed o se e al di e en ac s con ibu ing o ei he he do sal o he en al
ou e o language (Ca ani e al., 2005; F iede ici, 2009; Sau e al., 2008). This g oup o
e idence led o mos language scien is o alk abou do sal “pa hways” and en al “pa hways”
(Dick e al., 2014; F iede ici, 2011; F iede ici & Gie han, 2013). I illus a e his in Figu e 1.8,
which o e s a schema ic iew o he mul iple do sal and en al language pa hways.
Figu e 1.8. A schema ic iew o he do sal and en al pa hways. FC: F on al Co ex; OC: Occipi al
Co ex; TC: Tempo al Co ex. Taken om F iede ici & Gie han (2013).
And he las nuance, pa ly illus a ed in some o he de ails o egional ac i a ion
p esen ed hus a , conce ns he speci ics o he unc ional p ope ies o he egions
composing bo h pa hways. The e is conside able ag eemen on he idea o mul i unc ionali y
o b ain egions: he same egion can se e mul iple pu poses, and se e al di e en egions
can con ibu e o he same unc ion (Bullmo e & Spo ns, 2009). This p inciple can be applied
o he language ne wo ks, and as an example, we can ake back he idea o he OTC se ing
bo h lexico-seman ic and sub-lexical p ocesses (Le ma-Usabiaga e al., 2018; Whi e e al.,
2019).
Fo all he easons exposed abo e, in he sub-sec ions below, I will desc ibe in u he
de ail he connec ions and unc ions co e ed by each o he wo main language ne wo ks, and
he egions ha cons i u e hem.
27 OF 150
1.3.1. Do sal Rou e
In sec ion 1.2., we desc ibed how he le STG suppo s audi o y pe cep ion o linguis ic
s imuli, and mo e speci ically, phonological disc imina ion o language sounds. Two main whi e
ma e ib e bundles connec he STG mainly o he do sal PMC and pa ly o he he do sal
IFG (pa s ope cula is): he supe io longi udinal asciculus (SLF) and he a cua e asciculus
(AF). Focusing on he i s do sal pa hway, he STG is connec ed indi ec ly, ia he IPL, o he
do sal PMC by he SLF (Dick e al., 2014; F iede ici & Gie han, 2013). The IPL is known o be
impo an o ac i ely e ie ing and keeping ele an in o ma ion online (Cabeza e al., 2011;
Ses ie i e al., 2017), which in he case o language p ocessing, en ails phonological wo king
memo y (F iede ici & Gie han, 2013). The do sal PMC is esponsible o he mo o a icula ion
o speech (Hickok & Poeppel, 2007). As a consequence, lesions o any po ion o he SLF, o
any o he egions ha o m his connec ion, a e known o p oduce de ici s in he epe i ion o
speech (F iede ici & Gie han, 2013; Hickok & Poeppel, 2007). A second do sal pa hway
connec s he le pos e io STG o he pa s ope cula is o he IFG h ough he AF (and pa o
he SLF) (Ca ani e al., 2005; Thiebau De Scho en e al., 2012). The lack o a clea consensus
abou he composi ion o he ib es o ming his pa hway has some imes led esea che s o
name he ac SLF/AF (Dick e al., 2014; F iede ici & Gie han, 2013). The le pa s ope cula is
(BA 44), which lies nex o he PMC, has been p oposed o ake pa in p o iding op-down
p edic ions abou he linguis ic inpu (F iede ici & Gie han, 2013). Recen ly, he pos e io pa
o BA 44, lying nex o he PMC, has been linked o ac ion obse a ion and imi a ion, while he
an e io BA 44, connec ed o BA 45, is ele an o he unde s anding o complex language
ules and sequences p ocessing (F iede ici, 2023; Hamzei e al., 2016; Papi o e al., 2020).
Fo simpli ica ion, he connec ion be ween he le STG and le pos e io IFG (also ia le
PMC) is belie ed o be in ol ed in complex syn ac ic p ocessing (F iede ici & Gie han, 2013).
We can summa ise he ole o he do sal pa hways o language by saying ha hey a e in
cha ge o p ocessing he phonological aspec s o wo ds, e ie ing and keeping in memo y
hese sound uni s, and gene a ing p edic ions abou he inpu ha will help bo h in a icula ing
speech and in ecognising w i en and spoken phonological uni s.
1.3.2. Ven al Rou e
Gi en ha he p esen wo k is mainly conce ned wi h lexico-seman ic access, i is no
da ing o say ha we will mos ly (albei no exclusi ely) ocus on he egions cons i u ing hese
en al pa hways along his hesis. Fo his eason, he e I will simply desc ibe he connec ions
o he a eas o ming he en al pa hways, as well as he o e all unc ion o he en al ou es.
In he ollowing chap e s, I will add ess he di e se unc ional specialisa ion o he di e en
egions in ol ed in hese pa hways.
28 OF 150
Two main whi e ma e ac s a e known o connec he a eas a ound he OTC, and
occipi al co ex o e all, o he ATL and he an e io pa o he IFG (BA 45 and BA 47): he
in e io longi udinal asciculus (ILF) and he in e io on o-occipi al asciculus (IFOF). A hi d
ac , he uncina e asciculus, connec s he la e al and o bi al on al co ex, BA 47 and he
an e io cingula e co ex (ACC) wi h he ATL, amygdala and pa ahippocampal co ex (Dick e
al., 2014; Thiebau De Scho en e al., 2012). Addi ionally, and ou side he ypical
do sal/ en al amewo k, he pos e io OTC has been shown o display signi ican
connec i i y wi h he en al pa ie al lobe, p esumably h ough he pos e io AF, and he
in e io longi udinal asciculus (ILF) (Le ma-Usabiaga e al., 2018; López-Ba oso e al., 2020;
Moul on e al., 2019). In e es ingly, his pa e n has been shown o be associa ed wi h
inc easing eading expe ience (López-Ba oso e al., 2020; Moul on e al., 2019). The OTC
has al eady been desc ibed as an a ea ha is c ucial o he isual ecogni ion o lexical uni s,
and ha ecei es op-down in luences om phonological a eas. The ATL is known o be
in ol ed in seman ic p ocessing du ing complex sen ence eading, and o ep esen seman ic
knowledge o e all (Dick e al., 2014; F iede ici & Gie han, 2013). Finally, he an e io IFG (BA
45 and BA 47) is associa ed wi h seman ic ca ego isa ion and delibe a e access o seman ics
(Bad e & Wagne , 2007; F iede ici & Gie han, 2013), while also pa icipa ing in syn ax
p ocessing (F iede ici, 2011; F iede ici & Gie han, 2013). We will del e in o he unc ions o
ATL and he IFG in Chap e 2.
In sum, he en al pa hways ake on he ask o ca ego ising he linguis ic inpu in o
disc e e lexical uni s, link i o p e iously s o ed knowledge, and access i in a delibe a e
manne o in e ac wi h he wo ld a ound us.
29 OF 150
CHAPTER II. THE NEUROBIOLOGY OF SEMANTIC REPRESENTATIONS
The do sal/ en al iew cons i u es a a he use ul amewo k in unde s anding b ain
ac i a ion associa ed wi h speci ic subp ocesses du ing language p ocessing. Ne e heless,
as i was o iginally ailo ed o explain he neu oana omy o speech pe cep ion (Hickok &
Poeppel, 2004), he e a e a numbe o aspec s ha he do sal/ en al iew lea es ou . When
i comes o comp ehending how he b ain c ea es and accesses seman ic ep esen a ions,
se e al o he models a e especially help ul. Fu he mo e, language is no a uni a y p ocess,
and he egions cons i u ing he ne wo ks desc ibed hus a a e no exclusi ely ec ui ed o
speech pe cep ion. Cogni i e p ocesses like memo y a e key o unde s anding how he human
b ain builds, ep esen s, and accesses concep ual knowledge. In his chap e , we will ocus
on he mos ele an models ha aim a explaining he neu obiology o seman ic
ep esen a ions, bo h h ough language and memo y. Fo cla i y, I will di ide hem in o 1)
models ha a e ocused on he p ocesses suppo ing he access and use o seman ic
knowledge, and 2) hose ocused on he s o ed seman ic ep esen a ions, as well as
in eg a i e models ha combine aspec s om bo h iews.
2.1. MODELS FOCUSED ON THE PROCESS
In his sec ion, we will e iew wo heo e ical models ha a e especially conce ned wi h
he cogni i e p ocesses ha allow he access, manipula ion and con ol o he p e iously
acqui ed linguis ic in o ma ion: he Memo y, Uni ica ion and Con ol (MUC) model, p oposed
by Hagoo (2005) and he cogni i e con ol o seman ic memo y iew, pu o wa d by Bad e
and Wagne (2007). While he scope o he MUC model, p oposed by Hagoo includes
syn ac ic, phonological and seman ic p ocesses, he model p oposed by Bad e and Wagne
is mainly conce ned wi h he access o seman ic memo y.
2.1.1. Memo y, Uni ica ion and Con ol (Hagoo , 2005, 2013)
Un il well in o he 21s cen u y, he neu obiology o language was domina ed by he
B oca-We nicke model (Hagoo , 2013). As b ie ly in oduced in he p e ious sec ion, his iew
unde s ood ha he main wo neu ocogni i e componen s o language, p oduc ion and
pe cep ion, we e mainly based on he ac i i y o he le IFG and he le pos e io STG,
espec i ely (Hickok & Poeppel, 2007). Bu like mos o he models in he ea ly o mid 2000s
ha aimed a explaining how language is p ocessed a he neu al le el, Hagoo (2005, 2013)
buil on he idea ha he e is a u he di ision o labou ou side he classical B oca-We nicke
amewo k. The s a ing poin o Hagoo ’s model is he iden i ied need o explain bo h wha
a e he disc e e cogni i e skills ha language makes use o (cogni i e a chi ec u e), and how
hese skills a e suppo ed by b ain unc ion (neu al a chi ec u e). His MUC model ied o gi e
30 OF 150
esponse o hese wo essen ial needs, and i en ails ha e sa ile ne wo ks o egions om
bo h hemisphe es (al hough p edominan ly le ) ake on di e en componen s o language
p ocessing ha dynamically pa icipa e bo h in speech p oduc ion and comp ehension. In
pa icula , he de ines h ee main componen s: Memo y, Uni ica ion and Con ol (hence he
name, MUC). Figu e 2.1 illus a es he b ain dis ibu ion and connec ions o he componen s
in he MUC model.
Figu e 2.1. A) Dis ibu ion o he componen s o he MUC model a he neu al le el: Memo y (yellow),
Uni ica ion (blue) and Con ol (pink). Numbe s indica e B odmann a eas. B) Schema ic illus a ion o
he connec i i y in he le hemisphe e language ne wo k. Red ci cles ep esen b ain a eas: usi o m
gy us ( g), empo al gy i ( g), angula gy us (ag), empo al pole ( p), pa s o bi alis (o ), pa s
iangula is ( ) and pa s ope cula is (op). G ey a ows depic whi e ma e ac s: in e io longi udinal
asciculus (ILF), uncina e asciculus (UF), ex eme capsule (EC) and a cua e asciculus (AF). G een
lines ep esen in e aces wi h senso y mo o sys ems: isual co ex ( c), audi o y co ex ( c) and
mo o co ex (mc). Adap ed om Hagoo (2013).
The Memo y componen comp ises all linguis ic knowledge ha was acqui ed and
consolida ed in neoco ical s uc u es. This includes in o ma ion abou phonology o wo ds,
knowledge abou hei syn ac ic s uc u es (like g amma ical gende o wo d class), and
concep ual lexico-seman ic in o ma ion. In he MUC model, phonology in o ma ion is held in
he cen al o pos e io STG. In u n, he seman ic in o ma ion is dis ibu ed along se e al
di e en a eas, mainly a ound he MTG and ITG (Hagoo , 2005), bu also pos e io ly, in he
AG (Hagoo , 2013). The e ie al o syn ac ic in o ma ion is associa ed wi h he le pos e io
STG, al hough o he manipula ion o his kind o in o ma ion, he Uni ica ion and Con ol
componen s and hei espec i e neu al pa hways become especially ele an .
Simple e ie al o s o ed linguis ic in o ma ion is no su icien o co e he in eg a i e
use o lexico-seman ic in o ma ion and i s connec ion wi h bo h di e en pieces o seman ic
knowledge and he con ex in which hey appea . In his sense, he Uni ica ion componen
e e s o he combina ion o he i ems s o ed in memo y in no el ways in o de o p oduce
assembled meaning ul s uc u es. Uni ica ion was o iginally desc ibed h ough compu a ional
31 OF 150
models a he syn ac ic le el (Vosse & Kempen, 2000). A he sen ence le el, inpu wo ds
ha e associa ed syn ac ic s uc u es ( e ie ed om memo y) ha can be ambiguous. As
wo ds a e accessed, hei syn ac ic s uc u es a e selec ed and subjec o la e al inhibi ion
p ocesses ha esol e his ambigui y. This whole mechanism is mainly associa ed wi h he
ac i i y o he mid o pos e io le IFG (BA 45, 44) (Hagoo , 2005; Pe e sson, 2004). In he
MUC model, Hagoo ex ends he uni ica ion componen o he seman ic and phonological
sphe es. Seman ic Uni ica ion would en ail he p ocesses ha guide he selec ion o he
app op ia e meaning among all po en ial accep a ions o a wo d, o he link be ween he wo d
and he p eceding con ex . Phonological Uni ica ion e e s o he selec ion o ele an
segmen s wi hin an in ona ion o guide he ca ego isa ion o he phonological uni s o ming
wo ds and sen ences. Al hough seman ic and phonological Uni ica ion ha e been less
consis en ly s udied a he neu al le el, he MUC model p oposes ha seman ic Uni ica ion is
linked o he an e io le IFG (BA 45, 47), while phonological Uni ica ion is associa ed wi h he
pos e io le IFG (BA 44) and adjacen pos e io a eas including he supplemen a y mo o
a ea (SMA, BA 6) (Fedo enko e al., 2012; Hagoo , 2005, 2013). Figu e 2.2 depic s he
dis ibu ion o Uni ica ion p ocesses in he le IFG.
Figu e 2.2. Dis ibu ion o he seman ic, phonological and syn ac ic
Uni ica ion componen s in he le IFG. Taken om Hagoo (2013).
And inally, he Con ol componen e e s o he use o execu i e con ol o pu he uni s
o language in ela ion o ac ion and con ex . When p ocessing language, like, o ins ance,
du ing eading, ou b ain needs o ex ac he essen ial in o ma ion ha allows us o deciphe
he isual inpu in o meaning ul in o ma ion. In his ask, ou p e ious knowledge and
38 OF 150
model, hese con e gence a eas include o he egions ou side he ATL, like he AG and SMG,
he MTG and ITG, and he an e io OTC. Ano he ele an di e ence is he impo ance
conceded o hese con e gence a eas. While Binde and Desai ecognise he ole o
con e gence zones in ep esen ing seman ic abs ac in o ma ion, hey de end a model in
which senso y-mo o in o ma ion can gi e ise o mul imodal ep esen a ions wi hou he
exp ess need o an amodal hub, which is why hey e med hei iew embodied abs ac ion.
Unde his pe spec i e, concep ual ep esen a ions a e o med by di e en le els o
abs ac ion om senso y, mo o and a ec i e in o ma ion. The access o each le el o
abs ac ion is de e mined by se e al ac o s, like equency o amilia i y wi h he in o ma ion
e ie ed, o he demands o he ask du ing which he in o ma ion is used. Abs ac
ep esen a ions a e su icien in highly amilia con ex s, while in no el con ex s senso y, mo o
and a ec i e ep esen a ions a e addi ionally equi ed. Thus, hey desc ibe ou le els in he
neu oana omy o hese ep esen a ions (see Figu e 2.5): 1) senso y, mo o and a ec i e a eas
(e.g., p ecen al gy us o pos e io STG); 2) abs ac con e gence zones (e.g., AG o ITG); 3)
con ol a eas ha di ec he selec ion o in o ma ion acco ding o he goals (e.g. dmPFC and
IFG); and 4) a eas ha ac as an in e ace be ween seman ic and episodic in o ma ion (e.g.
p ecuneus and PCC). Impo an ly, hese modules a e lexible and highly in e ac i e h oughou
he le els o abs ac ion.
Figu e 2.5. Embodied abs ac ion neu oana omical model. Senso y, mo o and a ec i e egions
(yellow a eas) cons i u e he pe cep ual inpu o abs ac con e gence zones in pa ie al and empo al
co ices ( ed a eas). Con ol egions, like he IFG and dmPFC selec he in o ma ion di ec ed o he
goal a hand. In e ace a eas (in g een), like he p ecuneus and PCC, connec seman ic memo y o
39 OF 150
he episodic expe ience, h ough hei connec ions wi h he MTL. Taken om Binde & Desai (2011).
2.2.3. Con olled Seman ic Cogni ion (CSC) (Ralph e al., 2017)
Almos a decade a e he dis ibu ed-plus-hub heo y was pu o wa d, hei
p oponen s in eg a ed he mos ecen e idence ega ding he unc ion o he ATL and i s
unc ional seg ega ions, as well as o he ela ed s uc u es, in a e ised e iew o he
neu ocompu a ional unde pinnings o seman ic cogni ion (Ralph e al., 2017). They mainly pu
oge he wo lines o esea ch o p opose a double mechanism o seman ic cogni ion: 1) a
ep esen a ion sys em, dedica ed o he abs ac ion o pe cep ual inpu o di e se na u e and
hei in e ela ions (Lambon Ralph e al., 2010); and 2) a con ol sys em ha modula es he
ep esen a ions o adap o he con ex and i s demands (Bad e e al., 2005). Wi h his iew,
e e ed o as con olled seman ic cogni ion (CSC), he au ho s econcile some o he mos
ele an seman ic esea ch ocused on he p ocess, and ha ocused on he seman ic
ep esen a ions pe se.
The ep esen a ion sys em has al eady been desc ibed in he model p oposed by
Pa e son and colleagues (see subsec ion 2.2.1). Howe e , his upda ed e iew included some
in e es ing nuances o he ATL unc ion. Fi s , he ATL ole as an amodal hub ha in eg a es
di e en sou ces o in o ma ion seems o be mo e speci ically cen ed a ound he en ola e al
ATL (including ITG and medial FFG) (Abel e al., 2015; Shimo ake e al., 2015). Second, he
40 OF 150
ATL can be u he s uc u ally and unc ionally seg ega ed in o sub egions ha con ey
di e en so s o in o ma ion. F om neu oana omical esea ch, we know ha he ATL is
composed o g aded neu onal ensembles (Ding e al., 2009) ha ecei e s uc u al
connec ions om di e en a eas, wi h ce ain deg ee o p e e ence. Fo ins ance, he uncina e
asciculus connec s he o bi o on al co ex and an e io IFG mos s ongly o he empo opola
co ex, while he ex eme capsule and middle longi udinal asciculus connec he MTG and
IPL wi h he supe io pa o he ATL. In u n, he ILF connec s he isual co ex and he OTC
o he en al en omedial ATL (Binney e al., 2012). Impo an ly, his s uc u al connec i i y
is belie ed o be e lec ed in a unc ional specialisa ion o he ATL, in he o m o a g adien :
he en ola e al ATL ep esen s abs ac ca ego ies ha do no depend on he inpu modali y
o s imulus ca ego y (Visse e al., 2012); he medial ATL is less c oss-modal, and shows a
p e e ence o pic u e-based ma e ials and conc e e concep s (Cla ke & Tyle , 2015; Ho man
e al., 2015); he supe io ATL (an e io STG) shows a p e e ence o audi o y s imuli, spoken
language and abs ac concep s (Ho man e al., 2015); and he pola ATL (i s mos an e io
po ion) is p e e en ially ec ui ed by social concep s (Olson e al., 2013). Figu e 2.6 illus a es
his g aded s uc u al and unc ional seg ega ion o he ATL.
Figu e 2.6. A) Compu a ional model o he g aded unc ion o he ATL. B) Neu oana omical
amewo k o he unc ional specialisa ion o he ATL and i s inpu connec ions. Taken om Ralph e
al. (2017).
The second componen o he CSC, he con ol sys em, is an homologous o he
al eady desc ibed se o mechanisms o he cogni i e con ol o seman ic knowledge (see
sec ion 2.1), wi h some added pa icula i ies. Fi s , he a eas ha play an impo an ole in
execu i e con ol include a eas pos e io o he IFG, like he p e-SMA, and empo o-pa ie al
egions like he pos e io MTG and STG, and he IPL. The di e ences in he implica ion o
41 OF 150
on al as opposed o empo o-pa ie al a eas a e sub le: lesions o he PFC usually lead o
pe se e an esponses o a highe ex en han empo o-pa ie al lesions do. This led he
au ho s o p opose ha on al a eas migh ha e a mo e ex ensi e implica ion in inhibi o y
mechanisms han empo o-pa ie al egions (Ralph e al., 2017).
The CSC model uni ies many o he heo ies p oposed hus a in he ligh o he
upda ed e idence. While I p esen i in a sec ion dedica ed o he models ha y o explain
seman ic ep esen a ions a he neu al le el (pa ly because i is a e-elabo a ion o he
dis ibu ed-plus-hub model), he CSC model ocuses on bo h he ep esen a ional aspec s and
on he cogni i e p ocesses ha modula e such ep esen a ions. This makes i a a he
in eg a i e model ha is well wo h desc ibing.
Some c i ical inconsis encies a ise a e e alua ing he mos in luen ial models o he
ep esen a ion, access and manipula ion o seman ic in o ma ion. We ha e al eady poin ed
ou he disc epancies be ween embodied models and models based on amodal abs ac ions.
While he o me see concep ual in o ma ion as elying on a widely dis ibu ed ne wo k o
senso y-speci ic b ain egions, he la e unde s and ha his ype o knowledge depends bo h
on senso y and mo o in o ma ion and hei abs ac gene alisa ions. To da e, i is s ill unclea
o wha ex en seman ic in o ma ion elies on senso imo o ep esen a ions o gene alised
abs ac ions, and how hese ep esen a ions a y as a unc ion o he con ex (Ralph e al.,
2017). In a simila ein, whe he he e a e di e en neu al mechanisms o he ep esen a ion
o meaning, wi h o wi hou language p ocessing, is s ill an open ques ion (Fedo enko e al.,
2024). In his sense, language equi es a dynamic in e ac ion be ween unc ionally sepa a e
sys ems (e.g., a en ion, inhibi ion, mo o planning, e c.). Bu such sys ems a e no comple ely
independen , and o en o e lap. This makes i ha d o ease apa he ci cums ances unde
which hese ne wo ks a e ec ui ed, and in which cogni i e p ocesses hey ac ually pa icipa e.
Pe spec i es ha concei e he spa ial dis ibu ion o language ne wo ks as a g adien in which
some a eas pa icipa e dynamically in di e en cogni i e unc ions (e.g., Le ma-Usabiaga
e al., 2018; Ralph e al., 2017) do no challenge he idea ha such unc ional specialisa ion
exis s (Fedo enko e al., 2024). Ne e heless, his is s ill a poin o con lic in many speci ic
esea ch opics, including seman ic ep esen a ions (Fedo enko e al., 2024; Kanwishe ,
2010).
42 OF 150
CHAPTER III. THE STUDY OF NEURAL LEXICAL REPRESENTATIONS
Vi ually all he heo e ical models exposed hus a eme ge om wo ypes o
e idence: 1) neu opsychological s udies, like he ones ha ga e ise o he VWFA heo y, o
he s udies on ATL lesions in seman ic demen ia; and 2) neu oimaging s udies using a wide
a ie y o echniques, o which MRI has been pe haps he mos popula , due o i s po en ial
o manipula e a wide a ie y o condi ions and con ex s, and cap u e he associa ed b ain
esponses wi h high spa ial esolu ion. Al hough lesional s udies ha e been, and s ill a e,
essen ial o unde s anding b ain unc ion, neu oimaging s udies clea ly cons i u e he mos
lexible way o app oaching he analysis o cogni i e unc ioning a he neu al le el. The
possibili y o collec ing indica o s o b ain ac i i y du ing he engagemen in di e se asks and
con ex s, wi h a whole possible wo ld o manipula ions, is an ideal se ing o explo ing he
neu al unde pinnings o a gi en cogni i e p ocess. Taken o lexico-seman ic ep esen a ions,
we ha e a g ea example o his capaci y in he me a-analysis desc ibed in subsec ion 2.2.2:
by pu ing oge he 120 s udies ha used MRI o measu e b ain ac i a ion du ing simila
a ia ions o seman ic asks, we could ge a e y good app oxima ion o he egions ha a e
likely o be in ol ed du ing seman ic p ocessing. An addi ional poin ha we can ex ac om
his is he necessi y o ha ing good spa ial esolu ion, especially when i comes o explo ing
neu al ep esen a ions. Th oughou he p esen wo k, we ha e alluded o ine-g ained spa ial
di isions in se e al di e en b ain a eas, some hing ha MRI makes possible. Fo his eason,
we will ocus on MRI, he echnique employed in he wo ks p esen ed in his hesis.
Many o he conclusions d awn in he heo e ical models desc ibed in he p e ious
chap e we e possible due o he manipula ion o speci ic psycholinguis ic ea u es cap u ed
in language. Fo ins ance, he compa ison be ween abs ac and conc e e concep s in MRI
asks is one o he mos impo an con ibu o s o he e idence suppo ing p ac ically all
heo e ical amewo ks p esen ed (e.g., Binde e al., 2009; Ho man e al., 2015). Addi ionally,
some o he mos ecen indings abou he unc ional specialisa ion o he OTC we e possible
due o he use o manipula ions ha a ec ed lexical ea u es like he leng h o a wo d o
p ope ies ha a ec i s eadabili y (e.g, Le ma-Usabiaga e al., 2018; Whi e e al., 2019).
These possibili ies ha e been conside ably imp o ed wi h ecen ad ances in he ield
o compu a ional language modelling and wi h he popula isa ion o no el analy ical
app oaches applied in MRI da a analysis. The wo ks p esen ed in he cu en hesis
ha nessed he manipula ion o psycholinguis ic p ope ies, pa ially making use o
compu a ional language models and no el MRI mul i a ia e analy ical app oaches. Fo hese
easons, in his chap e I will desc ibe he mos ele an psycholinguis ic p ope ies associa ed
wi h lexico-seman ic access, ollowed by na u alis ic language p ocessing models, and a
43 OF 150
speci ic o m o mul i a ia e analysis app oach o MRI da a called ep esen a ional simila i y
analysis (RSA).
3.1. Psycholinguis ic P ope ies as a Window o Lexical Rep esen a ions
Whe he pe cep ibly o unconsciously, he chunks o in o ma ion ha we use in ou
e e yday li es e lec an ex ensi e a ie y o p ope ies ha a ec how ha speci ic pa o
seman ic knowledge is p ocessed. These chunks o in o ma ion, ha we can e e o as
concep s, a e cap u ed in wha we ha e been naming lexical uni s: wo ds. Wo ds can a y,
o ins ance, in how o en hey appea on media (lexical equency), o how amilia o us we
eel ha he con en ha he wo d e e s o (wo d amilia i y). They can e e o conc e e en i ies
ha we can pe cei e o eel in he wo ld a ound us, o be abs ac and no ha e any e iden
physical analogue in he eal wo ld (wo d conc e eness). This, in u n, will a ec how easily
we can c ea e a men al isualisa ion o sensa ion o he en i y e e ed o by he wo d (wo d
imageabili y). They can allude o li ing beings o inanima e en i ies, man made objec s (wo d
animacy), and he e e ed en i ies can be s onge examples o a gi en ca ego y (e.g., dog
as a mammal) han o he s (e.g. ba , also as a mammal) ( axonomic hie a chy). All hese
ea u es a e examples o seman ically- ela ed a iables ha e lec meaning ul p ope ies o
he concep ual en i ies. Impo an ly, manipula ing he deg ee o which wo ds a y in hese
con inua exe s a clea in luence o e he b ain ac i a ion pa e ns gene a ed by accessing
he concep s e e ed o. This has been used o in o m abou he implica ion o di e en b ain
ne wo ks in dis inc cogni i e p ocesses. As examples, he neu al e ec s o manipula ing
animacy (Coggan & Tong, 2023; G ill-Spec o e al., 2017; G ill-Spec o & Weine , 2014;
Jozwik e al., 2022) and axonomic hie a chy (Ri chie e al., 2021), o en s udied in he con ex
o isual ca ego isa ion o pic u es, ha e g ea ly con ibu ed o ou knowledge abou he
unc ional specialisa ion o he OTC (see Tho a e al., 2019, and subsec ion 1.1.2).
3.1.1. Wo d Conc e eness and Imageabili y
As men ioned ea lie , a g ea deal o wha we know abou how he b ain ep esen s
concep ual in o ma ion comes om he analysis o he conc e eness con inuum. This
con inuum is o en de ined subjec i ely, by asking a la ge sample o pa icipan s o a gi en
language o a e an ex ensi e se o wo ds om 1 (absolu ely no pe cep ible/abs ac ) o 7
(absolu ely pe cep ible/conc e e) (e.g., Duchon e al., 2013). The eade migh be wonde ing:
how can such a speci ic ea u e o language ell us so much abou such a complex
neu ocogni i e abili y? A po en ial answe o his is cap u ed by he dual-coding heo y (DCT)
(Pai io, 2010, 2014). This iew unde s ands ha ep esen a ions can be acqui ed and held
ei he h ough pe cep ual encoding o ia e bal encoding. Conc e e concep s ollow a double
44 OF 150
encoding p ocess, since he e e ed en i ies can be pe cei ed h ough ou senses, and
e balised h ough language. In con as , abs ac concep s a e mos ly acqui ed h ough
language, hus coun ing on weake suppo om pe cep ual expe ience. The DCT can explain
he well epo ed beha iou al e ec o conc e eness (i.e., abs ac concep s being ha de o
p ocess and mo e e o ul o lea n) (Feye eisen e al., 1988; Mk ychian e al., 2019; Palme
e al., 2013; Schwanen lugel e al., 1988). The ac ha abs ac wo ds a e usually lea ned
la e in li e, and a e on a e age less amilia han conc e e wo ds is also in line wi h he DCT
(S iem-Ami e al., 2018). And mos impo an ly, he neu oimaging indings o e he las 20
yea s a e o a g ea ex en consis en and in line wi h he DCT. Conc e e concep s mainly
ec ui dis ibu ed modali y-speci ic senso y and mo o a eas like he p ecuneus, PCC, FFG
and pa ahippocampal gy us (Ho man e al., 2015; Vignali e al., 2023; J. Wang e al., 2010),
bu also mul imodal and language a eas o some ex en (Binde e al., 2009; Ho man e al.,
2015; Vignali e al., 2023). In con as , abs ac wo ds a e associa ed wi h esponses in
mul imodal a eas like he ATL (Ho man e al., 2015; Vignali e al., 2023; J. Wang e al., 2010),
he le IFG (Binde e al., 2009; J. Wang e al., 2010) and le pos e io MTG and STG (Binde
e al., 2009; J. Wang e al., 2010). Figu e 3.1 shows he neu al dis ibu ion o he ac i a ions
associa ed wi h conc e e s. abs ac wo ds, aken om he me a-analyses e alua ing his
compa ison by Wang e al (2010) (bu see also Binde e al 2009).
Figu e 3.1. Ac i a ion likelihood o abs ac and conc e e wo ds, aken om he me a-analysis by
Wang e al (2010).
45 OF 150
Howe e , he e is no absolu e ag eemen in he in e p e a ion o he neu al e ec s o
abs ac as opposed o conc e e concep ual p ocessing. The mo e embodied pe spec i es
a gue ha many abs ac concep s a e g ounded in si ua ions, men al s a es, e en s o
ela ions be ween objec s, he same way ha conc e e concep s can e lec abs ac ea u es
(Bo ghi & Binko ski, 2014). In addi ion, abs ac concep s a e o en associa ed wi h highe
deg ees o a ec i e con en . Some s udies demons a ed ha wo ds ha can be g ounded in
emo ions a e lea ned ea lie in li e, and bene i om his emo ional alence g ounding, while
conc e e wo ds do no (Pona i e al., 2018). Mo eo e , some o he neu al e ec s associa ed
wi h abs ac wo d p ocessing can be pa ially explained by he in luence o a ec i e alence.
Fo ins ance, ac i a ions obse ed in he os al ACC could be ela ed wi h he p ocessing o
hedonic alue cap u ed by abs ac wo ds (Vigliocco e al., 2014). Simila ly, a ec i e alence
has been ound o be well ep esen ed in he supe io ATL only o abs ac wo ds p ocessing
(Mee smans e al., 2020).
A ac o ha is igh ly ela ed wi h wo d conc e eness is imageabili y (a measu e ha
is also ob ained om subjec i e a ings in no ma i e s udies). In ac , p e ious s udies ha e
epo ed conside ably high co ela ions be ween hese wo ac o s (Wes bu y e al., 2013).
None heless, he ex en o which a wo d can e oke a angible isualisa ion/sensa ion can add
some in o ma ion abou seman ic ep esen a ions. Fo ins ance, a wo d (e.g., e e nal) can be
judged as clea ly abs ac , bu as pa ially imageable (Wes bu y e al., 2013), pe haps owing
o i s capaci y o e oke bodily sensa ions and men al image y. This can p o ide addi ional
de ails abou he e ec s obse ed in MRI s udies. Fo example, al hough MRI e idence ound
when compa ing low- e sus high-imageabili y wo ds o e lap conside ably wi h conc e eness
neu al e ec s (i.e., ac i a ion o modali y-speci ic senso y a eas like he FFG when p ocessing
imageable concep s) (Bedny & Thompson-Schill, 2006; Lewis & Poeppel, 2014), addi ional
e ec s, like he in ol emen o he hippocampus when p ocessing easily imageable concep s
ha e been epo ed (Caplan & Madan, 2016; Kla e e al., 2005). Th oughou he es o he
hesis, while I conside he men ioned high collinea i y wi h wo d conc e eness, I assume ha
wo d imageabili y can s ill con ibu e o a be e unde s anding o seman ic neu al
ep esen a ions.
3.1.2. Wo d F equency and Familia i y
On a di e en ein, ano he ac o ha has con ibu ed eno mously o ou
unde s anding o he cogni i e and neu al mechanisms in ol ed in he access and
ep esen a ion o lexical knowledge is wo d equency. Ou b ains can ake ad an age o
epea ed exposu e o wo ds in di e en ways. Upon encoun e ing a new wo d, we ace he
46 OF 150
labo ious ask o deciphe ing i s undamen al o hog aphic and phonological componen s.
Th ough epea ed exposu e, hese no el wo ds can become consolida ed, inco po a ing hei
phonological, mo phological, and syn ac ic cha ac e is ics (Hagoo , 2013). They a e u he
connec ed o o he wo ds and concep s, and can become seman icised in o an amodal
seman ic ne wo k (see sec ion 2.2; Meme o a e al., 2024). This p og ession s eamlines he
eading p ocess, educing he need o exhaus i e pe cep ual analysis o he wo d, and
enhancing eading e iciency (Desai e al., 2020). We can connec his p ocess o he
dis inc ion be ween he do sal (o hog aphic-phonological) and en al (lexico-seman ic)
language ne wo ks explained in sec ion 1.3: eading unknown wo ds equi e he in ol emen
o he do sal phonological ne wo k, while wo ds ha a e well known o us may use he en al
lexico-seman ic ou e o p ocess he wo d as a whole, wi hou he need o deciphe i s
o hog aphic and phonological componen s.
In p ac ice, he wo d equency e ec (WFE) is a well-documen ed phenomenon in he
neu oscience o eading. I e e s o he ac ha low- equency wo ds pose g ea e p ocessing
challenges compa ed o high- equency ones. Ex ensi e beha iou al esea ch consis en ly
demons a ed as e esponses o high- equency wo ds as opposed o low- equency wo ds
ac oss a a ie y o asks such as wo d naming, lexical decision, and seman ic decision asks
(B ysbae e al., 2018). A he neu al le el, se e al MRI s udies ha e analysed he WFE. In
hose s udies, he WFE is de ined as heigh ened egional b ain ac i a ion o low- equency
wo ds, sugges ing mo e demanding cogni i e p ocessing. No ably, he IFG eme ges as a key
neu al locus o his e ec , al hough o he s udies ha e ound a WFE in he OTC, and less
consis en ly in o he a eas, like SMA o ACC. Table 3.1 p o ides a syn hesis o p e ious MRI
in es iga ions on he WFE, ou lining hei key indings, me hodological pa ame e s, and
sample sizes.
Table 3.1 P e ious MRI s udies examining he WFE, main indings, eading asks used and
sample sizes.
S udies
Regions showing he WFE
Reading Tasks
Sample
sizes (N)
Chee e al. (2002)
Le IFG (BA 44)
Silen eading s.
seman ic judgemen s
16
Fiebach e al. (2002)
Le IFG (BA 44, 45)
Lexical Decision Task
12
Chee e al. (2003)
Le ACC (BA 32), IFG (BA 44,
45), ITC (BA 37)
Seman ic judgemen +
24h Recogni ion
16
47 OF 150
Kuo e al. (2003)
Le P ecen al, SMA, IFG (BA
44), OTC
Co e naming +
Recogni ion
28
Joube e al. (2004)
Le IFG (BA 45, 47)
Silen eading
10
K onbichle e al.
(2004)
Le IFG (BA 45, 47), OTC
(mid)
Silen eading
13
Ca ei as e al. (2006)
Le IFG (BA 44)
Lexical Decision s
Reading aloud
16
G a es e al. (2007)
Le IFG (BA 45/47), OTC
(pos ), pSTG
Pic u e naming (o e )
59
Hauk e al. (2008)
Bil. IFG (BA 45, 47), OTC (an )
Silen eading
21
B uno e al. (2008)
Le P ecen al, IFG (BA 44, 45),
OTC, pSTG
Phonological Lexical
Decision Task
28
Ca ei as e al. (2009)
Bil. ACC/ IFG (BA 45, 47),
P ecuneus, SMA
Lexical Decision Task
20
Schus e e al. (2016)
Le IFG (BA 45), OTC
Silen Sen ence
Reading
56
Rundle e al. (2018)
Le ITC (BA 37); OTC
Silen eading +
seman ic ca ch ial
19
These indings ca y signi ican implica ions o he unde s anding o he neu al
ne wo ks in ol ed in p ocessing in o ma ion om phonological o seman ic. Howe e , he
in e p e a ion o hese esul s is di e se. In sho , while some au ho s a ibu e he e ec o he
sea ch and access o phonological in o ma ion (Ca ei as e al., 2009; Fiebach e al., 2002),
o he s in e p e i as an indica o o mo e e o ul sea ch o seman ic in o ma ion (Chee e al.,
2002). In u h, he phonological and seman ic in e p e a ions o he WFE a e no comple ely
incompa ible. P ocessing a wo d ha appea s o en on media can be easie bo h because he
access o i s o hog aphic (syllables) and phonological (phonemes) componen s is enhanced
a e epea ed exposu e o i . Bu as I indica ed be o e, a p o icien eade o en eads known
54 OF 150
and he concep ual ma ix (K iegesko e, 2008; Popal e al., 2019). Figu e 3.4 ep esen s a
schema ic applica ion o RSA.
In he las ew yea s, RSA has been success ully applied o explo e he neu al
ep esen a ions o a a ie y o dimensions in he ield o seman ics, lexical in o ma ion and
mnemonic ep esen a ions (Ca o a e al., 2017, 2021a; Liuzzi e al., 2023; Mee smans e al.,
2020, 2022; Viganò e al., 2021; Y. Wang e al., 2023; Yacoby e al., 2021). In Chap e 5, we
will co e an empi ical applica ion o RSA o disen angle he neu al ep esen a ions o , and
dynamic ela ions be ween, he mos impo an psycholinguis ic a iables, as well as NLP
models.
Figu e 3.4. Schema ic applica ion o RSA. A) RDM exp ession o oxel pa e n esponses o s imuli o
di e en na u e (he e, houses and aces). B) Mul iple possibili ies o he compa ison be ween he RDM
and an empi ical model. The empi ical model can be buil om a wide ange o ea u es, om
compu a ional, o da a om o he echniques (e.g. EEG o cell eco dings), and e en o he species.
Adap ed om K iegesko e e al (2008).
55 OF 150
CHAPTER IV. THE ROLE OF READING DEMANDS AND WORD FREQUENCY IN THE
ACCESS TO LEXICAL UNITS
4.1. RATIONALE
As we desc ibed in sec ion 3.1.2, esea ch on he WFE has ypically in ol ed di e se
me hodological se ings. The MRI me hodological p ocedu es used in p e ious s udies ha e
a ied g ea ly, including di e ences in p o ocols o mul iple compa ison co ec ions o e en
he absence o such co ec ions o s a is ical signi icance h esholds. This di e si y has
in luenced he obse ed e ec s o equency in he b ain egions desc ibed in sec ion 3.1.2,
making he WFE esul s di icul o in e p e . The inconsis encies in he loca ion o he neu al
ac i a ions ound has con ibu ed o a ce ain deg ee o di e si y in he in e p e a ion o hese
e ec s. In some o hese s udies, he WFE ex ends o he an e io pa o he IFG (i.e., pa s
o bi alis and pa s iangula is, BA 47 and 45, espec i ely), while in o he s i is associa ed wi h
he pos e io IFG (i.e., pa s ope cula is, BA 44), o bo h (see Table 3.1). In consequence, while
some esea che s associa e he obse ed WFE in he IFG wi h phonological p ocessing o
e ie al du ing lexical sea ch (Ca ei as e al., 2009; Fiebach e al., 2002), o he s sugges i
hinges on delibe a e access o seman ic in o ma ion (Chee e al., 2002). As b ie ly in oduced
in he p e ious chap e , bo h in e p e a ions a e no comple ely incompa ible. He e we p opose
ha , i he WFE elies on phonological p ocesses, hen i would be mainly (al hough no
exclusi ely) obse able in egions along he do sal eading ne wo k (i.e., IFG pa s ope cula is,
STG, and/o IPC) in ol ed in mapping isual pe cep s on o he phonological s uc u e o he
language. In con as , i he WFE elies on lexico-seman ic p ocessing, we p edic ha he
WFE will mainly (bu no exclusi ely) ely on he engagemen o egions along he en al
eading ne wo k (i.e., IFG pa s iangula is, IFG pa s o bi alis, and/o he OTC), in ol ed in
mapping o hog aphic-lexical s imuli o wo ds as a whole (Oli e e al., 2017; Sandak e al.,
2004). Indeed, he s udies de ending he seman ic in e p e a ion ha e ypically ound he WFE
in egions o he en al eading ne wo k, whe eas he s udies de ending he phonological
in e p e a ion ha e ypically ound he WFE o occu in egions along he do sal eading
ne wo k (see Table 3.1).
The ela ionship be ween phenomena such as he WFE and he unc ional
specialisa ion o he OTC has also spa ked deba e. The eade will emembe om sec ion
1.1.3 ha his egion con ains he pu a i e VWFA, which would be, acco ding o i s p oponen s,
exp essly dedica ed o he isual ecogni ion o wo d o ms. Con e sely, we can also b ing
back he idea o he OTC esponding o lexical and seman ic p ope ies (especially i s an e io
po ion), and e en o s imuli o he han isual wo ds. The ac ha he OTC has been ound
o espond o wo d equency is pa o he e idence agains he VWFA pe spec i e
(K onbichle e al., 2004; Kuo e al., 2003; Schus e e al., 2016). Ne e heless, while a numbe
56 OF 150
o s udies ha e ound a WFE in he OTC (B uno e al., 2008b; G a es e al., 2007; Hauk e
al., 2008; K onbichle e al., 2004; Rundle e al., 2018; Schus e e al., 2016), o he s ha e
ailed o ind such an e ec (Ca ei as e al., 2006, 2009; Chee e al., 2002, 2003; Fiebach e
al., 2002; Joube e al., 2004), and when ound, he loca ions o he e ec in he OTC a e
somewha inconsis en . These ac s ende he WFE in he OTC challenging o in e p e .
On he o he hand, i has been indica ed ha he engagemen o bo h he IFG and he
OTC could be modula ed by op-down p ocesses, such as hose imposed by eading
demands (P ice & De lin, 2011; Rundle e al., 2018; Yang & Ze in, 2014). P e ious MRI
s udies ha e used a a ie y o condi ions and asks, anging om silen eading o lexical
decision o seman ic decision asks (see Table 3.1). Some au ho s ha e p oposed ha , in he
case o he IFG, lexical decisions (Rundle e al., 2018) o naming e o s (Vogel e al., 2013)
migh be ampli ying he e ec s ound in his egion. In he case o he OTC, he p olonged
exposu e ime, combined wi h he demands o asks like seman ic judgemen o lexical
decision asks, could make his egion o be mo e engaged du ing he p ocessing o low-
equency wo ds (Schus e e al., 2016). Again, hese con as ing in e p e a ions a e especially
con o e sial in ega d o he main heo e ical accoun s o he OTC.
Finally, and adding o he men ioned limi a ions, mos p e ious MRI s udies on he
WFE ha e concen a ed on egional ac i a ion, neglec ing o examine he unc ional
connec i i y pa e ns ha migh unde lie his e ec . Gi en hese inconsis encies, ou p ima y
objec i e was o examine he WFE in he ac i a ion p o iles o egions wi hin he en al and
do sal eading ne wo ks using complemen a y analy ical app oaches. We also aimed o
in es iga e he po en ial in e ac ion be ween eading demands and wo d equency in hese
unc ional pa e ns. To achie e his, we employed wo e sions o a single-wo d eading ask:
a pe cep ual ask (low eading demand) and a seman ic ask (high eading demand). This
allowed us o assess he in luence o eading demands imposed by hese wo ypes o asks
on he WFE. Addi ionally, we aimed o de e mine whe he he e we e dis inguishable
unc ional connec i i y p o iles among en al and do sal eading egions ha espond o wo d
equency and eading demands. Consis en wi h p e ious e idence and heo ies ega ding
he di ision o labou be ween hese wo ne wo ks (e.g., Pugh e al., 2001), we expec ed o
obse e he WFE in egions o he lexico-seman ic en al eading ne wo k, such as he IFG
and he OTC, cha ac e ised by highe ac i a ion in hese a eas o low- equency wo ds
compa ed o high- equency wo ds. Fu he mo e, we p edic ed his e ec would be s onge
in he seman ic ask han in he pe cep ual ask and ha hese indings would be eplica ed
ac oss di e en analy ical app oaches. In e ms o unc ional connec i i y, we an icipa ed ha
he WFE would be associa ed wi h s onge unc ional connec i i y wi hin egions along he
57 OF 150
en al eading ne wo k. The me hods and esul s discussed in he ollowing sec ions we e
published elsewhe e as o Oc obe o 2023 (Sánchez e al., 2023)
4.2. METHODS
4.2.1. Pa icipan s
The o al sample o he s udy consis ed o 54 igh -handed na i e Spanish-speaking
pa icipan s, o 29.3 yea s o age on a e age (SD = 6.88 yea s; 30 emales). All pa icipan s
had no mal o co ec ed- o-no mal ision, and no known his o y o neu ological o psychia ic
illness. O he ini ial 57 pa icipan s, one pa icipan was excluded due o excessi e head
mo ion du ing scanning (see MRI da a analysis sec ion below), and wo pa icipan s we e
excluded due o he absence o eco ded esponses o he ca ch (i.e., Go) ials in he MRI
asks. Language p o iciency was assessed using bo h objec i e and subjec i e measu es. As
an objec i e measu e, we used an adap ed Spanish e sion o he Bos on Naming Tes (de
B uin e al., 2017). As a subjec i e measu e, pa icipan s illed in a language p o iciency sel -
a ed ques ionnai e, in which hey e alua ed hei own p o iciency, as well as language
exposu e. All pa icipan s ga e w i en in o med consen in compliance wi h he e hical
egula ions es ablished by he BCBL E hics Commi ee and he guidelines o he Helsinki
Decla a ion. All pa icipan s ecei ed mone a y compensa ion o hei pa icipa ion.
4.2.2 Ma e ials and P ocedu e
The expe imen al design consis ed o wo single-wo d eading Go/No-Go asks, one
pe cep ual (low eading demand) and he o he seman ic (high eading demand). In bo h asks,
all pa icipan s we e isually p esen ed wi h cha ac e s ings in hei na i e language (i.e.,
Spanish) ha could be wo ds o nonwo ds. In he pe cep ual ask, pa icipan s we e asked o
p ess a bu on any ime hey saw a colou ed le e wi hin a s ing. In he seman ic ask,
pa icipan s had o p ess a bu on any ime hey ead a wo d e e ing o an animal (ins ead o
s ings con aining a colou ed le e ). All s imuli we e p esen ed o 1.5 seconds on he cen e
o he sc een. The asks we e di ided in di e en unc ional uns ha we e coun e balanced
be ween pa icipan s.
Each ask included a o al o 80 wo ds, o which 40 we e high equency wo ds and 40
we e low equency wo ds, and 80 nonwo ds. Thus, wo se s o wo ds and co esponding
nonwo ds we e de eloped and hei use in ei he he pe cep ual ask (i.e., low eading
demand) and he seman ic ask (i.e., high eading demand) was coun e balanced be ween
subjec s. Wo d equency is objec i ely de ined by measu ing he numbe o appea ances o
a gi en wo d, pe million wo ds, on a la ge sample o ex sou ces (B ysbae e al., 2018). I s
mos commonly used measu e is he loga i hmic ans o ma ion o he equency pe million
58 OF 150
wo ds, he Zip scale, which anges om 1 ( e y low equency) o 7 ( e y high equency)
(Van Heu en e al., 2014). In bo h se s, low- equency wo ds we e nouns wi h a Zip measu e
lowe han 4, and high- equency wo ds we e nouns abo e his cu o . All wo d measu es we e
ob ained om EsPal (Duchon e al., 2013), and he wo se s o wo ds we e ma ched on
equency, leng h (i.e., 5-8 cha ac e s) and numbe o o hog aphic neighbou s. Nonwo d
s ings we e included as s imuli in he expe imen al design o add ess o he esea ch
ques ions no ele an o he p esen wo k. To educe he po en ial eading demands imposed
by nonwo ds as much as possible, hey we e designed so ha hey we e legal, legible s ings.
Fu he mo e, he wo se s o nonwo ds we e also ma ched in leng h. Addi ionally, we included
13% o Go ials (i.e., ei he wo ds wi h a colou ed le e o animal wo ds) as ca ch ials o
each o he wo eading asks. Nonwo ds and Go ials we e modelled as eg esso s o in e es
bu no analysed. The s imuli used o he pe cep ual and seman ic eading asks we e also
coun e balanced be ween subjec s.
4.2.3. MRI Da a Acquisi ion
Whole-b ain MRI da a we e ob ained on a 3-T Siemens TRIO whole-body MRI
scanne (Siemens Medical Solu ions) a he Basque Cen e on Cogni ion, B ain and Language
(BCBL), using a 32-channel whole-head coil. The a ea be ween he pa icipan s’ heads and
he coil was padded wi h oam in o de o educe head mo emen , and he pa icipan s we e
asked o s ay as s ill as possible. Snuggly i ing headphones (MR Con on) we e used o
dampen backg ound scanne noise and o allow communica ion be ween pa icipan s and
expe imen e s.
The unc ional images we e acqui ed using a g adien -echo echo-plana pulse
sequence wi h he ollowing pa ame e s: ime epe i ion (TR) = 2000 ms, ime echo (TE) = 25
ms, 35 con iguous 3-mm axial slices, 0-mm in e slice gap, lip angle= 90º, ield o iew = 218
mm, 64 x 64 ma ix. The i s ou olumes o each scan we e disca ded o allow T1-
equilib a ion e ec s. The o de o he condi ions o he s udy wi hin each un, as well as he
in e - ial in e als o a iable du a ion, we e de e mined wi h an algo i hm designed o
maximise he e iciency o he eco e y o he blood oxygen le el-dependen esponse: Op seq
II (Dale, 1999). High- esolu ion T1-weigh ed ana omical images we e also acqui ed wi h he
ollowing acquisi ion pa ame e s: TR = 2300 ms, TE = 2.97 ms, lip angle = 9º, Field o iew =
256 mm, 176 olumes pe un, oxel size = 1 cubic mm.
4.2.4. MRI Da a Analyses
S anda d SPM12 (Wellcome Depa men o Cogni i e Neu ology, London, UK)
p ep ocessing ou ines and analysis me hods we e employed. Images we e co ec ed o
59 OF 150
di e ences in iming o slice acquisi ion and ealigned o he i s olume by means o igid-
body mo ion ans o ma ion. Mo ion pa ame e s we e ex ac ed om his p ocess and we e
used, a e a pa ial smoo hing o 4-mm ull wid h a hal -maximum (FWHM) iso opic Gaussian
ke nel, o in o m addi ional mo ion co ec ion algo i hms implemen ed by he A i ac Repai
oolbox (A Repai ; S an o d Psychia ic Neu oimaging Labo a o y), in ended o epai ou lie
olumes wi h sudden scan- o-scan mo ion exceeding 0.5 mm and olumes whose signal
luc ua ions in global in ensi y was > 1.3 % SD away om he mean. The co ec ion o hese
ou lie olumes was pe o med ia linea in e pola ion be ween he nea es non-ou lie ime
poin s (Mazaika e al., 2009). Da a om 1 subjec equi ing mo e han 15% o hei olumes
o be epai ed was disca ded. Fo he inal sample o pa icipan s, he a e age pe cen age o
epai ed olumes was 1.8% (SD = 2.8%). A e olume epai , unc ional olumes we e co-
egis e ed o he T1 images using 12-pa ame e a ine ans o ma ion and spa ially no malised
o he Mon eal Neu ological Ins i u e (MNI) space by applying nonlinea ans o ms es ima ed
by de o ming he MNI empla e o each indi idual’s s uc u al olume. Du ing no malisa ion,
he olumes we e sampled o 3-mm cubic oxels. Func ional olumes we e hen smoo hed
wi h a 7-mm FWHM iso opic Gaussian ke nel. Due o he quad a ic ela ion be ween sepa a e
smoo hing ope a ions, he o al smoo hing applied o he unc ional da a was app oxima ely
equi alen o smoo hing wi h an 8-mm FWHM Gaussian ke nel. Finally, ime se ies we e
empo ally il e ed o elimina e con amina ion om slow equency d i (high-pass il e wi h a
cu o pe iod o 128 s).
S a is ical analyses we e pe o med on indi idual pa icipan da a using he gene al
linea model (GLM). MRI ime se ies da a we e modelled by a se ies o impulses con ol ed
wi h a canonical hemodynamic esponse unc ion (HRF). The mo ion pa ame e s o
ansla ion (i.e., x, y, and z) and o a ion (i.e., yaw, pi ch, and oll) we e included as co a ia es
o non-in e es in he GLM. Each ial was modelled as an e en , ime-locked o he onse o
he p esen a ion o each cha ac e s ing. The esul ing unc ions we e used as co a ia es in
a GLM, along wi h a basic se o cosine unc ions ha high-pass il e ed he da a. SPM12
FAST was used o empo al au oco ela ion modelling in his GLM due o i s op imal
pe o mance in e ms o emo ing esidual au oco ela ed noise in i s -le el analyses
(Olszowy e al., 2019). The leas -squa es pa ame e es ima es o he heigh o he bes - i ing
canonical HRF o each s udy condi ion we e used in pai wise con as s. Rega ding such
analyses, whole-b ain con as s we e compu ed by pe o ming one-sample - es s on he
con as images.
Region-o -in e es (ROI) analyses we e ca ied ou by using he MARSBAR oolbox o
SPM12 (B e e al., 2002). Six le -la e alised egions along he eading ne wo k we e
unc ionally iden i ied using wo di e en p ocedu es: I) g oup le el and II) indi idual-subjec
60 OF 150
le el. The g oup ROI iden i ica ion p ocedu e iden i ied ac i e oxels ob ained om he whole
b ain con as Wo ds > Null ac oss all pa icipan s, clus e Family-wise e o (FWE) co ec ed,
p < 0.001 oxel ex en . The egions iden i ied included pa s o bi alis (cen e o mass: -37, 27,
-8; mm3= 1656), pa s iangula is (cen e o mass: -46, 27, 14; mm3= 12704), pa s ope cula is
(cen e o mass: -48, 11, 20; mm3= 5856), MTG (cen e o mass: -46, -59, -2; mm3= 1008),
IPC (cen e o mass: -30, -53, 46; mm3= 2200) and OTC (cen e o mass: -43, -59, -17; mm3=
6296). As o he indi idual ROIs p ocedu e, he same six egions we e localised a he
indi idual-subjec le el. To his end, 5 mm adius sphe es we e c ea ed by selec ing he local
maxima in each indi idual subjec ’s Wo ds > Null con as (clus e FWE co ec ed, p < 0.001
oxel ex en ) ha all wi hin he ana omical mask o he six abo e-men ioned ROIs. Fo hose
subjec s ha had no oxels o e he h eshold ha ell wi hin he ana omical mask, he closes
local maxima ha allowed a sphe e o be buil alling wi hin he mask was selec ed. The
selec ion o he local maxima o indi idual ROIs in all pa icipan s we e sys ema ically
checked by wo au ho s (A.S. and P.M.P-A).
4.2.5. Func ional Connec i i y Analyses
We used he be a-se ies co ela ion me hod o compu e unc ional connec i i y
analyses (Rissman e al., 2004) by using cus om Ma lab sc ip s o SPM12. As in ROI
analyses, unc ional connec i i y analyses we e pe o med on bo h g oup and indi idual ROIs.
The occu ence o each e en was modelled wi h he canonical HRF, which allowed o he
ex ac ion o he pa ame e es ima es (i.e., be a co ela ions) associa ed wi h each condi ion
in e e y oxel. Following his, pai wise connec i i y be ween he 6 le -la e alised ROIs
desc ibed abo e was conduc ed. A e Bon e oni’s co ec ion, a alue o > .355 was
conside ed o show a signi ican unc ional connec i i y be ween nodes. Fu he con as s (i.e.,
- es s) on he be a co ela ions associa ed wi h low e sus high F equency, and pe cep ual
e sus seman ic Task, we e ca ied ou a e Fishe ’s Z ans o ms (Fishe , 1921) o he be a
Pea son’s co ela ions alues o make he null hypo hesis sampling dis ibu ion app oach
ha o he no mal dis ibu ion.
4.3. RESULTS
4.3.1. Beha iou al Pe o mance
All pa icipan s showed a high esponse a e o Go ials o e all, wi h an a e age
accu acy o 99.4% (SD = 0.02%) o he pe cep ual ask, and 95.4% (SD = 0.05%) o he
seman ic ask. This indica es ha pa icipan s we e ocused on he ins uc ions gi en by he
expe imen e and pe o ming he ask. As expec ed, accu acy was sligh ly, bu signi ican ly
highe o he pe cep ual ask ( = 4.880; p < 0.001). Likewise, on a e age, esponse imes
61 OF 150
we e signi ican ly as e o he pe cep ual han o he seman ic ask ( = -8.697; p < 0.001;
pe cep ual M = 560 ms, SD = 90 ms; seman ic M = 760 ms, SD = 140 ms).
4.3.2. Whole-b ain esul s
When con as ing all ials including wo ds agains baseline, he a e aged ac i a ion
map o wo ds ac oss all subjec s (see Figu e 4.1A) included egions in he occipi al co ex,
such as he lingual gy us and he cuneus, he OTC, IPC, MTG, he middle and supe io on al
gy us, he p ecen al and pos cen al gy i, and he di e en sub egions wi hin he IFG (pa s
o bi alis, pa s iangula is and pa s ope cula is).
Figu e 4.1. Whole-b ain con as s and low-high equency simple con as s. A) Resul s o he Wo ds
e sus Null con as ac oss all subjec s; B) Low > High equency con as in he pe cep ual (low eading
demand) ask (in ed), and he seman ic (high eading demand) ask (in g een). p < 0.05 FWE co ec ed
clus e wise (p < 0.001 unco ec ed oxel-ex en h eshold).
Addi ionally, we compu ed he whole-b ain Low > High equency con as s in bo h he
pe cep ual and he seman ic asks sepa a ely. This con as e lec s he b ain dis ibu ion o
he WFE in ei he ask (Figu e 4.1B). In he pe cep ual ask, no oxels su i ed he h eshold
62 OF 150
o p < 0.05 (FWE co ec ed clus e wise, wi h a p < 0.001 unco ec ed oxel-ex en h eshold).
On he o he hand, he seman ic ask showed a WFE exceeding he es ablished h eshold in
he whole le IFG, wi h global maximae loca ed in he an e io IFG (pa s iangula is and pa s
o bi alis).
4.3.3. ROI analysis
Fo each o he selec ed ROIs, a 2x2 ANOVA wi h F equency and Task as ac o s, and
pe cen signal change (PSC) as he dependen measu e, was ca ied ou . These analyses
we e ollowed by simple pos -hoc pai wise - es s o planned compa isons, o which Bayes
ac o s (BF) a e epo ed below. Resul s om g oup ROIs a e epo ed i s , ollowed by
indi idual ROIs esul s. Figu e 4.2 shows an o e iew o he esul s ega ding he WFE by
Task, ac oss all g oup ROIs. An addi ional 2x2x2 ANOVA, wi h ROI ype (g oup e sus
indi idual ROI) x F equency x Task was ca ied ou o de e mine any possible e ec s a ising
om he ype o ROIs used (g oup s indi idual). Aside om a signi ican ROI ype x Task
in e ac ion in IFG pa s iangula is (F = 8.078, p = 0.004), we ound no signi ican main o
in e ac i e e ec s o he ac o ROI ype and, he e o e, only esul s om g oup ROIs a e
shown in Figu e 4.2. Table 4.1 shows a summa y o he esul s om he g oup and indi idual
ROI ANOVAs.
Table 4.1. S a is ical esul s om he g oup and indi idual ROI ANOVAs. Main e ec s o
F equency and Task, as well as hei in e ac ion a e epo ed. As e isks deno e a s a is ically
signi ican e ec .
ROI names
G oup ROI
Indi idual ROIs
F equency
Task
In e ac ion
F equency
Task
In e ac ion
O bi alis
F = 10.347
p = 0.002*
𝜂2 = 0.163
F = 7.222
p = 0.009*
𝜂2 = 0.119
F = 5.595
p = 0.021*
𝜂2 = 0.095
F = 7.058
p = 0.010*
𝜂2 = 0.121
F = 5.870
p = 0.018*
𝜂2 = 0.103
F = 3.211
p = 0.079 𝜂2
= 0.059
T iangula is
F = 6.043
p = 0.017*
𝜂2 = 0.102
F = 11.946
p = 0.001*
𝜂2 = 0.183
F = 4.812
p = 0.017*
𝜂2 = 0.083
F = 4.710
p = 0.034*
𝜂2 = 0.083
F = 43.873
p < 0.001*
𝜂2 = 0.457
F = 2.785
p = 0.101
𝜂2 = 0.050
Ope cula is
F = 5.384
p = 0.024*
𝜂2 = 0.092
F = 9.345
p = 0.003*
𝜂2 = 0.149
F = 1.957
p = 0.16
𝜂2 = 0.035
F = 8.035
p = 0.006*
𝜂2 = 0.133
F = 20.109
p < 0.001*
𝜂2 = 0.278
F = 0.835
p = 0.364
𝜂2 = 0.015
IPC
F = 0.596
p = 0.443
𝜂2 = 0.011
F = 4.425
p = 0.040*
𝜂2 = 0.077
F = 0.309
p = 0.580
𝜂2 = 0.005
F = 0.012
p = 0.715
𝜂2 < 0.001
F = 4.022
p = 0.050*
𝜂2 = 0.071
F = 0.134
p = 0.715
𝜂2 = 0.002
MTG/STG
F = 0.367
p = 0.547
𝜂2 = 0.006
F = 2.169
p = 0.146
𝜂2 = 0.039
F = 0.008
p = 0.928
𝜂2 < 0.001
F = 0.077
p = 0.781
𝜂2 = 0.001
F = 0.972
p = 0.328
𝜂2 = 0.018
F = 0.023
p = 0.879
𝜂2 < 0.001
63 OF 150
OTC
F = 0.013
p = 0.909
𝜂2 < 0.001
F = 7.331
p = 0.009*
𝜂2 = 0.121
F = 0.725
p = 0.398
𝜂2 = 0.013
F = 0.060
p = 0.807
𝜂2 = 0.001
F = 9.841
p = 0.002*
𝜂2 = 0.159
F = 1.704
p = 0.197
𝜂2 = 0.031
Ven al Ne wo k
Pa s o bi alis. Following he g oup ROI app oach, a main e ec o F equency was
ound o his egion. We also ound a signi ican main e ec o Task. These main e ec s we e
quali ied by a signi ican Task x F equency in e ac ion. Simple-e ec pos -hoc analyses
e ealed ha his in e ac ion was due o low- equency wo ds showing s onge egional
ac i a ion han high- equency wo ds in he seman ic ( = -3.907, p < 0.001, BF = 91.709), bu
no in he pe cep ual ask ( = -0.687, p = 0.495, BF = 0.186). Resul s om he g oup ROI o
he pa s o bi alis we e eplica ed wi h indi idual ROIs, al hough he Task x F equency
in e ac ion esul ed ma ginally signi ican , possibly due o di e ences in signal in ensi ies
de i ed om bo h app oaches.
Figu e 4.2. ROI analyses. A) G oup ROIs employed, ob ained om he Wo ds-Null con as . B) Resul s om he
g oup ROI analyses. ROIs a e ep esen ed in he X axis, whe eas he di e ence o he pa ame e es ima es
be ween egional ac i a ion o low e sus high equency wo ds is depic ed in he Y axis. Boxes wi h s aigh lines
ep esen he pe cep ual ask, whe eas boxes wi h do ed lines depic he seman ic ask. Red as e isks a he
bo om pa indica e ha he egion showed a signi ican main e ec o F equencyTask (*p < .05, BF > 1; **p < .01,
BF > 5; ***p < .001, BF > 10). Blue as e isks a he bo om pa indica e ha he egion showed a signi ican main
e ec o TaskF equency (*p < .05, BF > 1; **p < .01, BF > 5; ***p < .001, BF > 10). Black as e isks o e he boxes
indica e ha he egion showed a signi ican Task x F equency in e ac ion, he e depic ed as a signi ican di e ence
70 OF 150
4.5. CONCLUSIONS
By applying di e en analy ical app oaches o a la ge da ase o 54 indi iduals, he e
we o e obus e idence ha he ac i i y in he le pa s ope cula is is modula ed by wo d
equency, possibly e lec ing phonological p ocessing du ing lexical sea ch. The ac i a ion o
he pa s ope cula is and IPC a e also modula ed by eading demands, possibly e lec ing
s onge phonological p ocessing du ing seman ic wo d eading. The same eading demand
e ec was obse ed in he OTC, bu in he absence o any e ec o wo d equency, which
seems o suppo he no ion ha his a ea is in luenced by op-down p ocesses. This has
po en ial implica ions o heo ies abou he ole o he OTC in p e-lexical and lexical
p ocesses. Finally, he WFE is modula ed by eading demands in pa s o bi alis and pa s
iangula is, since he WFE was only p esen unde seman ic eading demands. This is
in e p e ed as an indica o o he ole o he an e io IFG in con olled access o seman ics.
These e ec s, occu ing a he egional ac i a ion le el, seem o unde line he ole o he
en al eading ne wo k in he access o lexico-seman ic in o ma ion.
71 OF 150
CHAPTER V. NEURAL REPRESENTATIONS OF LEXICO-SEMANTIC KNOWLEDGE:
SIMILARITY OF SUB-LEXICAL AND LEXICAL MODELS WITH MULTIVARIATE BRAIN
RESPONSES
5.1. RATIONALE
As e lec ed in Chap e 3, in e p e ing he e ec s o psycholinguis ic p ope ies like
wo d conc e eness, amilia i y o equency a he b ain le el can be oublesome, gi en he
complex in e ela ion be ween such p ope ies, and hei in e ac ion wi h expec a ions
imposed by he ask. While we ied o ackle he la e ac o in he p e ious chap e , he s udy
was agnos ic o complex in e ac ions be ween wo d p ope ies, as acknowledged in sec ion
4.4.5. The in es iga ion p esen ed in his chap e was o iginally designed o o e come he
limi a ions o he s udy desc ibed in Chap e 4. I also a emp ed o imp o e ou unde s anding
o neu al ep esen a ions associa ed wi h phonological p ocessing as compa ed o hose
linked o seman ic p ocessing. Finally, he cu en s udy also ied o o e new insigh s in o
he na u e o he b ain ep esen a ions elici ed by NLPs, hus add essing some o hei
concep ual limi a ions as men ioned in sec ion 3.2.
Mo e speci ically, in Chap e 3 I alluded o he di e ences in a ec i e alence be ween
abs ac and conc e e wo ds, as well as hei con ex ual ep esen a ion. I also men ioned how
abs ac wo ds a e, on a e age, lea ned la e in li e and less amilia han conc e e wo ds. The
high collinea i y be ween conc e eness and imageabili y, o be ween equency and amilia i y
has also been poin ed ou . A he neu al le el, he well documen ed neu al e ec s o seman ic
a iables like wo d equency, conc e eness o amilia i y, mos ly come om in es iga ions
ha add essed hei s udy in isola ion (wi h some excep ions men ioned in Chap e 3). Bu in
u h, he in o ma ion abou concep ual neu al ep esen a ions ha hese a iables can p o ide
when s udied oge he , in a con olled ashion, has been a ely explo ed. And mos impo an ly,
he implica ions o uni a ia e designs ha mainly ocus on one o wo o he abo e-men ioned
ac o s, o bo h lexico-seman ic and sublexical p ocessing, a e conside ably limi ed.
In he p esen s udy, ou gene al objec i es we e o disen angle he b ain
ep esen a ions o se e al di e en language p ope ies, om sublexical (i.e., big am and
biphone equency, phonological neighbou s, o hog aphic dis ance) o seman ic (i.e.,
conc e eness, equency, amilia i y), and o in es iga e whe he hese p ope ies a e be e
ep esen ed oge he o in isola ion in b ain a eas c i ical o language p ocessing. Addi ionally,
we had he speci ic objec i e o imp o ing ou unde s anding o he linguis ic p ope ies ha
a e e lec ed in na u alis ic language models (i.e. wo d ec o s). To answe he gene al and
speci ic objec i es, we ca e ully selec ed a la ge pool o wo ds ha he e ogeneously a ied in
sublexical and lexico-seman ic p ope ies, and es ed hem wi h MRI in a simple wo d eading
ask. This allowed us o es ima e se e al di e en models based on di e se combina ions o
72 OF 150
all o he abo e-men ioned language ea u es, and o compa e hem, in mul i a ia e simila i y
analyses, wi h b ain ac i a ion pa e ns elici ed by eading hese wo ds. Compa ing models
ha a e in en ionally “biased” o ep esen di e en combina ions o ei he seman ic,
phonological o bo h kinds o ea u es a he same ime, has he po en ial o e eal how he
b ain egions ha a e key o language p ocessing ep esen lexico-seman ic knowledge. Fo
ins ance, compa ing phonological and seman ic models in he IFG will e eal o wha ex en
he di e en sub egions o he IFG and adjacen a eas pa icipa e in sublexical, lexico-
seman ic o bo h p ocesses. Simila ly, con as ing seman ic e sus phonological models in he
OTC would e eal whe he indeed he an e io OTC is implica ed in seman ic p ocessing o
some deg ee, while i s pos e io subdi ision shows a p e e ence o sublexical in o ma ion.
Speci ically, he e we expec 1) ha he le IFG will show a dissocia ion in which he an e io
IFG ac i a ion is be e p edic ed by seman ic models, whe eas in i s pos e io subdi isions,
bo h seman ic and phonological models will yield compa able co ela ion alues wi h b ain
ac i a ion pa e ns (Sánchez e al., 2023); 2) ha he le OTC will show a dissocia ion om
an e io o pos e io , whe e he an e io bu no he pos e io OTC shows sensi i i y o
seman ically- ela ed models, in line wi h p e ious ecen e idence (Le ma-Usabiaga e al.,
2018); and 3) ha a eas ha ac as “seman ic hubs”, such as he an e io IFG, ATL o he
STG, will show especially good b ain-model simila i y when using models buil om seman ic
ea u es, including wo d ec o s (Binde e al., 2009; Ho man e al., 2015; J. Wang e al.,
2010).
5.2. METHODS
5.2.1. Pa icipan s
A o al o 30 Spanish-speaking, igh -handed pa icipan s (23 emales) aged be ween
19 and 40 yea s old (a e age = 28.5 ± 6.933 yea s) ook pa in he s udy. All pa icipan s
spoke Spanish as hei i s language, had no mal o co ec ed- o-no mal ision, and had no
his o y o epo ed neu ological o psychia ic diso de s. O he ini ial 32 pa icipan s, wo o
hem we e disca ded due o excessi e head mo ion. All pa icipan s ecei ed mone a y
compensa ion o hei olun a y pa icipa ion and ga e hei in o med consen o ake pa in
he s udy, in compliance wi h he egula ions es ablished by he BCBL E hics Commi ee and
he guidelines o he Helsinki Decla a ion.
5.2.2. S imuli and Ma e ials
The s imuli employed included a o al o 960 spanish wo ds. Hal o hese wo ds we e
es ed in he MRI, and he o he hal we e es ed ou side he scanne (see below). All wo ds
we e nouns ex ac ed om EsPal (Duchon e al., 2013), anging om 4 o 10 le e s in leng h,
73 OF 150
and including conc e eness, amilia i y and imageabili y subjec i e a ings. These a ings we e
a ailable in EsPal, and come om no ma i e da a desc ibed elsewhe e (Duchon e al., 2013).
They a e subjec i e a ings abou he wo d, anging om 1 o 7, whe e 1 means comple ely
abs ac (in he case o conc e eness), comple ely no amilia (in he case o amilia i y), o
e e ing o an objec ha is comple ely impossible o imagine (in he case o imageabili y).
Addi ionally, all wo ds coun ed on objec i e measu es o equency o occu ence, big am
equency, biphone equency, numbe o phonological neighbou s, and o hog aphic
Le ensh ein dis ance (OLD20) (see Ya koni e al., 2008). Each subse o 480 wo ds was spli
in o 240 abs ac and 240 conc e e wo ds. Each wo d g oup was also subsequen ly di ided in
4 condi ions: a) 60 wo ds wi h low amilia i y and low equency; b) 60 wo ds wi h low amilia i y
and high equency; c) 60 wo ds wi h high amilia i y and low equency; and d) 60 wo ds wi h
high amilia i y and high equency. The cu o poin s o each o he a iables (conc e eness,
amilia i y and equency) we e based on hei app oxima e median alues. Thus, abs ac
concep s include wo ds wi h a a ing o 4.5/7 o lowe , and conc e e concep s include wo ds
abo e his alue. Likewise, he cu o alue o amilia i y was se o 4.5/7. Fo wo d equency
we used a Zip (B ysbae e al., 2018) scaled alue o 3.5 as a cu o . Al hough disc e e g oups
we e used o gene al con as s and o simpli y he uni a ia e analyses, we coun ed on a
con inuum o all h ee ac o s o in e es , hus allowing us o pe o m RSA and hie a chical
eg ession analyses. The wo subse s o wo ds we e ma ched in all a iables o in e es , as
well as in wo d leng h and phonological neighbou s. Figu e 5.1 shows he dis ibu ion o he
subse o wo ds es ed in he MRI ask along he con inuum o he a iables o in e es . The
beha iou al subse , no shown he e (bu see Figu e 6.1), shows a simila pa e n. A o al o
560 p onounceable non-wo ds, buil om he spanish wo ds by using Wuggy (Keulee s &
B ysbae , 2010), we e also employed. O he o al, 80 non-wo ds we e used in he MRI, and
he emaining 480 non-wo ds we e used in he beha iou al es s. Bo h subse s we e ma ched
in leng h and phonological neighbou s.
74 OF 150
Figu e 5.1. Dis ibu ion o he 480 wo ds used in he MRI ask along he con inuum o conc e eness,
amilia i y and equency (zip ). Each do ep esen s a wo d. The eigh colou s ep esen each o he
disc e e ca ego ies c ea ed based on he median cu o s.
All ma e ials explained abo e we e es ed in wo lexical decision asks, one inside he
scanne and ano he one ou side he scanne . The lexical decision ask allowed us o ha e
pa icipan s eading and p ocessing wo ds, while pe mi ing he egis a ion o eac ion imes
(RTs) and a oiding excessi e o e load due o complex decision making. The wo asks
di e ed in he p opo ion o wo ds and pseudowo ds and in he in e - ial in e als (ITI), bu
we e o he wise iden ical. Since we needed a high numbe o obse a ions, and RSA equi es
a su icien spacing be ween ials (o e 6 seconds), in he lexical decision ask inside he
scanne we used 14 % o non-wo d ials (80 i ems), o he sake o ime. The “p ope ” lexical
decision ask ou side he scanne (wi h 50% o non-wo ds and 50% o wo ds), ac ed as a
con ol o any po en ial ask e ec s de i ed om he di e en ial p opo ion o wo ds e sus
pseudowo ds in he lexical decision ask inside he scanne .
5.2.3. P ocedu e
Fi s ly, high esolu ion T1 images we e acqui ed. Nex , du ing he unc ional MRI BOLD
sequence, pa icipan s pe o med he lexical decision ask: hey we e old o pay a en ion o
75 OF 150
and ead a se ies o le e s ings ha migh o migh no o m eal wo ds in Spanish. They
we e ins uc ed o decide whe he he wo d exis s o no by p essing he co esponding bu on.
Bu on assignmen was coun e balanced o all pa icipan s, and he o de o p esen a ion o
he i ems was unique o e e y wo pa icipan s (i.e. a o al o 15 p ede ined coun e balanced
o de s) o ensu e ha no e ec s o o de in luence pa e n simila i y (Mum o d e al., 2014).
The s imuli we e p esen ed on he cen e o he sc een o 1 second, ollowed by a a iable
ITI o a leas 6 seconds. This allowed he haemodynamic esponse o e u n o baseline,
which allowed us o accu a ely model each ial sepa a ely. The ask was di ided in o 6
iden ical unc ional uns o 11:40 minu es each.
Immedia ely a e he MRI session, pa icipan s pe o med he “p ope ” lexical decision
ask wi h he emaining 480 wo ds and 480 non-wo ds ou side he scanne . The ask was he
same, excep o he p opo ion o non-wo ds, and he ITI, which was sho e (1 second) in he
“p ope ” lexical decision ask (longe ITIs we e only equi ed by he MRI ask). The o de o
he s imuli was also coun e balanced by subjec .
5.2.4. MRI Da a Acquisi ion and P ep ocessing
Whole-b ain Images we e acqui ed on a 3-T SIEMENS’s Magne om P isma- i
scanne , wi h 64-channel head coil, a he Basque Cen e on Cogni ion B ain and Language
(BCBL). Fi s ly, high- esolu ion T1-weigh ed ana omical images we e ob ained wi h he
ollowing acquisi ion pa ame e s: TR = 2530 ms, TE = 2.36 ms, lip angle = 7º, Field o iew =
256 mm, 176 olumes pe un, oxel size = 1 cubic mm. A e ha , 6 unc ional uns we e
acqui ed. Each MRI un consis ed o a mul iband g adien -echo echo-plana imaging
sequence wi h he ollowing pa ame e s: TR = 1000 ms, TE = 35 ms, lip angle = 56º Field o
iew = 210 mm, 690 olumes pe un, oxel size = 2.4 cubic mm, accele a ion ac o = 5. The
i s 6 olumes o each un we e emo ed o ensu e T1-equilib a ion e ec s. The o de o he
ials in each un, as well as he in e - ial in e als o a iable du a ion, we e de e mined wi h
an op imised algo i hm designed o maximise he e iciency o he eco e y o BOLD esponse:
Op seq II (Dale, 1999).
All images we e p ep ocessed by using cus om sc ip s based on AFNI (Cox, 1996).
The T1-weigh ed image was skull-s ipeed and co- egis e ed o he unc ional images by
means o linea a ine ans o ma ions. Al hough slice iming co ec ion was no compulso y
due o simul aneous acquisi ion o mul iple slices wi h mul iband sequence and a sho
epe i ion ime (i.e. 1000 ms), he emaining olumes we e co ec ed o po en ial di e ences
in iming o slice acquisi ion and ealigned o he minimum ou lie olume by means o 12
pa ame e igid-body mo ion ans o ma ion. Fo uni a ia e analyses, all images we e
76 OF 150
no malised o he MNI152 s anda d space (2009Lin) by means o non-linea ans o ma ions,
a a esolu ion o 2 cubic mm. The esul ing images we e smoo hed wi h a 4-mm ull wid h a
hal -maximum (FWHM) iso opic Gaussian ke nel, and inally, scaled o ge a mean oxel
signal o 100. Rega ding mul i a ia e analyses, bo h RSA sea chligh and RSA based on
egions o in e es (ROIs) we e ca ied ou . All mul i a ia e analyses we e pe o med in
indi idual subjec space, on unsmoo hed, unscaled images ob ained p io o he no malisa ion
o he MNI space.
5.2.5. Uni a ia e analyses
S a is ical analyses we e pe o med on indi idual subjec da a using he gene al linea
model (GLM), and hen submi ed o g oup le el analyses. MRI ime se ies da a we e
modelled by a se ies o impulses con ol ed wi h a canonical hemodynamic esponse unc ion
(HRF). The mo ion pa ame e s o ansla ion (i.e., x, y, and z) and o a ion (i.e., yaw, pi ch,
and oll), along wi h a basic se o cosine unc ions ha high-pass il e ed he da a, we e
included as co a ia es o non-in e es in he GLM. Each o he 8 ca ego ical condi ions,
composed o 60 ials, was modelled as an e en , ime-locked o he onse s o he p esen a ion
o each wo d ha belonged o hei co esponding ca ego y. The leas -squa es pa ame e
es ima es o he heigh o he bes - i ing canonical gamma HRF o each ca ego ical condi ion
we e used in pai wise con as s. Whole-b ain main e ec s and in e ac ions o he a iables o
in e es we e explo ed by pe o ming a 2x2x2 wi hin-subjec ANOVA (Conc e eness x
Familia i y x F equency). Simple e ec s we e explo ed by pe o ming pai wise con as s,
compu ed as one-sample - es s on he con as images. Gi en he nes ed na u e o he
ca ego ical condi ions (see S imuli and Ma e ials), he main con as images (Conc e eness,
Familia i y and F equency) we e ob ained by me ging oge he he co esponding ca ego ical
condi ions. Thus, he conc e eness con as was ob ained by pe o ming a one-sample - es
be ween he ac i a ions coming om he me ged 240 abs ac wo ds e sus he ac i a ion o
he me ged 240 conc e e wo ds. Likewise, he amilia i y con as was ob ained by pe o ming
a one-sample - es be ween he ac i a ions o he me ged 240 low- amilia i y wo ds e sus
he ac i a ion o he me ged high- amilia i y wo ds. And inally, he equency con as was
ob ained om compa ing he me ged 240 low- equency wo ds agains he me ged 240 high-
equency wo ds. A e applying a alse disco e y a e (FDR) clus e -le el co ec ion o p <
0.01 o he unco ec ed oxel-ex en h eshold o p < 0.001, a clus e was deemed o be
signi ican ly ac i e i i exceeded 14 oxels.
5.2.6. RSA sea chligh
In o de o explo e he whole-b ain ep esen a ions o he di e en lexical ea u es ha
a e objec s o in e es in he p esen s udy, a sea chligh -based RSA was ca ied ou . Fo his,
77 OF 150
we i s ly de ined 6 di e en models based on he dissimila i y be ween each pai o wo ds in
hei key ea u es: a) wo d conc e eness; b) wo d amilia i y; c) wo d equency (zip ); d)
seman ic ea u es ( he combina ion o a, b and c, plus imageabili y; e) phonological ea u es
(a combina ion o big am equency, biphone equency, numbe o phonological neighbou s,
numbe o le e s, and ; and ) wo d ec o s (Wo d2Vec) eco e ed om se e al di e en
Spanish sou ces (Almeida & Bilbao, 2018; Bilbao-Jayo & Almeida, 2018). To ensu e ha he
models we e exp essed in as compa able measu es as possible, we employed euclidean
dis ances in he case o unique a iables (conc e eness, amilia i y and equency),
mahalanobis dis ances in he case o composi e phonological and seman ic models (gi en
hei po en ial co a iabili y) and cosine dis ances in he case o wo d ec o s (gi en he
po en ial in luence o he size o he es ima ed ec o ). All measu es we e hen no malised o
p oduce a ange be ween 0 ( he i ems a e he same) and 1 ( he i ems a e comple ely di e en ).
Figu e 5.2 shows a g aphical ep esen a ion o he models ( ep esen a ional dissimila i y
ma ices, RDMs), along wi h a co ela ion plo displaying he simila i y be ween each pai o
models. O no e, he models ep esen he wo ds o de ed by conc e eness o isualisa ion
and consis ency pu poses.
Figu e 5.2. G aphical ep esen a ion o he models employed and pai wise co ela ion be ween hem.
A) Rep esen a ion ma ix o he “simple” models. F om op o bo om: Conc e eness, Familia i y,
F equency. Each ow/column ep esen s a di e en wo d ( ial). B) Rep esen a ion ma ix o he
“combina o ial” models. F om op o bo om: Seman ic, Phonological, Wo d2Vec. C) Pai wise
co ela ion ma ix be ween he models. O no e, A and B display dissimila i ies ma ices, wi h 1 meaning
absolu e dissimila i y, and 0 meaning absolu e simila i y. In u n, C displays co ela ion, and he e o e,
1 means pe ec co ela ion, while 0 means absolu e absence o co ela ion.
78 OF 150
A e he p ep ocessing o he images, be a alues o each ial (each wo d) we e
es ima ed wi h he GLM. A o al o 36 pa ame e s co esponded o polynomial e ms ha
es ima ed changes in signal due o d i (6 e ms pe un). Ano he 6 eg esso s we e mo ion
pa ame e s (3 o a ional, 3 ansla ional). Each wo d was es ima ed as a sepa a e eg esso
by con ol ing he onse o he s imulus wi h a canonical gamma HRF. The esul ing images
con aining he be a alues we e hen masked o include b ain oxels only.
The sea chligh i sel was pe o med wi h cus om sc ip s based on Py hon
(h ps://gi hub.com/Ab ahamSV/simplyRSA). All possible 5 mm sphe es con aining a leas
50% o oxels alling wi hin he b ain mask we e p ede ined o inc ease compu ing e iciency.
Fo each sphe e, he b ain simila i y ma ix was compu ed as he cosine dis ance be ween
each pai o ial-speci ic ec o ised pa e ns. The b ain simila i y ma ix was hen compa ed
o each o he 6 desc ibed models by means o anks co ela ion (Spea man’s ho). The esul
o each sphe e (i.e. each cen e o mass coo dina es) was hen ead back in o he b ain mask,
and no malised o he MNI152 empla e o a e aging, esul ing in he whole-b ain
ep esen a ion o he simila i y wi h each model. Since we we e in e es ed in in es iga ing he
whole-b ain di e ences be ween phonological, seman ic, and na u alis ic models (Wo d2Vec),
and gi en he collinea i y be ween he h ee a iables o in e es wi h he seman ic model (see
Figu e 5.2), only he h ee combina o ial models we e analysed in he sea chligh . The
con ibu ion o he h ee a iables o in e es was explo ed in ROI-based analyses.
5.2.7. ROI-based RSA
In o de o add ess he main objec i es and o es he p oposed hypo heses, we
conduc ed RSA in b ain a eas ha a e key o concep ual ep esen a ions associa ed wi h he
a iables o in e es , acco ding o he p e ious li e a u e men ioned abo e (Binde e al., 2009;
Ho man e al., 2015; J. Wang e al., 2010). A o al o 7 le -la e alised ROIs we e ana omically
de ined by using he HCP a las a ailable in AFNI (Glasse e al 2016): IFG pa s o bi alis, IFG
pa s iangula is, IFG pa s ope cula is, ATL, pos e io STG, an e io OTC ( usi o m FG4) and
pos e io OTC ( usi o m FG2). As in he sea chligh RSA, he b ain simila i y ma ix was
ob ained o each ROI by compu ing he cosine simila i y be ween each pai o ial-speci ic
ec o ised pa e ns. The b ain simila i y ma ix was compa ed wi h each o he 6 models, and
submi ed o a boo s ap analysis o 100.000 i e a ions, which yielded he employed 95%CI.
Each i e a ion was compu ed as he Spea man ank’s co ela ion be ween he b ain RDM and
a andom pe mu a ion o he Wo d2Vec model. The esul ing a e aged co ela ion be ween
each o he ROIs and each model was hen compa ed o he 95 h pe cen ile o he boo s ap
sample, by means o Pea son and Filon’s Z, adap ed om he coco R package (Diedenho en
79 OF 150
& Musch, 2015). Al hough we we e mainly in e es ed in analysing he di e ences be ween he
combina o ial models, in he ROI analyses we included he simple models in o de o
in es iga e he con ibu ion o each o he a iables o in e es o he seman ic model.
5.3. RESULTS
5.3.1. Beha iou al Pe o mance
On a e age, he global accu acy o all i ems in he MRI ask was 95.031% ± 4.229,
and he a e age RT o all i ems was 0.677 sec ± 0.084. In o de o e alua e he po en ial
ela ions be ween he h ee a iables o in e es and RTs, we pe o med Pea son’s co ela ion
es s o e all co ec esponses by each pa icipan . Con idence in e als (CIs) o he alues
a e epo ed. Ac oss he 30 subjec s, he a e age co ela ion be ween Conc e eness and RT
was = -0.052 ± 0.045, wi h a 95%CI = [-0.069, -0.036], and an a e aged p = 0.309 ± 0.255 .
The a e aged co ela ion be ween Familia i y and RT was = -0.284 ± 0.083, wi h a 95% CI
= [-0.315, -0.253], and an a e aged p = 0.001 ± 0.005. Finally, he a e age co ela ion be ween
F equency and RT was = -0.246 ± 0.059, wi h a 95%CI = [-0.269, -0.224], and an a e aged
p = 0.001 ± 0.007.
We also assessed he di e ences in he lexical e ec be ween all he 8 ca ego ical
bins desc ibed abo e, by calcula ing T alues and e ec sizes o he con as non-wo d’s RT
minus each bin’s RT. Non-wo ds end o show slowe RTs han wo ds o e all. A signi ican
di e ence be ween non-wo ds and any o he 8 bins would mean ha wo ds in such bin a e
easie o access han non-wo ds (i.e., lexical e ec ). In his sense, he highe he e ec size
associa ed wi h he con as , he g ea e he lexical e ec . In u n, i he con as does no show
a signi ican di e ence, o i he e ec size associa ed is conside ably low, i can be conside ed
ha he wo ds in he bin a e as easy o access as non-wo ds (i.e., no lexical e ec ). Below,
we epo a e aged alse disco e y a e (FDR)-co ec ed p alues (i.e. q- alues) and a e aged
Cohen’s d alues. The esul s demons a ed ha all bins showed a lexical e ec . Howe e ,
he bins ha combined highly amilia and highly equen wo ds showed he g ea es e ec
sizes, wi h sligh di e ences be ween abs ac and conc e e wo ds. In con as , bins ha
con ained lowe - amilia i y and lowe - equency wo ds showed mode a e o high e ec sizes,
hus indica ing a less p onounced lexical e ec o wo ds in hese bins (see Table 5.1 below).
Table 5.1. Lexical e ec s esul ing om con as ing each o he ca ego ical bins de ined agains
nonwo ds.
86 OF 150
Figu e 5.5. Resul s om he boo s ap analysis in he seman ic hub ROIs o each o he 6 models. Rho
alues a e shown in he X axis. FDR-co ec ed q alues a e epo ed o signi ican abo e-pe cen ile
co ela ions. Single as e isks ep esen co ela ions signi ican ly abo e he 95h pe cen ile o he boo s ap
no su i ing he FDR co ec ion. Double as e isks ep esen signi ican abo e-pe cen ile co ela ions
su i ing he FDR co ec ion.
The IFG pa s o bi alis ROI did no show signi ican abo e-pe cen ile co ela ions wi h
any o he combina o ial models. When looking a he simple models, only he F equency
model showed a signi ican abo e-pe cen ile co ela ion, which did no su i e he FDR
co ec ion (Z = 1.663, q = 0.336).
The IFG pa s iangula is ROI displayed a signi ican abo e-pe cen ile co ela ion wi h
he Seman ic model only, bu i ailed o each signi icance a e FDR co ec ion (Z = 2.369, q
= 0.093). Rega ding he simple models, we ound a signi ican co ela ion wi h F equency (Z
= 3.789, q = 0.001).
In he pos e io IFG pa s ope cula is, he co ela ions we e signi ican ly abo e he ixed
pe cen ile o he Seman ic model, al hough his co ela ion did no su i e he FDR co ec ion
87 OF 150
(Z = 2.052 , q = 0.168). Howe e , his ROI showed a signi ican co ela ion su i ing he FDR
co ec ions wi h he F equency model (Z = 5.264, q < 0.001)
The ATL ROI showed no signi ican co ela ions wi h any o he combina o ial models,
o any o he simple models. Simila ly, he STG ROI did no show any signi ican co ela ion
wi h any o he combina o ial o simple models.
Figu e 5.6 shows he main con as s agains he 95 h pe cen ile o he boo s ap
dis ibu ion o each o he OTC ROIs (an e io FG4, pos e io FG2). While he pos e io OTC
did no show any signi ican co ela ion wi h any o he combina o ial o simple models, he
an e io OTC displayed a signi ican co ela ion wi h he F equency model ha became only
ma ginally signi ican a e FDR co ec ions (Z = 2.654, q = 0.055 ).
Figu e 5.6. Resul s om he boo s ap analysis in he OTC ROIs o each o he 6
models. As in Figu e 5.5, X axis ep esen s Rho alues. Single as e isks ep esen
co ela ions signi ican ly abo e he 95h pe cen ile ha did no su i e he FDR
co ec ion.
88 OF 150
5.4. DISCUSSION
In he p esen s udy, ou goal was o in es iga e he neu al ep esen a ional space o
se e al linguis ic p ope ies ha a e known o be associa ed wi h he access o lexico-seman ic
knowledge. We u he sough o disen angle he in e ela ions be ween hese a iables, along
wi h na u alis ic language models. The main indings a e discussed below.
5.4.1. Associa ions Be ween Wo d P ope ies
One o ou gene al objec i es was o explo e how se e al psycholinguis ic a iables,
om phonology o seman ics, a e ela ed in e ms o b ain ac i a ion. The e is collinea i y
be ween wo d conc e eness and a iables like a ec i e alence (Mee smans e al., 2020;
Vigliocco e al., 2014), equency o occu ence o subjec i e amilia i y (S iem-Ami e al.,
2018). And ye , neu oimaging s udies e y o en in es iga e hese a iables in isola ion, while
igno ing he mos common collinea a iables. He e, we ound ha some o he commonly
obse ed neu al e ec s o wo d conc e eness can be quali ied by ele an in e ac ions wi h
wo d amilia i y. Ou esul s show ha his is he case o he igh pa ahippocampal gy us,
ex ending o he hippocampus, whe e highly amilia wo ds showed g ea e ac i a ion han
less amilia wo ds, only o wo ds on he lowe end o he conc e eness con inuum. Being he
i s ga e o he medial empo al lobe om he isual inpu s, he pa ahippocampal gy us is
known o be in ol ed in lea ning and seman ic memo y, and speci ically, in he binding o i ems
o hei con ex (C. B. Ma in & Ba ense, 2023). Some s udies ha e ound highe esponses in
he pa ahippocampal gy us o mo e amilia isual objec s (C. B. Ma in e al., 2013), and o
conc e e e sus abs ac wo ds (Binde e al., 2009; Ho man e al., 2015). Ou da a eplica ed
hese esul s, and ex ended hem, by showing ha high amilia i y e okes g ea e
pa ahippocampal esponses also in he linguis ic domain. This e ec o wo d amilia i y was
especially e iden in abs ac wo ds. A po en ial in e p e a ion o his in e ac ion is ela ed wi h
he abo e-men ioned ole o he pa ahippocampal gy us in con ex ual binding. In his sense,
he e ie al o amilia in o ma ion could be mo e e o ul o abs ac wo ds, gi en ha hey
a e known o appea in a wide a ie y o con ex s, which makes hem mo e ambiguous and
ha de o access (Ho man e al., 2013; Schwanen lugel e al., 1988)
In addi ion, we ound a signi ican in e ac ion be ween wo d equency and amilia i y.
This is pe haps less su p ising, gi en he known collinea i y be ween hese wo a iables
(Tanaka-Ishii & Te ada, 2011). Ne e heless, o ou knowledge, his is he i s s udy o u he
explo e he in e ac ion o hese wo a iables a he neu al le el, wi h po en ial concep ual
implica ions. The ques ion aises hen as o whe he hese wo p ope ies ap in o he same
cogni i e p ocess. Ou da a sugges ha , albei o a high ex en collinea , he e ec s o
amilia i y and equency add up o p oduce s onge lexical e ec s. Ou beha iou al da a show
89 OF 150
ha wo ds became easies o access when hey a e associa ed wi h bo h highe equency o
occu ence in media (objec i e equency), and wi h highe a es o subjec i e amilia i y
(subjec i e equency). Aligning o his esul , ou MRI da a show ha equency and amilia i y
add up o p oduce s onge e ec s in a eas o he en al language ne wo k (i.e. IFG pa s
o bi alis and pa s iangula is), also including he le pa ahippocampal gy us. While he e ec s
o equency ha e been associa ed o bo h phonological access (Ca ei as e al., 2009;
Fiebach e al., 2002), and seman ic access (Chee e al., 2002, 2003) du ing wo d p ocessing,
he cogni i e e ec s o amilia i y ha e been almos exclusi ely associa ed o he ease o
access o seman ic knowledge (Ne eu & Kaushanskaya, 2023; Shinozuka e al., 2021). Thus,
gi en he ole o he en al eading ne wo k in ac i e e ie al o lexico-seman ic in o ma ion
(wo d meaning), i seems ha he added e ec s o equency and amilia i y a e mainly linked
o seman ic access, wi hou con es ing he idea ha wo d equency is also associa ed wi h
phonological access.
5.4.2. An e io - o-Pos e io dissocia ion in he le IFG
In line wi h p e ious indings by ou g oup (Sánchez e al., 2023), we expec ed o ind
e idence o an an e io - o-pos e io unc ional dissocia ion in he le IFG, whe e i s an e io
subdi isions would show highe simila i y wi h seman ically- ela ed models, whe eas he
pos e io po ions o he IFG would show equi alen simila i ies wi h bo h seman ic and
phonological models. Ou esul s pa ially suppo his no ion. F om he RSA sea chligh , we
could demons a e ha a seman ic model buil om psycholinguis ic p ope ies showed abo e
h eshold simila i ies in he whole le IFG, wi h he highes co ela ions obse ed in he pa s
iangula is. Simila ly, a na u alis ic language model, belie ed o ep esen ine-g ained
seman ic ela ions, showed abo e h eshold simila i ies in he bila e al pa s o bi alis only.
Howe e , while he pos e io IFG (pa s ope cula is) displayed abo e- h eshold co ela ions
wi h he seman ic model, we did no obse e signi ican co ela ions wi h a phonological model
in his a ea. In u n he phonological model showed signi ican co ela ions in mo e pos e io
a eas, in SMA (BA 6), po en ially unde lining i s ole in mo o planning and phonological sea ch
(Ca ei as e al., 2006, 2009). A possibili y hen is ha he di ision o labou be ween an e io
and pos e io IFG expands beyond his a ea. Unde his in e p e a ion he an e io IFG is
s ongly associa ed wi h seman ic lexical access (al hough no exclusi ely), as illus a ed by
i s high simila i y wi h bo h a psycholinguis ic seman ic model, and a na u alis ic language
model. As we mo e o he pos e io IFG, we s ill obse e, o a ce ain deg ee, in ol emen in
seman ic p ocessing, al hough his in ol emen becomes less p onounced. And going u he ,
he a eas immedia ely pos e io /do sal o he IFG seem o be almos exclusi ely in ol ed in
mo o planning associa ed wi h phonological p ocessing (Hagoo , 2005, 2013).
90 OF 150
This iew could in pa explain he esul s om he ROI-based RSA. We ound ha
only he wo d equency model showed signi ican co ela ions wi h he b ain ac i a ion pa e n
in he pa s iangula is, and especially, in pa s ope cula is. Gi en he abo e-men ioned hyb id
na u e o wo d equency (pu a i ely con eying bo h seman ic p ocessing and phonological
sea ch), he ac ha he pa s ope cula is is he a ea whe e he equency model showed he
highes signi ican co ela ions, illus a e he dynamic ole o his a ea in bo h phonological and
seman ic p ocesses. Ne e heless, he ROI-based analyses ailed o eplica e some o he
indings om he sea chligh (i.e., we did no ind a signi ican co ela ion wi h seman ic models
in he an e io IFG). Two ac o s could accoun o his inconsis ency. Fi s ly, he boo s ap
analysis applied in he ROIs is conside ably mo e s ingen han he whole-b ain sea chligh ,
by es ablishing a highe h eshold o a con as o be conside ed signi ican , a e he due
co ec ions. And secondly, al hough we belie e ha he employed lexical decision ask is
op imal o he objec i es o he s udy (as explained abo e), i migh ha e no maximised
seman ic p ocessing, o which op-down modula ions ha e been demons a ed o be
especially ele an (Sánchez e al., 2023). Fu u e s udies should in es iga e he ela ion
be ween mul i a ia e ep esen a ions o psycholinguis ic p ope ies in he IFG and a ying ask
demands. Bu no wi hs anding hese limi a ions, ou esul s pinpoin he unc ional dynamics
o he IFG and associa ed a eas, and hei ole in he di e en cogni i e p ocesses ha
acili a e lexical access.
5.4.3. In ol emen o he le an e io OTC in lexico-seman ic p ocessing
We expec ed o ind a unc ional dissocia ion be ween he le an e io and pos e io
OTC, whe e i s an e io , bu no pos e io , subdi ision would show sensi i i y o seman ic
models. Ou esul s co obo a ed his hypo hesis. We ound signi ican co ela ions wi h he
seman ic model a ound he an e io OTC. Al hough wi h a smalle clus e size, we also ound
a co ela ion wi h he Wo d2Vec model in his a ea. Fu he mo e, we ound a nega i e
co ela ion wi h he phonological model in his an e io OTC a ea. This could be indica ing
ha phonological p ocessing migh no be as ele an in he an e io OTC. In u n, he
phonological model showed high co ela ions wi h p ima y and seconda y isual a eas,
including V3 and V4. As in he IFG, a po en ial in e p e a ion is ha he expec ed g adien
could ex end pos e io ly, whe e seconda y isual a eas would be sensi i e o lowe -le el
ea u es o isual linguis ic inpu . A eas V3 and V4 a e known o hei ole in ea ly isual
p ocessing o low-le el isual ea u es (Fu lan & Smi h, 2016; Pasupa hy e al., 2020). The
a ea V3 has been associa ed wi h he pe cep ion o mo ion ela ed wi h speech ecogni ion
(Jeschke e al., 2023), and A ea V4 has o en been conside ed he ea lies s ep o he
ca ego isa ion o isual inpu (Okazawa e al., 2016; Pasupa hy e al., 2020). Addi ionally,
91 OF 150
adjacen a eas ha e been associa ed wi h spa ial and ac ion- ela ed p ocessing, gi en hei
co-ac i a ion wi h he pa ie al and, p ecisely, wi h he SMA (Maliko ic e al., 2016).
Pe haps he mos compelling piece o e idence suppo ing his iew is ha he an e io
OTC showed signi ican (albei ma ginal) co ela ions wi h wo d equency, a seman ically
ela ed a iable, while he pos e io OTC did no . This is in line wi h p e ious ecen e idence
indica ing a dissocia ion be ween he an e io and pos e io OTC (Le ma-Usabiaga e al.,
2018; P ice & De lin, 2011; Whi e e al., 2019). Acco dingly, we would ha e expec ed ha , in
u n, he pos e io OTC showed ce ain sensi i i y o phonological models. Howe e , as i
happened wi h he IFG, we did no ind a signi ican co ela ion wi h he phonological model in
he mo e pos e io OTC subdi ision, and hus he esul s should be aken ca e ully.
5.4.4. Linguis ic P ope ies s Wo d Vec o s in Seman ic Hubs
Ou las speci ic objec i e was o imp o e ou unde s anding o he neu al e ec s
cap u ed by na u alis ic language models, and hei ela ion wi h psycholinguis ic p ope ies.
In his sense, i is ypically assumed ha hese models ep esen , o a high ex en , seman ic
ela ions be ween wo ds (Abna e al., 2018; G a e e al., 2019; Liuzzi e al., 2023). In ou
s udy, his was co obo a ed by he signi ican co ela ion be ween a seman ic model buil om
psycholinguis ic a iables and a wo d ec o model.
A he neu al le el, we hypo hesised ha b ain a eas ha a e highly in ol ed in abs ac
p ocessing and e ie al o concep ual in o ma ion (some imes e e ed o as seman ic hubs)
(Pa e son e al., 2007), would show especially good b ain simila i y wi h he seman ic and he
Wo d2Vec models. This hypo hesis was con i med by he RSA sea chligh . We ound
signi ican co ela ions in a eas like he an e io IFG, he STG, o he ATL, wi h some
di e ences be ween he seman ic model and he Wo d2Vec model. The co ela ions wi h he
seman ic model we e mo e ex ensi e, especially in he IFG and he STG, while no being
ound in he ATL. In u n he Wo d2Vec model showed less widesp ead simila i ies in he le
an e io IFG, in e io ATL and pa ahippocampal gy us. This could be indica ing ha he
Wo d2Vec model migh be mo e speci ic o seman ically ela ed p ocesses (see Ca o a e al.,
2017, 2021) han he seman ic model buil om psycholinguis ic a iables. Fo example, he
STG, an a ea associa ed bo h wi h phonological and seman ic p ocessing (F iede ici, 2012;
Lau e al., 2008; MacG ego e al., 2012), showed high simila i ies wi h bo h he phonological
and he seman ic model, bu no wi h he Wo d2Vec model. In addi ion, co ela ions wi h he
le in e io ATL we e only ound wi h he Wo d2Vec model. I should be poin ed ou ha he
phonological model yielded signi ican co ela ions in he supe io ATL. A ela i ely ecen
e iew (see sec ion 2.2.3) poin ed ou ha he ATL displays a g aded unc ional specialisa ion,
and hus in o ma ion o di e se na u e is ep esen ed along he ATL (Ralph e al., 2017). In
92 OF 150
his sense, he supe io ATL is connec ed wi h he audi o y co ex and hence i is no su p ising
ha phonological ep esen a ions ec ui his a ea (Zhang e al., 2024). In con as , he
simila i ies wi h he Wo d2Vec model we e ound in he in e io ATL (in e io empo al/an e io
FFG), an a ea ha migh con ey in o ma ion o di e se na u e (Ralph e al., 2017).
Al hough hese esul s seem p omising in he a emp o disen angle he neu al e ec s
associa ed wi h na u alis ic language models, i should be no ed ha he ROI analyses ailed
o eplica e he indings. Once mo e, we can e e o he wo ac o s explained abo e o
accoun o his. Fi s ly, because we used a lexical decision ask, seman ic op-down
in luences a e no maximised. And secondly, and adding o his, he ROI con as s we e mo e
s ingen , and hus, a sub le seman ic e ec a ising om a lexical decision ask, migh ha e
been o e looked when ollowing ROI analyses.
5.5. CONCLUSIONS
The p esen s udy o e s new insigh s in o he b ain ep esen a ions o lexico-seman ic
in o ma ion. We ound ha some o he mos commonly epo ed seman ic e ec s, like he le
pa ahippocampal esponses o conc e eness, migh be quali ied by amilia i y e ec s.
Fu he mo e, we ound conside able suppo o he iew ha he le IFG and adjacen
pos e io do sal a eas (BA 6, SMA) show a unc ional dissocia ion, whe e an e io a eas o he
IFG a e especially ele an o he access o lexico-seman ic knowledge, while he pos e io
IFG pa icipa es o a lesse ex en in seman ic p ocesses, and he SMA is in ol ed in
accessing and ep esen ing phonological lexical in o ma ion. A somewha di e en pa e n
was ound in he OTC, ex ending o he adjacen pos e io a eas o he p ima y and seconda y
isual co ex: while he an e io OTC exhibi ed sensi i i y o seman ically- ela ed p ope ies
du ing wo d eading, he pos e io OTC did no , and pos e io adjacen a eas V3 and V4
showed high simila i y wi h a model ep esen ing phonological p ope ies. Finally, we ound
ha bo h na u alis ic language models and seman ic psycholinguis ic models yield signi ican
simila i ies in b ain ne wo ks ha hold seman ic ep esen a ions. Bu while he seman ic
psycholinguis ic model showed a mo e widesp ead pa e n, including he whole IFG and STG,
he wo d ec o s model displayed mo e speci ic ac i a ions in seman ic hubs. Ou esul s
pinpoin he speci ic cogni i e p ocesses, om phonological o seman ic, in ol ed in he
access o wo ds in b ain ne wo ks ha a e c ucial o eading and seman ic memo y.
93 OF 150
CHAPTER VI. BEHAVIOURAL CORRELATES OF LEXICO-SEMANTIC
REPRESENTATIONS
6.1. RATIONALE
As i has been desc ibed in p e ious chap e s, he neu al e ec s o psycholinguis ic
a iables obse ed in mul iple MRI s udies, in di e en b ain a eas, may in pa depend on
op-down in luences imposed by he demands o he ask. In many ins ances, like in he s udy
desc ibed in Chap e 4 (also see Mee smans e al., 2022; So o e al., 2020), he a en ional
equi emen s o he ask can be in en ionally manipula ed o explo e he di e en cogni i e
se ings unde which he condi ions s udied elici a gi en b ain esponse. Howe e , in o he
ci cums ances, mos o hese ac o s may play a signi ican ole, wi hou any explici con ol
o e hem, ei he because hey a e unknown o us, o because i is cos ul o include such
manipula ions and/o con ols (as was he la e case o he s udy p esen ed in Chap e 5, due
o he ime cons ain s).
I we ocus only on he beha iou al aspec s o a cogni i e ask, decision making i sel
is in luenced by a conside ably ex ensi e a ie y o ac o s. Some o hem o e li le
in o ma ion abou he cogni i e p ocesses o in e es , and a e una oidable, like noise o pu ely
mo o esponse planning; o he s migh be a iables o in e es , and can be ha nessed o be e
unde s and a gi en beha iou , like indi idual di e ences and he manipula ions ha al e he
cogni i e p ocesses induced by he ask i sel . Whils mos s a is ical app oaches ha e he
po en ial o assess he in luence o known and con olled a iables o no in e es , i is usually
ha d o ease apa hose unknown noise ac o s ha in luence a gi en decision. This has been
he s a ing poin o , and main eason o use an app oach designed o be e accoun o
decision making p ocesses, he d i di usion model (DDM) (Ra cli , 1978; Ra cli e al., 2016;
Ra cli & McKoon, 2008). This g oup o compu a ional models y o in e la en cogni i e
p ocesses om he obse able beha iou (usually ep esen ed by ial- o- ial RTs) du ing
decision making in cogni i e asks (Mye s e al., 2022). Essen ially, he DDM simula es he
ime i akes o each a decision (o a bounda y), in one way o he o he , when a choice needs
o be made. In i s mos common applica ion, ou pa ame e s a e de ined ha ep esen
di e en aspec s o he e idence accumula ion p ocess in o de o make a decision: 1) he
bounda y sepa a ion (a) indica es he dis ance om one bounda y ( esponse A) o he o he
( esponse B), wi h g ea e bounda y sepa a ion indica ing a highe necessa y amoun o
e idence equi ed o each a decision; 2) he s a ing poin (z) ep esen s he ini ial s a e in he
decision making p ocess, and can be close o ei he bounda y i he condi ions a ou one o
o he esponse; 3) he d i , which gi es name o he app oach, ep esen s he speed a which
e idence is accumula ed o each a bounda y, wi h la ge posi i e alues indica ing as e
e idence accumula ion owa ds he uppe bounda y ( esponse A), and la ge nega i e alues
94 OF 150
indica ing as e e idence accumula ion owa ds he lowe bounda y ( esponse B); and 4) he
non-decision ime ep esen s he ime i akes o he ne ous sys em o gi e a mo o esponse,
once he decision bounda y is eached (Mye s e al., 2022; Shinn e al., 2020). Impo an ly, all
hese pa ame e s y o build a simula ion ha bes i s he RTs obse ed in a gi en ask in a
ial- o- ial manne . I is common o de ine di e en d i s ha ep esen he di e en abili y o
he condi ions in he s udy o a ec decision ime (Mye s e al., 2022). O he mo e
sophis ica ed models include dynamically changing pa ame e s in o de o imp o e he
p edic ions, and hus he i o he model (Shinn e al., 2020).
We can ha ness his ype o compu a ional modelling o p edic he a iables ha
con ibu e he mos o he decision making p ocess. Addi ionally, by cus omising he model
pa ame e s, we can explo e how complex he p ocess is, and o wha ex en RTs a e “in la ed”
by his complexi y in he decision making (e.g., Muelle e al., 2017). This can be especially
ele an when i comes o accoun ing o he b ain ac i a ion pa e ns ha a e ob ained unde
ce ain cogni i e manipula ions, and o es ima ing how in luenced hey a e by op-down
in luences no di ec ly ela ed wi h he cogni i e p ocess s udied. Mo eo e , DDMs o e he
possibili y o compa e he co ela ions be ween he d i s yielded by he models and b ain
ac i i y pa e ns. This can e eal whe he he abili y o any se o a iables o in luence he
decision making p ocess is indeed associa ed wi h he pa e ns o b ain ac i a ion obse ed
in esponse o hose a iables. This is why I decided o employ DDMs o u he analyse he
pe o mance in he lexical decision asks desc ibed in Chap e 5, bo h inside and ou side he
scanne ( e e ed o as MRI and Beha iou al espec i ely om his poin on). Since
manipula ing he ask (as in Chap e 4) was no possible in he s udy desc ibed in Chap e 5,
pa o he cogni i e p ocesses associa ed wi h he neu al ep esen a ions obse ed could no
be di ec ly accoun ed o . Speci ically, as desc ibed abo e, a limi a ion o he p e ious s udy
was ha i was unclea whe he he lexical decision ask employed could be limi ing he ex en
o seman ic p ocessing, o on he con a y, boos ing sublexical p ocessing by making
pa icipan s ead he wo ds in a supe icial and au oma ed way. DDMs o e a di ec way o
compa ing he abili y o di e en combina ions o pa ame e s o be e p edic he obse ed
RTs. I also has he po en ial o accoun o non-decision ime in a ial-wise manne . This
could also allow us o examine he co ela ion be ween he “pu e” lexical decision ime (no
in luenced by e ec s o no in e es in he esponses gi en), and he neu al e ec s obse ed
in Chap e 5.
The gene al objec i e o his chap e was o examine he deg ee o which sublexical
(big am equency, numbe o le e s, and o hog aphic dis ance) and/o lexico-seman ic (wo d
equency, amilia i y, and conc e eness) p ope ies, o any combina ion o hese a iables,
a e capable o in luencing he decision making p ocess in lexical decision asks. Addi ionally,
his chap e has he speci ic objec i e o analysing he link be ween ine-g ained beha iou al
95 OF 150
measu es and he neu al e ec s obse ed. In pa icula , we sough o examine he associa ion
be ween he model-b ain simila i ies om Chap e 5 on he one hand, and he abili y o
di e en seman ic and sublexical psycholinguis ic a iables o in luence decision making on
he o he . To gi e an answe o he gene al objec i e, we will compa e he i o DDMs
ep esen ing di e en combina ions o he men ioned a iables o in e es . The speci ic
objec i e will be add essed by ob aining he bes - i ing d i alues o he seman ic and
sublexical DDMs in o de o es ima e hei associa ion wi h b ain simila i ies ob ained om he
RSA analysis, which embody he neu al ep esen a ions associa ed wi h seman ic o
sublexical p ocesses desc ibed in he p e ious chap e . We expec ha a DDM buil om
lexico-seman ic p ope ies (wo d equency, amilia i y and conc e eness) will be he bes DDM
a p edic ing RTs a e accoun ing o model complexi y, and ha wo d equency will be he
mos signi ican con ibu o o he d i a e. This esul would indica e ha , e en in a lexical
decision ask, lexical p ocessing o seman ic p ope ies eme ges, as ou esul s om p e ious
chap e s seem o indica e. In addi ion, we hypo hesised ha , i he neu al e ec s obse ed in
Chap e 5 a e explained by lexical decision making, hen we should expec signi ican
co ela ions be ween he d i alues om he DDMs and b ain-model simila i ies. We expec
he highes co ela ions be ween he bes - i ing seman ic d i s and he neu al ep esen a ions
in hose a eas ha showed signi ican simila i ies wi h he seman ic and wo d ec o models
(an e io IFG, an e io OTC and ATL). A he same ime, i o e all RTs a e associa ed wi h
RSA model simila i ies, we would expec equally high co ela ions be ween model simila i ies
and o e all RTs in hose a eas ha a e mo e hea ily in luenced by complex decision-making,
like he pos e io IFG, and he OTC. In con as , i RSA simila i ies a e unbiased by aw RTs,
hen we should no expec any signi ican co ela ions be ween his a iable and simila i y
alues.
6.2. METHODS
6.2.1. Pa icipan s
The pa icipan s analysed he e we e he same 30 desc ibed in he p e ious chap e
(see sec ion 5.2.1). All pa icipan s spoke Spanish as hei i s language and had no epo ed
his o y o neu ological o psychia ic illnesses. All o hem ecei ed mone a y compensa ion
o hei olun a y pa icipa ion and ga e hei in o med consen o ake pa in he s udy, in
compliance wi h he egula ions es ablished by he BCBL E hics Commi ee and he guidelines
o he Helsinki Decla a ion.
102 OF 150
Figu e 6.3. A) Dis ibu ion o RTs (in seconds) ac oss subjec s in esponse o wo ds in he Beha iou al
(le ) and MRI ( igh ) asks o co ec (in blue) and inco ec (in ed) esponses. B) Dis ibu ion o RTs (in
seconds) ac oss subjec s in esponse o nonwo ds in he Beha iou al (le ) and MRI ( igh ) asks o co ec
(in blue) and inco ec (in ed) esponses.
6.3.2. D i -Di usion Resul s
Subsequen ly, i alues o he 7 DDMs buil we e checked o di e ences be ween he
di e en models in he MRI lexical decision ask. The model ha showed he bes i , a e
con olling o he numbe o pa ame e s included, was he Seman ic DDM (AIC=-579.315),
ollowed by he Familia i y DDM (AIC=-565.096) and he F equency DDM (AIC=-549.906).
The Conc e eness DDM (AIC=-491.970) s ill pe o med be e han he wo Random DDMs,
bu conside ably wo se han he Seman ic, Familia i y and F equency DDMS. LRT es s u he
indica ed ha he Seman ic DDM pe o med be e han he F equency (p<.001), Familia i y
(p=.058) and Conc e eness (p<.001) DDMs. Las ly, he Sublexical DDM (AIC=-444.727) did
no pe o m be e han he Random 3 DDM (AIC=-464.929).
As o nonwo ds, he o e all i o he model was conside ably lowe (mean AIC=-
15.227). I should be conside ed ha he d i in his model was kep ixed o all ials, as
103 OF 150
opposed o he es o he models, whose d i s dynamically a ied wi h he a iables o
in e es . Figu e 6.4 shows he AIC alues, a e aged ac oss subjec s, o each o he 7 wo d
models es ed and he nonwo ds model.
Figu e 6.4. A) Fi o he DDMs in he MRI lexical decision ask, based on he ans o ma ion o NLL alues
o AIC. Lowe alues indica e be e i . No ice ha he Nonwo ds model is es ima ed di e en ly ( ixed d i )
han he es o he models (dynamically changing d i ). B) Dis ibu ion o he bes - i ing d i s om he
Seman ic DDM ac oss subjec s. C) Dis ibu ion o he bes - i ing d i s om he Sublexical DDM.
The bes - i ing d i s o he Seman ic DDM we e also analysed in o de o u he
examine he con ibu ions o each o he a iables o he decision making p ocess. Consis en
wi h he analysis o he AIC alues, he highes con ibu o o he d i o he Seman ic DDM
was Familia i y (mean d i =1.413), ollowed by F equency (mean d i =1.144). Conc e eness
showed d i s ha we e close o ze o (mean d i =0.023), indica ing ha his a iable had li le
in luence on he o e all d i o he model.
Addi ionally, he bes - i ing d i s o he Sublexical DDM we e explo ed. Al hough he
i o his model was no be e han he Random 3 DDM, he d i s can p o ide addi ional
in o ma ion abou he con ibu ion o he a iables included in he model. Numbe o le e s
d i alues we e conside ably high (mean d i =1.257), being he only a iable in he model
104 OF 150
whose d i s signi ican ly di e ed om ze o. O no e, O hog aphic dis ance (OLD20) d i s
showed high a iabili y ac oss subjec s.
6.3.3. Co ela ions be ween DDMs and B ain Rep esen a ions
We sys ema ically inspec ed he co ela ions be ween each o he d i s ob ained om
he DDM analyses in he MRI ask (see Figu e 6.4[B,C]) and he ROI-based RSA simila i ies
yielded by he seman ic, phonological and wo d2 ec models. O all se en ROIs, only he IFG
pa s o bi alis, he ATL and FG2 yielded signi ican co ela ions su i ing he FDR co ec ion.
Wi hin hese a eas, we only obse ed associa ions be ween some o he seman ic DDM d i s
and he seman ic RSA model. The esul s a e desc ibed below, and ep esen ed in Figu e 6.5.
Figu e 6.5. Signi ican RSA-DDM MRI d i co ela ions su i ing he FDR co ec ion o an alpha o
.05. Each do ep esen s a subjec .
In IFG pa s o bi alis, he Familia i y d i was posi i ely associa ed wi h he seman ic
RSA model simila i ies ( =0.563, q=.013), while he Conc e eness d i was nega i ely
105 OF 150
co ela ed wi h he same seman ic RSA model ( =-0.589, q=.013). The hie a chical linea
eg ession analysis in his ROI indica ed ha a model wi h Familia i y d i alone explained
25.3% o he a iance (R2=0.253, F=9.162, p=.005), wi h Familia i y d i signi ican ly
p edic ing RSA simila i y (𝛽=.009, p=.005, CI=[.003,.015]). Adding F equency d i o he
model helped explain up o 30.2 % o he a iance (R2=0.302, F=5.615, p=.009), wi h
Familia i y s ill being a signi ican p edic o o RSA simila i y (𝛽=.007, p=.036), bu wi h
F equency no being a signi ican p edic o (𝛽=-.003, p=.192). Finally, including Conc e eness
d i in he model summed up o 32.4% o he a iance explained (R2=324, F=3.997, p=.018).
Howe e , in his model, none o he d i s we e signi ican p edic o s o RSA simila i y
(Familia i y: 𝛽=.005, p=.203; F equency: 𝛽=-.001., p=.644; Conc e eness: 𝛽=-.005, p=.370).
This was no likely due o mul icollinea i y, gi en he conside ably low VIF alues associa ed
wi h he h ee d i s (Familia i y=1.787, F equency=2.014, Conc e eness=2.931). The ollow-
up ANOVA indica ed ha he abo e-men ioned inc eases in he a iance explained did no
each signi icance a e he addi ions o F equency d i (F=1.786, p=0.192) o Conc e eness
d i (F=0.833, p=0.370) o he model.
In he ATL, he Conc e eness d i was nega i ely co ela ed wi h he seman ic RSA
model simila i ies ( =-0.582, q=0.017). The eg ession model wi h Conc e eness d i alone
explained 29.6% o he a iance (R2=0.296, F=11.37, p=.002), wi h Conc e eness d i as a
signi ican p edic o o RSA simila i y (𝛽=-.009, p=.002, CI=[-.015, -.004]). Adding Familia i y
d i o he model did no con ibu e o he explained a iance (R2=0.297, F=5.489, p=.01).
Finally, a e adding F equency d i o he model, he a iance explained inc eased o 31.6%
(R2=0.316, F=3.855, p=.02), bu Conc e eness d i was he only signi ican p edic o o he
RSA simila i y (Conc e eness: 𝛽=-.012, p=.018, CI=[-.023, -.002]; Familia i y: 𝛽<-.001, p=.828,
CI=[-.008, .006]; F equency: 𝛽=.002, p=.408, CI=[-.004, .009]). The ollow-up ANOVA
indica ed ha he obse ed inc ease in he a iance explained a e adding F equency d i as
a hi d p edic o was no signi ican (F=0.709, p=.407).
As o he FG2 ROI, Conc e eness d i was also nega i ely co ela ed wi h he
seman ic RSA model simila i y ( =-0.549, q=.037). In he subsequen hie a chical linea
eg ession analysis, a model wi h Conc e eness d i alone explained 28.1% o he a iance
(R2=0.281, F=10.54, p=.003), wi h Conc e eness d i as a signi ican p edic o o RSA
simila i y (𝛽=-.012, p=.003, CI=[-.021, -.005]). A e adding Familia i y d i , he model
explained 30.1% o he a iance (R2=0.301, F=5.607, p=.009), al hough wi h Conc e eness as
he only ma ginally signi ican p edic o o RSA simila i y (Conc e eness: 𝛽=-.010, p=.076,
CI=[-.021, .001]; Familia i y: 𝛽=.004, p=.389, CI=[-.005, .014]). Finally, adding F equency d i
o he model did no inc ease he a iance explained by he model (R2=0.303, F=3.625,
106 OF 150
p=.026). The ollow-up ANOVA e ealed ha he inc ease in he a iance explained a e
adding Familia i y d i was no s a is ically signi ican (F=0.74, p= .397).
We also explo ed he associa ions be ween aw RTs and RSA model simila i ies. In
his sense, al hough we obse ed signi ican o ma ginally signi ican nega i e co ela ions in
he mid/pos e io IFG (pa s ope cula is: =-0.334, p=.075; pa s iangula is: =-0.37, p=.047),
hese co ela ions did no su i e he FDR co ec ion (pa s ope cula is: q=.264; pa s
iangula is: q=.264). No addi ional e ec s we e ound in any o he ROIs, o any o he RSA
models.
6.4. DISCUSSION
In his wo k, we aimed o quan i y he deg ee o which di e en psycholinguis ic
p ope ies a e able o in luence he decision-making p ocess in wo di e en lexical decision
asks, unde di e en iming and en i onmen al condi ions, and wi h a ixed o de . We
add essed his by employing D i -Di usion Model (DDM) analyses. This app oach addi ionally
se ed he speci ic aim o assessing he link be ween he abili y o lexico-seman ic and
sublexical p ope ies o in luence decision-making ( ep esen ed by he d i s o he DDMs),
and seman ic and phonological b ain ep esen a ions ound in Chap e 5. The main indings
a e discussed below.
6.4.1. Psycholinguis ic P ope ies and Decision-Making
In ou p e ious s udy, we could no ensu e ha ha ing pa icipan s simply decide
whe he a gi en s ing o le e s o ms a wo d ha exis s in hei language o no could
e ec i ely igge lexico-seman ic p ocessing (i.e., he p ocessing o wo ds as a whole and
hei meaning). Because we we e awa e o his po en ial limi a ion, we sough o examine he
abili y o sublexical s. lexico-seman ic p ope ies o in luence beha iou du ing lexical
decisions. As we expec ed, we ound ha lexico-seman ic, mo e so han sublexical p ope ies,
d o e mos o he e idence accumula ion a e owa ds eaching a decision in wo di e en
lexical decision asks. A DDM based on wo d-by-wo d Familia i y, F equency and
Conc e eness was he bes model o accoun o e idence accumula ion a e, e en a e
con olling o model complexi y. In con as , a DDM buil om sublexical a iables like he
numbe o le e s, o hog aphic dis ance and big am equency, did no pe o m be e han a
DDM buil om h ee andomly gene a ed a iables. The mos ele an con ibu o o he
decision a e was wo d Familia i y, ollowed by wo d F equency, a iables ha a e ine i ably
associa ed wi h lexical p ocessing o meaning (Chee e al., 2002, 2003; Ne eu &
Kaushanskaya, 2023; Shinozuka e al., 2021). In his sense, hese wo a iables sum up o
con ibu e o deciding ha a s ing is a eal wo d. Al hough in ui i ely one would expec ha
107 OF 150
his ype o ask should be pe o med wi hou he need o seman ic p ocessing, i is appa en
om ou esul s ha lexico-seman ic p ope ies a e igge ed au oma ically in ypical eade s,
e en in a seemingly simple linguis ic ask.
I has been ypically assumed ha lexical access migh no be s ic ly equi ed o
pe o m a lexical decision ask, and ha i a he elies on au oma ic isual wo d ecogni ion
(Col hea , 2004; Col hea e al., 2022). I should be no ed ha ou esul s a e no aken as a
di ec indica ion ha lexical access is a uni a y p ocess. Jus as wo d F equency is likely o
ap in o bo h phonological access (Ca ei as e al., 2009; Fiebach e al., 2002), and seman ic
access (Chee e al., 2002, 2003), i is possible ha bo h wo d Familia i y and wo d leng h
a ec lexical decisions simul aneously. In he p esen wo k, wi hin a sublexical model buil
om phonological and o hog aphic ea u es, a a iable like he numbe o le e s cons i u ing
he wo d was likely o in luence he e idence accumula ion a e du ing lexical decision, gi en
he obse ed la ge d i alues associa ed wi h i . Al hough we di ided sublexical and lexico-
seman ic a iables in o wo sepa a e models, i is possible ha bo h kinds o a iables
con e ge o in luence he ease wi h which a lexical decision is eached. This in e p e a ion
would be in line wi h p e ious ecen indings indica ing ha bo h sublexical and lexico-
seman ic p ope ies in luence he spa io empo al neu al dynamics o lexical p ocessing
(Woolnough e al., 2021). The ho ough explo a ion o his complex spa io empo al in e ac ion,
howe e , was beyond he scope o his wo k, and su ice i o say ha lexico-seman ic
p ope ies such as wo d Familia i y and F equency, we e he mos impo an de e minan s o
he obse ed lexical decision imes, ollowed by a sublexical a iable like wo d leng h.
Ou indings we e eplica ed in wo di e en e sions o he same ask (inside he MRI
s. ou side he scanne ), which a ied in he iming condi ions and e en in he p opo ion o
nonwo d s imuli, and each es ed in a di e en se o wo ds. Hence, he in luence o lexico-
seman ic p ocessing du ing lexical decision-making became appa en in ou esul s.
6.4.2. Associa ion be ween D i Ra e and B ain Rep esen a ions
In e p e ing he associa ions be ween neu al e ec s and he obse ed beha iou is
o en imes a challenging endea ou . In Chap e 5, we desc ibed how b ain a eas such as he
an e io IFG, he ATL o he an e io OTC, a e ec ui ed o seman ic ep esen a ions, as
illus a ed by hei ep esen a ional simila i y wi h psycholinguis ic seman ic and wo d ec o
RSA models. In his wo k, we aimed o go a s ep u he and inspec whe he hese
ep esen a ions we e associa ed wi h seman ic d i s, which embody he abili y o lexico-
seman ic p ope ies o in luence decision making. I such associa ions a e g ea e han any
po en ial link be ween aw RTs and he neu al ep esen a ions, hen such inding would u he
suppo he in e p e a ion o hese neu al e ec s in e ms o seman ic b ain ep esen a ions.
We hypo hesised ha seman ic and wo d ec o ep esen a ions would show high co ela ions
108 OF 150
wi h seman ic d i s in he an e io IFG, ATL and an e io OTC, while aw RTs would also
show signi ican co ela ions in he pos e io IFG and he OTC. Ou esul s pa ially suppo ed
hese expec a ions.
In he an e io IFG, Familia i y d i a ose as a good p edic o o b ain simila i ies wi h
a seman ic model buil om psycholinguis ic p ope ies, bu no wi h a model buil om wo d
ec o s. Wo d Familia i y is a subjec i e measu e o he equency wi h which indi iduals a e
exposed o concep ual in o ma ion. I has been associa ed wi h seman ic access, wo d
lea ning and memo y (Ne eu & Kaushanskaya, 2023), and i has been ound o elici IFG
esponses ha could be linked o he le el o in e nalisa ion o lexical con en (Shinozuka e
al., 2021). In his sense, ou da a seem o indica e ha i is hese cha ac e is ics ha mainly
d i e seman ic ep esen a ions in an a ea ha is igh ly associa ed wi h access o seman ics
such as he an e io IFG. Ne e heless, i should be no ed ha his inding was no eplica ed
when using a wo d ec o model, and hence i should be aken cau iously. Al hough F equency
and Conc e eness d i s sligh ly imp o ed he p edic ions o seman ic simila i ies in he an e io
IFG, he con ibu ion o hese wo a iables was limi ed. On he one hand, as indica ed abo e,
because wo d F equency aps in o sublexical and lexico-seman ic p ocesses, i is no
su p ising ha i s con ibu ion o p edic seman ic b ain ep esen a ions was modes . On he
o he hand, in ou s udy, wo d Conc e eness seemed o in e ac wi h wo d Familia i y a he
neu al le el (see sec ion 5.4.1), bu i did no seem o ha e a ele an impac on decision
making in ou asks. This ac could explain why i did no a ise as a good p edic o o seman ic
ep esen a ions in he an e io IFG. Po en ially, his migh be indica ing ha when including
o he ac o s ha be e capi alise on he equency o exposu e o in o ma ion, a key ea u e
in he modula ion o he an e io IFG unc ion (as shown in Chap e 4), wo d Conc e eness
becomes less ele an o lexical access.
A inding ha is mo e challenging o in e p e was he nega i e co ela ion be ween
Conc e eness d i and seman ic b ain ep esen a ions in he an e io IFG, ATL and he
pos e io OTC. D i can adop nega i e alues, meaning, in his scena io, ha he highe he
speci ic d i a iable (i.e. Conc e eness), he as e he esponse owa ds he lowe end (i.e.,
as e e idence accumula ion o he “nonwo d” esponse). In o he wo ds, ou s udy included
pa icipan s o whom highly conc e e wo ds we e ha de o ecognise as wo ds. In many
cases, his kind o p o ile was associa ed wi h highe seman ic simila i ies in he men ioned
ROIs. A possible explana ion is ha in hese cases, Conc e eness is no eally a ec ing
seman ic p ocessing, bu ins ead e lec ing sublexical p ocessing associa ed wi h speci ic
ea u es o abs ac wo ds. Fo ins ance, many abs ac wo ds sha e e mina ions like -ción (-
ion in English) o -dad (-i y in English), which migh ha e helped in ecognising hese s ings
as wo ds. I his was he case, hen i is plausible ha in hose pa icipan s wi h nega i e
Conc e eness d i s, he seman ic RSA model was pe o ming be e because Conc e eness
109 OF 150
did no compe e wi h he o he wo seman ic a iables, hus leading o a clea e , less a iable
e ec o e neu al ep esen a ions. This in e p e a ion would be consis en wi h ou obse a ion
ha , when ying o p edic seman ic simila i ies wi h Conc e eness as he s a ing a iable in
he hie a chy o a linea eg ession model, he simila i y baseline ( ep esen ed by he cons an
e m) was g ea ly di e en om 0. Mo eo e , his possibili y also aligns wi h he abo e-
men ioned iew o lexical access as a dynamic p ocess, in which lexico-seman ic and
sublexical mechanisms in e play, each aking o e unde speci ic ci cums ances (e.g., So o
e al., 2020).
Finally, we expec ed o obse e signi ican co ela ions be ween aw RTs and model
simila i ies in he pos e io IFG and he OTC, owing o he in luence o decision-making no
ela ed o lexical p ocessing (i.e., noise). Al hough his was he case in he pos e io IFG, such
co ela ions did no su i e he mul iple compa isons co ec ions. I is possible ha b ain
ep esen a ions ob ained om mul i a ia e analyses do no exac ly e lec ac o s ha a ec
BOLD signal in ensi y pe se (Haynes & Rees, 2006), such as decision-making in complex
scena ios. In u n, RSA is be e sui ed o explo ing complex concep ual s uc u es (F isby e
al., 2023), while being less sensi i e o single ac o s (Lewis-Peacock & No man, 2014).
Ne e heless, ou indings seem o indica e ha he b ain ep esen a ions ound in Chap e 5
we e unlikely o be due o noise a ibu ed o complex decision-making.
Some limi a ions should be men ioned. Fi s ly, he iming and ha dwa e di e ences
be ween he MRI and he beha iou al asks did no allow a balanced, comp ehensi e
compa ison be ween he esul s o bo h asks. As a consequence, all such compa isons we e
me ely desc ip i e, and lack he po en ial o d aw any kind o in e ence. Secondly, al hough
we explo ed di e en combina ions o psycholinguis ic p ope ies in o a sublexical and a
lexico-seman ic model, hyb id models ha inco po a ed bo h ypes o a iables we e no
u he explo ed. F om ou esul s, i seems ha lexical access is no a uni a y p ocess, and
al hough i is hea ily in luenced by lexico-seman ic p ope ies in ypical eade s, some
sublexical wo d p ope ies, like wo d leng h, seem o in e play in his mechanism. Fu u e
s udies should u he explo e his in e p e a ion. And las ly, we decided o employ he ROIs
used in Chap e 5 o a clea in e p e a ion o he e ec s. Howe e , because hese ROIs do
no necessa ily o e lap wi h he sup a- h eshold simila i ies ound in he whole-b ain
sea chligh , i is possible ha some o he associa ions be ween b ain ep esen a ions and he
d i s ob ained in his wo k we e o e looked. Simila ly, some o he e ec s ha spanned
ou side hese ROIs, like hose ound in he pa ahippocampal gy us, we e no explo ed.
Ne e heless, ou decision o employing he p e-buil ROIs was based on ou aims and
hypo heses, and hese ques ions may be add essed in u u e analyses.
110 OF 150
6.5. CONCLUSIONS
We ound obus e idence ha lexico-seman ic p ocessing can be igge ed du ing a
seemingly simple lexical decision ask. In ac , lexico-seman ic p ope ies, and especially,
wo d Familia i y and wo d F equency, a ose as he mos e iden con ibu o s o decision
making du ing lexical decisions. Howe e , o he sublexical a iables, like wo d leng h, seemed
o in e play wi h his lexico-seman ic p ocessing o in luence lexical access in ou s udy. A he
neu al le el, he abili y o wo d Familia i y o in luence lexical access was ound o be igh ly
associa ed wi h seman ic ep esen a ions in he an e io IFG. Al hough o he seman ic
p ope ies, such as wo d Conc e eness, seem o d i e pa o he seman ic ep esen a ions in
a eas like he ATL o he OTC, hese associa ions we e possibly due o spu ious in e ac ions
wi h sublexical p ocessing, and hence should be u he in es iga ed in u u e s udies. In sum,
ou esul s highligh he dynamic na u e o sublexical and seman ic lexical p ocessing
mechanisms, and he ole o he an e io IFG in seman ic access du ing eading.
111 OF 150
GENERAL DISCUSSION
The gene al objec i e o his hesis was o gi e a comp ehensi e iew o he neu al
unde pinnings o he access and use o concep ual in o ma ion du ing eading. Speci ically,
he hesis aimed o imp o e ou unde s anding o how, om pe cep ual linguis ic in o ma ion,
ou b ain enables he access o lexical concep ual in o ma ion, and how i uses i o adap o
he ask a hand. To his end, I in es iga ed sublexical o lexico-seman ic measu able wo d
p ope ies and hei associa ed b ain ac i a ion pa e ns and ep esen a ions. We can highligh
wo main con ibu ions. Fi s ly, he iden i ica ion o unc ional dissocia ions wi hin (and beyond)
b ain a eas ha a e key in he access o lexico-seman ic in o ma ion, like he IFG (ex ending
o mo o planning a eas such as he SMA) and OTC (ex ending o he IOG), which a e
dynamically ec ui ed by sublexical and lexico-seman ic p ocesses du ing lexical sea ch. And
secondly, his hesis unde sco es he impo ance o using psycholinguis ic analyses and
na u al language p ocessing in combina ion o disen angle neu al ep esen a ions o seman ic
in o ma ion in con e gence a eas like he IFG, ATL and he OTC. These key indings a e
discussed in he con ex o he mos ecen e idence a ailable.
Func ional Dissocia ions in he IFG and OTC and hei Dynamic Na u e
The idea ha he di e en subdi isions o he le IFG a e ec ui ed di e en ially o
speci ic language and memo y p ocesses was pu o wa d nea ly wo decades ago. In eading,
he ac i i y o he an e io IFG (pa s o bi alis) has been epea edly associa ed wi h access o
meaning and o e ie al o in o ma ion ha is no eadily a ailable, as when we p ocess
un amilia wo ds (Bad e & Wagne , 2005; Hagoo , 2005). The a eas pos e io o he an e io
IFG ha e been associa ed wi h he selec ion among he compe ing in o ma ion ha is a ailable
(Bad e & Wagne , 2007), as i is he case o syn ac ic p ocessing associa ed wi h he middle
IFG (pa s iangula is), o phonological selec ion and mo o planning implica ing he pos e io
IFG (pa s ope cula is) (Hagoo , 2013). Mo e ecen ly, di e en iews ha e been p oposed
ha suppo a a he in e ac i e (and no so he me ic) amewo k, in which di e en sys ems
(like decla a i e memo y and language comp ehension) in e play (Roge , Banjac, e al., 2022),
and ely on o e lapping la ge-scale ne wo ks (Fedo enko e al., 2024; Roge , Banjac, e al.,
2022). This includes he IFG, a complex s uc u e in which di e se unc ions, om decla a i e
memo y o language p oduc ion, o ac ion pe cep ion, con ey (F iede ici, 2023). In his sense,
se e al key ne wo ks (salience ne wo k, SAL; on opa ie al ne wo k, FPN; and de aul mode
ne wo k, DMN) con e ge in he IFG, and wi h his con e gence, i is plausible ha unc ions
ha pe ain o phonology and a icula ion, syn ac ic and seman ic p ocessing, as well as
moni o ing and con ol o inpu and ou pu linguis ic p ocessing also con ey (see Roge ,
Banjac, e al., 2022 o a de ailed e iew).
118 OF 150
Hie a chical O ganiza ion o he P e on al Co ex. Jou nal o Cogni i e
Neu oscience, 19(12), 2082–2099. h ps://doi.o g/10.1162/jocn.2007.19.12.2082
Bad e, D., Pold ack, R. A., Pa é-Blagoe , E. J., Insle , R. Z., & Wagne , A. D. (2005).
Dissociable Con olled Re ie al and Gene alized Selec ion Mechanisms in
Ven ola e al P e on al Co ex. Neu on, 47(6), 907–918.
h ps://doi.o g/10.1016/j.neu on.2005.07.023
Bad e, D., & Wagne , A. D. (2005). F on al Lobe Mechanisms ha Resol e P oac i e
In e e ence. Ce eb al Co ex, 15(12), 2003–2012.
h ps://doi.o g/10.1093/ce co /bhi075
Bad e, D., & Wagne , A. D. (2006). Compu a ional and neu obiological mechanisms
unde lying cogni i e lexibili y. P oceedings o he Na ional Academy o Sciences,
103(18), 7186–7191. h ps://doi.o g/10.1073/pnas.0509550103
Bad e, D., & Wagne , A. D. (2007). Le en ola e al p e on al co ex and he cogni i e
con ol o memo y. Neu opsychologia, 45(13), 2883–2901.
h ps://doi.o g/10.1016/j.neu opsychologia.2007.06.015
Banaszkiewicz, A., Ma uszewski, J., Bola, Ł., Szczepanik, M., Kossowski, B., Ru kowski, P.,
Szwed, M., Emmo ey, K., Jedno óg, K., & Ma chewka, A. (2021). Mul imodal imaging
o b ain eo ganiza ion in hea ing la e lea ne s o sign language. Human B ain
Mapping, 42(2), 384–397. h ps://doi.o g/10.1002/hbm.25229
Ba nha , A. S., & Goldinge , S. D. (2015). O hog aphic and phonological neighbo hood
e ec s in handw i en wo d pe cep ion. Psychonomic Bulle in & Re iew, 22(6), 1739–
1745. h ps://doi.o g/10.3758/s13423-015-0846-z
Bašnáko á, J., Webe , K., Pe e sson, K. M., Van Be kum, J., & Hagoo , P. (2014). Beyond
he Language Gi en: The Neu al Co ela es o In e ing Speake Meaning. Ce eb al
Co ex, 24(10), 2572–2578. h ps://doi.o g/10.1093/ce co /bh 112
Beauchamp, M. S., A gall, B. D., Bodu ka, J., Duyn, J. H., & Ma in, A. (2004). Un a eling
mul isenso y in eg a ion: Pa chy o ganiza ion wi hin human STS mul isenso y co ex.
Na u e Neu oscience, 7(11), 1190–1192. h ps://doi.o g/10.1038/nn1333
119 OF 150
Bedny, M., & Thompson-Schill, S. L. (2006). Neu oana omically sepa able e ec s o
imageabili y and g amma ical class du ing single-wo d comp ehension. B ain and
Language, 98(2), 127–139. h ps://doi.o g/10.1016/j.bandl.2006.04.008
Benne , J., Reinha d , J., Ch is ine , M., Wengen o h, M., S ippich, C., Schneide , P., &
Bla ow, M. (2023). Tempo al hie a chy o co ical esponses e lec s co e-bel -
pa abel o ganiza ion o audi o y co ex in musicians. Ce eb al Co ex, 33(11), 7044–
7060. h ps://doi.o g/10.1093/ce co /bhad020
Bilbao-Jayo, A., & Almeida, A. (2018). Au oma ic poli ical discou se analysis wi h mul i-scale
con olu ional neu al ne wo ks and con ex ual da a. In e na ional Jou nal o
Dis ibu ed Senso Ne wo ks, 14(11), 155014771881182.
h ps://doi.o g/10.1177/1550147718811827
Bilecen, D., Sei i z, E., Sche le , K., Henning, J., & Schul e, A.-C. (2002). Ampli opici y o
he Human Audi o y Co ex: An MRI S udy. Neu oImage, 17(2), 710–718.
h ps://doi.o g/10.1006/nimg.2002.1133
Binde , J. R., & Desai, R. H. (2011). The neu obiology o seman ic memo y. T ends in
Cogni i e Sciences, 15(11), 527–536. h ps://doi.o g/10.1016/j. ics.2011.10.001
Binde , J. R., Desai, R. H., G a es, W. W., & Conan , L. L. (2009). Whe e Is he Seman ic
Sys em? A C i ical Re iew and Me a-Analysis o 120 Func ional Neu oimaging
S udies. Ce eb al Co ex, 19(12), 2767–2796. h ps://doi.o g/10.1093/ce co /bhp055
Binney, R. J., Pa ke , G. J. M., & Lambon Ralph, M. A. (2012). Con e gen Connec i i y and
G aded Specializa ion in he Ros al Human Tempo al Lobe as Re ealed by
Di usion-Weigh ed Imaging P obabilis ic T ac og aphy. Jou nal o Cogni i e
Neu oscience, 24(10), 1998–2014. h ps://doi.o g/10.1162/jocn_a_00263
Bo ghi, A. M., & Binko ski, F. (2014). Wo ds as Social Tools: An Embodied View on Abs ac
Concep s. Sp inge New Yo k. h ps://doi.o g/10.1007/978-1-4614-9539-0
Bou ie , S. E., & Engel, S. A. (2006). Beha io al De ici s and Co ical Damage Loci in
Ce eb al Ach oma opsia. Ce eb al Co ex, 16(2), 183–191.
h ps://doi.o g/10.1093/ce co /bhi096
120 OF 150
B aunsdo , M., Blazquez F eches, G., Roumazeilles, L., Eiche , N., Schu z, M., Ui hol, S.,
B yan , K. L., & Ma s, R. B. (2021). Does he empo al co ex make us human? A
e iew o s uc u al and unc ional di e si y o he p ima e empo al lobe.
Neu oscience & Biobeha io al Re iews, 131, 400–410.
h ps://doi.o g/10.1016/j.neubio e .2021.08.032
B e , M., An on, J.-L., Valab egue, R., & Poline, J.-B. (2002). Region o in e es analysis
using an SPM oolbox. 1.
B ouwe , G. J., & Heege , D. J. (2009). Decoding and Recons uc ing Colo om Responses
in Human Visual Co ex. The Jou nal o Neu oscience, 29(44), 13992–14003.
h ps://doi.o g/10.1523/JNEUROSCI.3577-09.2009
B uno, J. L., Zumbe ge, A., Manis, F. R., Lu, Z.-L., & Goldman, J. G. (2008a). Sensi i i y o
o hog aphic amilia i y in he occipi o- empo al egion. Neu oImage, 39(4), 1988–
2001. h ps://doi.o g/10.1016/j.neu oimage.2007.10.044
B uno, J. L., Zumbe ge, A., Manis, F. R., Lu, Z.-L., & Goldman, J. G. (2008b). Sensi i i y o
o hog aphic amilia i y in he occipi o- empo al egion. Neu oImage, 39(4), 1988–
2001. h ps://doi.o g/10.1016/j.neu oimage.2007.10.044
B ysbae , M., Mande a, P., & Keulee s, E. (2018). The Wo d F equency E ec in Wo d
P ocessing: An Upda ed Re iew. Cu en Di ec ions in Psychological Science, 27(1),
45–50. h ps://doi.o g/10.1177/0963721417727521
Bullmo e, E., & Spo ns, O. (2009). Complex b ain ne wo ks: G aph heo e ical analysis o
s uc u al and unc ional sys ems. Na u e Re iews Neu oscience, 10(3), 186–198.
h ps://doi.o g/10.1038/n n2575
Cabeza, R., Mazuz, Y. S., S okes, J., K agel, J. E., Woldo , M. G., Cia amelli, E., Olson, I.
R., & Mosco i ch, M. (2011). O e lapping Pa ie al Ac i i y in Memo y and Pe cep ion:
E idence o he A en ion o Memo y Model. Jou nal o Cogni i e Neu oscience,
23(11), 3209–3217. h ps://doi.o g/10.1162/jocn_a_00065
Cao, F., Khalid, K., Za e i, R., Bolge , D. J., Bi an, T., & Boo h, J. R. (2010). Neu al
co ela es o p iming e ec s in child en du ing spoken wo d p ocessing wi h
121 OF 150
o hog aphic demands. B ain and Language, 114(2), 80–89.
h ps://doi.o g/10.1016/j.bandl.2009.07.005
Caplan, J. B., & Madan, C. R. (2016). Wo d Imageabili y Enhances Associa ion-memo y by
Inc easing Hippocampal Engagemen . Jou nal o Cogni i e Neu oscience, 28(10),
1522–1538. h ps://doi.o g/10.1162/jocn_a_00992
Ca o a, F., K iegesko e, N., Nili, H., & Pul e mülle , F. (2017). Rep esen a ional Simila i y
Mapping o Dis ibu ional Seman ics in Le In e io F on al, Middle Tempo al, and
Mo o Co ex. Ce eb al Co ex, 27(1), 294–309.
h ps://doi.o g/10.1093/ce co /bhw379
Ca o a, F., Nili, H., Pul e mülle , F., & K iegesko e, N. (2021a). Dis inc on o- empo al
subs a es o dis ibu ional and axonomic simila i y among wo ds: E idence om
RSA o BOLD signals. Neu oImage, 224, 117408.
h ps://doi.o g/10.1016/j.neu oimage.2020.117408
Ca o a, F., Nili, H., Pul e mülle , F., & K iegesko e, N. (2021b). Dis inc on o- empo al
subs a es o dis ibu ional and axonomic simila i y among wo ds: E idence om
RSA o BOLD signals. Neu oImage, 224, 117408.
h ps://doi.o g/10.1016/j.neu oimage.2020.117408
Ca ei as, M., A ms ong, B. C., Pe ea, M., & F os , R. (2014). The wha , when, whe e, and
how o isual wo d ecogni ion. T ends in Cogni i e Sciences, 18(2), 90–98.
h ps://doi.o g/10.1016/j. ics.2013.11.005
Ca ei as, M., Mechelli, A., & P ice, C. J. (2006). E ec o wo d and syllable equency on
ac i a ion du ing lexical decision and eading aloud. Human B ain Mapping, 27(12),
963–972. h ps://doi.o g/10.1002/hbm.20236
Ca ei as, M., Riba, J., Ve ga a, M., Heldmann, M., & Mün e, T. F. (2009). Syllable
cong uency and wo d equency e ec s on b ain ac i a ion. Human B ain Mapping,
30(9), 3079–3088. h ps://doi.o g/10.1002/hbm.20730
Ca ani, M., Jones, D. K., & F y che, D. H. (2005). Pe isyl ian language ne wo ks o he
human b ain. Annals o Neu ology, 57(1), 8–16. h ps://doi.o g/10.1002/ana.20319
122 OF 150
Celesia, G. G. (2005). Chap e 2 Ana omy and physiology o he isual pa hways and co ex.
In G. G. Celesia (Ed.), Handbook o Clinical Neu ophysiology (Vol. 5, pp. 45–63).
Else ie . h ps://doi.o g/10.1016/S1567-4231(09)70199-6
Chao, L. L., Weisbe g, J., & Ma in, A. (2002). Expe ience-dependen Modula ion o
Ca ego y- ela ed Co ical Ac i i y. Ce eb al Co ex, 12(5), 545–551.
h ps://doi.o g/10.1093/ce co /12.5.545
Chee, M. W. L., Hon, N. H. H., Caplan, D., Lee, H. L., & Goh, J. (2002). F equency o
Conc e e Wo ds Modula es P e on al Ac i a ion du ing Seman ic Judgmen s.
Neu oImage, 16(1), 259–268. h ps://doi.o g/10.1006/nimg.2002.1061
Chee, M. W. L., Wes phal, C., Goh, J., G aham, S., & Song, A. W. (2003). Wo d equency
and subsequen memo y e ec s s udied using e en - ela ed MRI. Neu oImage,
20(2), 1042–1051. h ps://doi.o g/10.1016/S1053-8119(03)00335-5
Chen, R., Wang, F., Liang, H., & Li, W. (2017). Syne gis ic P ocessing o Visual Con ou s
ac oss Co ical Laye s in V1 and V2. Neu on, 96(6), 1388-1402.e4.
h ps://doi.o g/10.1016/j.neu on.2017.11.004
Chen, Y., Da is, M. H., Pul e mülle , F., & Hauk, O. (2013). Task modula ion o b ain
esponses in isual wo d ecogni ion as s udied using EEG/MEG and MRI. F on ie s
in Human Neu oscience, 7. h ps://doi.o g/10.3389/ nhum.2013.00376
Chen, Y., Da is, M. H., Pul e mülle , F., & Hauk, O. (2015). Ea ly Visual Wo d P ocessing Is
Flexible: E idence om Spa io empo al B ain Dynamics. Jou nal o Cogni i e
Neu oscience, 27(9), 1738–1751. h ps://doi.o g/10.1162/jocn_a_00815
Chia ello, C., Vaden, K. I., & Ecke , M. A. (2018). O hog aphic in luence on spoken wo d
iden i ica ion: Beha io al and MRI e idence. Neu opsychologia, 111, 103–111.
h ps://doi.o g/10.1016/j.neu opsychologia.2018.01.032
Cla ke, A., & Tyle , L. K. (2015). Unde s anding Wha We See: How We De i e Meaning
F om Vision. T ends in Cogni i e Sciences, 19(11), 677–687.
h ps://doi.o g/10.1016/j. ics.2015.08.008
Coggan, D. D., & Tong, F. (2023). Spikiness and animacy as po en ial o ganizing p inciples
123 OF 150
o human en al isual co ex. Ce eb al Co ex, 33(13), 8194–8217.
h ps://doi.o g/10.1093/ce co /bhad108
Cohen, L. (2003). Visual Wo d Recogni ion in he Le and Righ Hemisphe es: Ana omical
and Func ional Co ela es o Pe iphe al Alexias. Ce eb al Co ex, 13(12), 1313–
1333. h ps://doi.o g/10.1093/ce co /bhg079
Cohen, L., & Dehaene, S. (2004). Specializa ion wi hin he en al s eam: The case o he
isual wo d o m a ea. Neu oImage, 22(1), 466–476.
h ps://doi.o g/10.1016/j.neu oimage.2003.12.049
Cohen, L., Dehaene, S., Naccache, L., Lehé icy, S., Dehaene-Lambe z, G., Héna , M.-A.,
& Michel, F. (2000). The isual wo d o m a ea: Spa ial and empo al cha ac e iza ion
o an ini ial s age o eading in no mal subjec s and pos e io spli -b ain pa ien s.
B ain, 123(2), 291–307. h ps://doi.o g/10.1093/b ain/123.2.291
Cohen, L., Jobe , A., Le Bihan, D., & Dehaene, S. (2004). Dis inc unimodal and mul imodal
egions o wo d p ocessing in he le empo al co ex. Neu oImage, 23(4), 1256–
1270. h ps://doi.o g/10.1016/j.neu oimage.2004.07.052
Cohen, L., Lehé icy, S., Chochon, F., Leme , C., Ri aud, S., & Dehaene, S. (2002).
Language‐speci ic uning o isual co ex? Func ional p ope ies o he Visual Wo d
Fo m A ea. B ain, 125(5), 1054–1069. h ps://doi.o g/10.1093/b ain/aw 094
Col hea , M. (2004). A e The e Lexicons? The Qua e ly Jou nal o Expe imen al
Psychology Sec ion A, 57(7), 1153–1172.
h ps://doi.o g/10.1080/02724980443000007
Col hea , M., Da elaa , E., Jonasson, J. T., & Besne , D. (2022). Access o he In e nal
Lexicon. In S. Do nič, A en ion and Pe o mance VI (1s ed., pp. 535–555).
Rou ledge. h ps://doi.o g/10.4324/9781003309734-29
Col hea , M., Ras le, K., Pe y, C., Langdon, R., & Ziegle , J. (2001). DRC: A dual ou e
cascaded model o isual wo d ecogni ion and eading aloud. Psychological Re iew,
108(1), 204–256. h ps://doi.o g/10.1037/0033-295X.108.1.204
Comb isson, E., & Je bi, K. (2015). Exceeding chance le el by chance: The ca ea o
124 OF 150
heo e ical chance le els in b ain signal classi ica ion and s a is ical assessmen o
decoding accu acy. Jou nal o Neu oscience Me hods, 250, 126–136.
h ps://doi.o g/10.1016/j.jneume h.2015.01.010
Conan , L. L., Lieben hal, E., Desai, A., Seidenbe g, M. S., & Binde , J. R. (2020).
Di e en ial ac i a ion o he isual wo d o m a ea du ing audi o y phoneme
pe cep ion in you h wi h dyslexia. Neu opsychologia, 146, 107543.
h ps://doi.o g/10.1016/j.neu opsychologia.2020.107543
Cox, R. W. (1996). AFNI: So wa e o Analysis and Visualiza ion o Func ional Magne ic
Resonance Neu oimages. Compu e s and Biomedical Resea ch, 29(3), 162–173.
h ps://doi.o g/10.1006/cbm .1996.0014
Dale, A. M. (1999). Op imal expe imen al design o e en ‐ ela ed MRI. Human B ain
Mapping, 8, 109–114.
Damasio, A. R., & Damasio, H. (1983). The ana omic basis o pu e alexia. Neu ology,
33(12), 1573–1573. h ps://doi.o g/10.1212/WNL.33.12.1573
Dampu é, J., Ros, C., Roue , J., & Vibe , N. (2012). How Wo d Familia i y Facili a es Visual
Sea ch o Ve bal Ma e ial. Applied Cogni i e Psychology, 26(2), 271–288.
h ps://doi.o g/10.1002/acp.1821
de B uin, A., Ca ei as, M., & Duñabei ia, J. A. (2017). The BEST Da ase o Language
P o iciency. F on ie s in Psychology, 8. h ps://doi.o g/10.3389/ psyg.2017.00522
Dȩbska, A., Wójcik, M., Chyl, K., Dziȩgiel-Fi e , G., & Jedno óg, K. (2023). Beyond he
Visual Wo d Fo m A ea – a cogni i e cha ac e iza ion o he le en al
occipi o empo al co ex. F on ie s in Human Neu oscience, 17, 1199366.
h ps://doi.o g/10.3389/ nhum.2023.1199366
Dehaene, S., & Cohen, L. (2007). Cul u al Recycling o Co ical Maps. Neu on, 56(2), 384–
398. h ps://doi.o g/10.1016/j.neu on.2007.10.004
Dehaene, S., & Cohen, L. (2011). The unique ole o he isual wo d o m a ea in eading.
T ends in Cogni i e Sciences, 15(6), 254–262.
h ps://doi.o g/10.1016/j. ics.2011.04.003
125 OF 150
Dehaene, S., Le Clec’H, G., Poline, J.-B., Le Bihan, D., & Cohen, L. (2002). The isual wo d
o m a ea: A p elexical ep esen a ion o isual wo ds in he usi o m gy us:
Neu o epo , 13(3), 321–325. h ps://doi.o g/10.1097/00001756-200203040-00015
Dehaene, S., Naccache, L., Cohen, L., Bihan, D. L., Mangin, J.-F., Poline, J.-B., & Ri iè e,
D. (2001). Ce eb al mechanisms o wo d masking and unconscious epe i ion
p iming. Na u e Neu oscience, 4(7), 752–758. h ps://doi.o g/10.1038/89551
Déje ine, J. (1892). Con ibu ion a l’é ude ana omoclinique e clinique des di e en es
a ie es de ceci e e bal. Comp e Rendus Hebdomadai es Séances e Mémoi es de
La Socié é de Biologie, 4, 61–90.
Deniz, F., Tseng, C., Wehbe, L., Dup é La Tou , T., & Gallan , J. L. (2023). Seman ic
Rep esen a ions du ing Language Comp ehension A e A ec ed by Con ex . The
Jou nal o Neu oscience, 43(17), 3144–3158.
h ps://doi.o g/10.1523/JNEUROSCI.2459-21.2023
Desimone, R., Schein, S. J., Mo an, J., & Unge leide , L. G. (1985). Con ou , colo and
shape analysis beyond he s ia e co ex. Vision Resea ch, 25(3), 441–452.
h ps://doi.o g/10.1016/0042-6989(85)90069-0
De lin, J. T., Jamison, H. L., Gonne man, L. M., & Ma hews, P. M. (2006). The Role o he
Pos e io Fusi o m Gy us in Reading. Jou nal o Cogni i e Neu oscience, 18(6), 911–
922. h ps://doi.o g/10.1162/jocn.2006.18.6.911
DeWi , I., & Rauschecke , J. P. (2012). Phoneme and wo d ecogni ion in he audi o y
en al s eam. P oceedings o he Na ional Academy o Sciences, 109(8).
h ps://doi.o g/10.1073/pnas.1113427109
Diaz, M. T., Ka imi, H., T ou man, S. B. W., Ge el, V. H., Cosg o e, A. L., & Zhang, H.
(2021). Neu al sensi i i y o phonological cha ac e is ics is s able ac oss he li espan.
Neu oImage, 225, 117511. h ps://doi.o g/10.1016/j.neu oimage.2020.117511
Dick, A. S., Be nal, B., & T emblay, P. (2014). The Language Connec ome: New Pa hways,
New Concep s. The Neu oscien is , 20(5), 453–467.
h ps://doi.o g/10.1177/1073858413513502
126 OF 150
Diedenho en, B., & Musch, J. (2015). coco : A Comp ehensi e Solu ion o he S a is ical
Compa ison o Co ela ions. PLOS ONE, 10(4), e0121945.
h ps://doi.o g/10.1371/jou nal.pone.0121945
Ding, S., Van Hoesen, G. W., Cassell, M. D., & Po emba, A. (2009). Pa cella ion o human
empo al pola co ex: A combined analysis o mul iple cy oa chi ec onic,
chemoa chi ec onic, and pa hological ma ke s. Jou nal o Compa a i e Neu ology,
514(6), 595–623. h ps://doi.o g/10.1002/cne.22053
Dobbins, I. G., & Wagne , A. D. (2005). Domain-gene al and Domain-sensi i e P e on al
Mechanisms o Recollec ing E en s and De ec ing No el y. Ce eb al Co ex, 15(11),
1768–1778. h ps://doi.o g/10.1093/ce co /bhi054
Downing, P. E., Chan, A. W.-Y., Peelen, M. V., Dodds, C. M., & Kanwishe , N. (2006).
Domain Speci ici y in Visual Co ex. Ce eb al Co ex, 16(10), 1453–1461.
h ps://doi.o g/10.1093/ce co /bhj086
Duchon, A., Pe ea, M., Sebas ián-Gallés, N., Ma í, A., & Ca ei as, M. (2013). EsPal: One-
s op shopping o Spanish wo d p ope ies. Beha io Resea ch Me hods, 45(4),
1246–1258. h ps://doi.o g/10.3758/s13428-013-0326-1
Dziȩgiel-Fi e , G., Beck, J., & Jedno óg, K. (2023). The ole o he le en al
occipi o empo al co ex in speech p ocessing—The in luence o isual dep i a ion.
F on ie s in Human Neu oscience, 17, 1228808.
h ps://doi.o g/10.3389/ nhum.2023.1228808
Eps ein, R. A., & Bake , C. I. (2019). Scene Pe cep ion in he Human B ain. Annual Re iew
o Vision Science, 5(1), 373–397. h ps://doi.o g/10.1146/annu e - ision-091718-
014809
Fa ell, S., & Lewandowsky, S. (Eds.). (2018). Model Compa ison. In Compu a ional
Modeling o Cogni ion and Beha io (pp. 241–272). Camb idge Uni e si y P ess;
Camb idge Co e. h ps://doi.o g/10.1017/CBO9781316272503.011
Fedo enko, E., Duncan, J., & Kanwishe , N. (2012). Language-Selec i e and Domain-
Gene al Regions Lie Side by Side wi hin B oca’s A ea. Cu en Biology, 22(21),
127 OF 150
2059–2062. h ps://doi.o g/10.1016/j.cub.2012.09.011
Fedo enko, E., I ano a, A. A., & Rege , T. I. (2024). The language ne wo k as a na u al kind
wi hin he b oade landscape o he human b ain. Na u e Re iews Neu oscience,
25(5), 289–312. h ps://doi.o g/10.1038/s41583-024-00802-4
Fe nandes, M. A., Mosco i ch, M., Ziegle , M., & G ady, C. (2005). B ain egions associa ed
wi h success ul and unsuccess ul e ie al o e bal episodic memo y as e ealed by
di ided a en ion. Neu opsychologia, 43(8), 1115–1127.
h ps://doi.o g/10.1016/j.neu opsychologia.2004.11.026
Feye eisen, P., Van de Bo gh , F., & Se on, X. (1988). The ope a i i y e ec in naming: A
e-analysis. Neu opsychologia, 26(3), 401–415. h ps://doi.o g/10.1016/0028-
3932(88)90094-2
Fiebach, C. J., F iede ici, A. D., Mülle , K., & C amon, D. Y. on. (2002). MRI E idence o
Dual Rou es o he Men al Lexicon in Visual Wo d Recogni ion. Jou nal o Cogni i e
Neu oscience, 14(1), 11–23. h ps://doi.o g/10.1162/089892902317205285
Fische -Baum, S., B uggemann, D., Gallego, I. F., Li, D. S. P., & Tamez, E. R. (2017).
Decoding le els o ep esen a ion in eading: A ep esen a ional simila i y app oach.
Co ex, 90, 88–102. h ps://doi.o g/10.1016/j.co ex.2017.02.017
Fishe , R. A. (1921). On he p obable e o o a coe icien o co el- a ion deduced om a
small sample. Me on, 1, 3–32.
F ankland, S. M., & G eene, J. D. (2015). An a chi ec u e o encoding sen ence meaning in
le mid-supe io empo al co ex. P oceedings o he Na ional Academy o Sciences,
112(37), 11732–11737. h ps://doi.o g/10.1073/pnas.1421236112
F iede ici, A. D. (2009). Pa hways o language: Fibe ac s in he human b ain. T ends in
Cogni i e Sciences, 13(4), 175–181. h ps://doi.o g/10.1016/j. ics.2009.01.001
F iede ici, A. D. (2011). The B ain Basis o Language P ocessing: F om S uc u e o
Func ion. Physiological Re iews, 91(4), 1357–1392.
h ps://doi.o g/10.1152/phys e .00006.2011
F iede ici, A. D. (2012). The co ical language ci cui : F om audi o y pe cep ion o sen ence
134 OF 150
Sciences, 107(6), 2717–2722. h ps://doi.o g/10.1073/pnas.0907307107
Lau, E. F., Phillips, C., & Poeppel, D. (2008). A co ical ne wo k o seman ics:
(De)cons uc ing he N400. Na u e Re iews Neu oscience, 9(12), 920–933.
h ps://doi.o g/10.1038/n n2532
Le ma-Usabiaga, G., Ca ei as, M., & Paz-Alonso, P. M. (2018). Con e ging e idence o
unc ional and s uc u al seg ega ion wi hin he le en al occipi o empo al co ex in
eading. P oceedings o he Na ional Academy o Sciences, 115(42), E9981–E9990.
h ps://doi.o g/10.1073/pnas.1803003115
Lewis, G., & Poeppel, D. (2014). The ole o isual ep esen a ions du ing he lexical access
o spoken wo ds. B ain and Language, 134, 1–10.
h ps://doi.o g/10.1016/j.bandl.2014.03.008
Lewis-Peacock, J. A., & No man, K. A. (2014). Mul i oxel Pa e n Analysis o Func ional MRI
Da a. In M. S. Gazzaniga & G. R. Mangun (Eds.), The Cogni i e Neu osciences (5 h
ed.). The MIT P ess. h ps://doi.o g/10.7551/mi p ess/9504.003.0099
Lim, S. W. H. (2016). The In luence o O hog aphic Neighbo hood Densi y and Wo d
F equency on Visual Wo d Recogni ion: Insigh s om RT Dis ibu ional Analyses.
F on ie s in Psychology, 7. h ps://doi.o g/10.3389/ psyg.2016.00401
Liuzzi, A. G., Mee smans, K., Pee e s, R., De Deyne, S., Dupon , P., & Vandenbe ghe, R.
(2024). Seman ic ep esen a ions in in e io on al and la e al empo al co ex du ing
pic u e naming, eading, and epe i ion. Human B ain Mapping, 45(2), e26603.
h ps://doi.o g/10.1002/hbm.26603
Liuzzi, A. G., Mee smans, K., S o ms, G., De Deyne, S., Dupon , P., & Vandenbe ghe, R.
(2023). Independency o Coding o A ec i e Simila i ies and o Wo d Co-
occu ences in Tempo al Pe isyl ian Neoco ex. Neu obiology o Language, 4(2),
257–279. h ps://doi.o g/10.1162/nol_a_00095
López-Ba oso, D., & De Diego-Balague , R. (2017). Language Lea ning Va iabili y wi hin
he Do sal and Ven al S eams as a Cue o Compensa o y Mechanisms in Aphasia
Reco e y. F on ie s in Human Neu oscience, 11, 476.
135 OF 150
h ps://doi.o g/10.3389/ nhum.2017.00476
López-Ba oso, D., Thiebau De Scho en, M., Mo ais, J., Kolinsky, R., B aga, L. W.,
Gue ei o-Tauil, A., Dehaene, S., & Cohen, L. (2020). Impac o li e acy on he
unc ional connec i i y o ision and language ela ed ne wo ks. Neu oImage, 213,
116722. h ps://doi.o g/10.1016/j.neu oimage.2020.116722
Lude sdo e , P., Wimme , H., Richlan, F., Schu z, M., Hu zle , F., & K onbichle , M. (2016).
Le en al occipi o empo al ac i a ion du ing o hog aphic and seman ic p ocessing
o audi o y wo ds. Neu oImage, 124, 834–842.
h ps://doi.o g/10.1016/j.neu oimage.2015.09.039
MacG ego , L. J., Pul e mülle , F., Van Cas e en, M., & Sh y o , Y. (2012). Ul a- apid
access o wo ds in he b ain. Na u e Communica ions, 3(1), 711.
h ps://doi.o g/10.1038/ncomms1715
Malcolm, G. L., G oen, I. I. A., & Bake , C. I. (2016). Making Sense o Real-Wo ld Scenes.
T ends in Cogni i e Sciences, 20(11), 843–856.
h ps://doi.o g/10.1016/j. ics.2016.09.003
Maliko ic, A., Amun s, K., Schleiche , A., Mohlbe g, H., Kujo ic, M., Palome o-Gallaghe , N.,
Eickho , S. B., & Zilles, K. (2016). Cy oa chi ec u e o he human la e al occipi al
co ex: Mapping o wo ex as ia e a eas hOc4la and hOc4lp. B ain S uc u e and
Func ion, 221(4), 1877–1897. h ps://doi.o g/10.1007/s00429-015-1009-8
Ma in, A. (2007). The Rep esen a ion o Objec Concep s in he B ain. Annual Re iew o
Psychology, 58(1), 25–45. h ps://doi.o g/10.1146/annu e .psych.57.102904.190143
Ma in, A., Haxby, J. V., Lalonde, F. M., Wiggs, C. L., & Unge leide , L. G. (1995). Disc e e
Co ical Regions Associa ed wi h Knowledge o Colo and Knowledge o Ac ion.
Science, 270(5233), 102–105. h ps://doi.o g/10.1126/science.270.5233.102
Ma in, C. B., & Ba ense, M. D. (2023). Pe cep ion and Memo y in he Ven al Visual S eam
and Medial Tempo al Lobe. Annual Re iew o Vision Science, 9(1), 409–434.
h ps://doi.o g/10.1146/annu e - ision-120222-014200
Ma in, C. B., McLean, D. A., O’Neil, E. B., & Kohle , S. (2013). Dis inc Familia i y-Based
136 OF 150
Response Pa e ns o Faces and Buildings in Pe i hinal and Pa ahippocampal
Co ex. Jou nal o Neu oscience, 33(26), 10915–10923.
h ps://doi.o g/10.1523/JNEUROSCI.0126-13.2013
Mazaika, P. K., Hoe , F., Glo e , G. H., & Reiss, A. L. (2009). Me hods and So wa e o
MRI Analysis o Clinical Subjec s. Human B ain Mapping Con e ence, Flo ence,
I aly.
McCandliss, B. D., Cohen, L., & Dehaene, S. (2003). The isual wo d o m a ea: Expe ise
o eading in he usi o m gy us. T ends in Cogni i e Sciences, 7(7), 293–299.
h ps://doi.o g/10.1016/S1364-6613(03)00134-7
McClelland, J. L., & Roge s, T. T. (2003). The pa allel dis ibu ed p ocessing app oach o
seman ic cogni ion. Na u e Re iews Neu oscience, 4(4), 310–322.
h ps://doi.o g/10.1038/n n1076
McClelland, J. L., & Rumelha , D. E. (1981). An in e ac i e ac i a ion model o con ex
e ec s in le e pe cep ion: I. An accoun o basic indings. Psychological Re iew,
88(5), 375–407. h ps://doi.o g/10.1037/0033-295X.88.5.375
Mee smans, K., B u ae s, R., Jamoulle, T., Liuzzi, A. G., De Deyne, S., S o ms, G., Dupon ,
P., & Vandenbe ghe, R. (2020). Rep esen a ion o associa i e and a ec i e seman ic
simila i y o abs ac wo ds in he la e al empo al pe isyl ian language egions.
Neu oImage, 217, 116892. h ps://doi.o g/10.1016/j.neu oimage.2020.116892
Mee smans, K., S o ms, G., De Deyne, S., B u ae s, R., Dupon , P., & Vandenbe ghe, R.
(2022). O ien ing o di e en dimensions o wo d meaning al e s he ep esen a ion o
wo d meaning in ea ly p ocessing egions. Ce eb al Co ex, 32(15), 3302–3317.
Scopus. h ps://doi.o g/10.1093/ce co /bhab416
Meme o a, K. S., Knyaze a, V. M., S anke ich, L. N., Malanchuk, I. G., & Aleksand o , A. A.
(2024). Func ional Connec i i y o B ain Regions Resul ing om Lea ning Un amilia
Wo ds: Wo d F equency E ec . In A. V. Samsono ich & T. Liu (Eds.), Biologically
Inspi ed Cogni i e A chi ec u es 2023 (Vol. 1130, pp. 595–603). Sp inge Na u e
Swi ze land. h ps://doi.o g/10.1007/978-3-031-50381-8_63
137 OF 150
Mikolo , T., Su ske e , I., Chen, K., Co ado, G., & Dean, J. (2013). Dis ibu ed
Rep esen a ions o Wo ds and Ph ases and hei Composi ionali y.
h ps://doi.o g/10.48550/ARXIV.1310.4546
Mishkin, M., Unge leide , L. G., & Macko, K. A. (1983). Objec ision and spa ial ision: Two
co ical pa hways. T ends in Neu osciences, 6, 414–417.
h ps://doi.o g/10.1016/0166-2236(83)90190-X
Mi chell, T. M., Shinka e a, S. V., Ca lson, A., Chang, K.-M., Mala e, V. L., Mason, R. A., &
Jus , M. A. (2008). P edic ing Human B ain Ac i i y Associa ed wi h he Meanings o
Nouns. Science, 320(5880), 1191–1195. h ps://doi.o g/10.1126/science.1152876
Mk ychian, N., Blago ech chenski, E., Ku makae a, D., Gnedykh, D., Kos omina, S., &
Sh y o , Y. (2019). Conc e e s. Abs ac Seman ics: F om Men al Rep esen a ions
o Func ional B ain Mapping. F on ie s in Human Neu oscience, 13, 267.
h ps://doi.o g/10.3389/ nhum.2019.00267
Moe el, M., De Ma ino, F., & Fo misano, E. (2012). P ocessing o Na u al Sounds in Human
Audi o y Co ex: Tono opy, Spec al Tuning, and Rela ion o Voice Sensi i i y. The
Jou nal o Neu oscience, 32(41), 14205–14216.
h ps://doi.o g/10.1523/JNEUROSCI.1388-12.2012
Moo e, C. J., & P ice, C. J. (1999). Th ee Dis inc Ven al Occipi o empo al Regions o
Reading and Objec Naming. Neu oImage, 10(2), 181–192.
h ps://doi.o g/10.1006/nimg.1999.0450
Mo ales, J., Lau, H., & Fleming, S. M. (2018). Domain-Gene al and Domain-Speci ic
Pa e ns o Ac i i y Suppo ing Me acogni ion in Human P e on al Co ex. The
Jou nal o Neu oscience, 38(14), 3534–3546.
h ps://doi.o g/10.1523/JNEUROSCI.2360-17.2018
Moul on, E., Bouhali, F., Monzal o, K., Poupon, C., Zhang, H., Dehaene, S., Dehaene-
Lambe z, G., & Dubois, J. (2019). Connec i i y be ween he isual wo d o m a ea
and he pa ie al lobe imp o es a e he i s yea o eading ins uc ion: A longi udinal
MRI s udy in child en. B ain S uc u e and Func ion, 224(4), 1519–1536.
138 OF 150
h ps://doi.o g/10.1007/s00429-019-01855-3
Muelle , C. J., Whi e, C. N., & Kuchinke, L. (2017). Elec ophysiological co ela es o he d i
di usion model in isual wo d ecogni ion. Human B ain Mapping, 38(11), 5616–
5627. h ps://doi.o g/10.1002/hbm.23753
Mum o d, J. A., Da is, T., & Pold ack, R. A. (2014). The impac o s udy design on pa e n
es ima ion o single- ial mul i a ia e pa e n analysis. Neu oImage, 103, 130–138.
h ps://doi.o g/10.1016/j.neu oimage.2014.09.026
Mye s, C. E., In e ian, A., & Mous a a, A. A. (2022). A p ac ical in oduc ion o using he d i
di usion model o decision-making in cogni i e psychology, neu oscience, and heal h
sciences. F on ie s in Psychology, 13, 1039172.
h ps://doi.o g/10.3389/ psyg.2022.1039172
Nas , S., Liu, N., De aney, K. J., Yue, X., Rajimeh , R., Unge leide , L. G., & Too ell, R. B.
H. (2011). Scene-Selec i e Co ical Regions in Human and Nonhuman P ima es. The
Jou nal o Neu oscience, 31(39), 13771–13785.
h ps://doi.o g/10.1523/JNEUROSCI.2792-11.2011
Ne eu, A., & Kaushanskaya, M. (2023). Pai ed-associa e e sus c oss-si ua ional: How do
e bal wo king memo y and wo d amilia i y a ec wo d lea ning? Memo y &
Cogni ion, 51(7), 1670–1682. h ps://doi.o g/10.3758/s13421-023-01421-7
No d , M., Gomez, J., Na u, V., Jeska, B., Ba ne , M., & G ill-Spec o , K. (2019). Lea ning o
Read Inc eases he In o ma i eness o Dis ibu ed Ven al Tempo al Responses.
Ce eb al Co ex, 29(7), 3124–3139. h ps://doi.o g/10.1093/ce co /bhy178
No d , M., Gomez, J., Na u, V. S., Rezai, A. A., Finzi, D., Kula , H., & G ill-Spec o , K.
(2021). Co ical ecycling in high-le el isual co ex du ing childhood de elopmen .
Na u e Human Beha iou , 5(12), 1686–1697. h ps://doi.o g/10.1038/s41562-021-
01141-5
No d , M., Gomez, J., Na u, V. S., Rezai, A. A., Finzi, D., Kula , H., & G ill-Spec o , K.
(2023). Longi udinal de elopmen o ca ego y ep esen a ions in en al empo al
co ex p edic s wo d and ace ecogni ion. Na u e Communica ions, 14(1), 8010.
139 OF 150
h ps://doi.o g/10.1038/s41467-023-43146-w
No man, K. A., Polyn, S. M., De e, G. J., & Haxby, J. V. (2006). Beyond mind- eading: Mul i-
oxel pa e n analysis o MRI da a. T ends in Cogni i e Sciences, 10(9), 424–430.
h ps://doi.o g/10.1016/j. ics.2006.07.005
OECD. (2023). PISA 2022 Resul s (Volume I): The S a e o Lea ning and Equi y in
Educa ion. OECD. h ps://doi.o g/10.1787/53 23881-en
Okazawa, G., Tajima, S., & Koma su, H. (2016). G adual De elopmen o Visual Tex u e-
Selec i e P ope ies Be ween Macaque A eas V2 and V4. Ce eb al Co ex, 4867–
4880. h ps://doi.o g/10.1093/ce co /bhw282
Oli e , M., Ca ei as, M., & Paz-Alonso, P. M. (2017). Func ional Dynamics o Do sal and
Ven al Reading Ne wo ks in Bilinguals. Ce eb al Co ex, 27(12), 5431–5443.
h ps://doi.o g/10.1093/ce co /bhw310
Olson, I. R., McCoy, D., Klobusicky, E., & Ross, L. A. (2013). Social cogni ion and he
an e io empo al lobes: A e iew and heo e ical amewo k. Social Cogni i e and
A ec i e Neu oscience, 8(2), 123–133. h ps://doi.o g/10.1093/scan/nss119
Pai io, A. (2010). Dual coding heo y and he men al lexicon. The Men al Lexicon, 5(2), 205–
230. h ps://doi.o g/10.1075/ml.5.2.04pai
Pai io, A. (2014). In elligence, dual coding heo y, and he b ain. In elligence, 47, 141–158.
h ps://doi.o g/10.1016/j.in ell.2014.09.002
Palme , S. D., MacG ego , L. J., & Ha elka, J. (2013). Conc e eness e ec s in single-
meaning, mul i-meaning and newly acqui ed wo ds. B ain Resea ch, 1538, 135–150.
h ps://doi.o g/10.1016/j.b ain es.2013.09.015
Papi o, G., F iede ici, A. D., & Zacca ella, E. (2020). The opog aphical o ganiza ion o
mo o p ocessing: An ALE me a-analysis on six ac ion domains and he ele ance o
B oca’s egion. Neu oImage, 206, 116321.
h ps://doi.o g/10.1016/j.neu oimage.2019.116321
Pasupa hy, A., Popo kina, D. V., & Kim, T. (2020). Visual Func ions o P ima e A ea V4.
Annual Re iew o Vision Science, 6(1), 363–385. h ps://doi.o g/10.1146/annu e -
140 OF 150
ision-030320-041306
Pa amadilok, C., Plan on, S., & Bonna d, M. (2019). Spoken language coding neu ons in
he Visual Wo d Fo m A ea: E idence om a TMS adap a ion pa adigm.
Neu oImage, 186, 278–285. h ps://doi.o g/10.1016/j.neu oimage.2018.11.014
Pa e son, K., Nes o , P. J., & Roge s, T. T. (2007). Whe e do you know wha you know?
The ep esen a ion o seman ic knowledge in he human b ain. Na u e Re iews
Neu oscience, 8(12), 976–987. h ps://doi.o g/10.1038/n n2277
Paz-Alonso, P. M., Oli e , M., Le ma-Usabiaga, G., Caballe o-Gaudes, C., Quiñones, I.,
Suá ez-Coalla, P., Duñabei ia, J. A., Cue os, F., & Ca ei as, M. (2018). Neu al
co ela es o phonological, o hog aphic and seman ic eading p ocessing in dyslexia.
Neu oImage: Clinical, 20, 433–447. h ps://doi.o g/10.1016/j.nicl.2018.08.018
Peelen, M. V., & Downing, P. E. (2005). Selec i i y o he Human Body in he Fusi o m
Gy us. Jou nal o Neu ophysiology, 93(1), 603–608.
h ps://doi.o g/10.1152/jn.00513.2004
Pe ei a, F., Bo inick, M., & De e, G. (2013). Using Wikipedia o lea n seman ic ea u e
ep esen a ions o conc e e concep s in neu oimaging expe imen s. A i icial
In elligence, 194, 240–252. h ps://doi.o g/10.1016/j.a in .2012.06.005
Pe e sson, K. (2004). A i icial syn ac ic iola ions ac i a e B oca?s egion. Cogni i e
Science, 28(3), 383–407. h ps://doi.o g/10.1016/j.cogsci.2003.12.003
Plan on, S., Chanoine, V., Sein, J., An on, J.-L., Naza ian, B., Pallie , C., & Pa amadilok, C.
(2019). Top-down ac i a ion o he isuo-o hog aphic sys em du ing spoken
sen ence p ocessing. Neu oImage, 202, 116135.
h ps://doi.o g/10.1016/j.neu oimage.2019.116135
Plau , D. C., McClelland, J. L., Seidenbe g, M. S., & Pa e son, K. (1996). Unde s anding
no mal and impai ed wo d eading: Compu a ional p inciples in quasi- egula
domains. Psychological Re iew, 103(1), 56–115. h ps://doi.o g/10.1037/0033-
295X.103.1.56
Pohl, W. (1973). Dissocia ion o spa ial disc imina ion de ici s ollowing on al and pa ie al
141 OF 150
lesions in monkeys. Jou nal o Compa a i e and Physiological Psychology, 82(2),
227–239. h ps://doi.o g/10.1037/h0033922
Pona i, M., No bu y, C. F., & Vigliocco, G. (2018). Acquisi ion o abs ac concep s is
in luenced by emo ional alence. De elopmen al Science, 21(2), e12549.
h ps://doi.o g/10.1111/desc.12549
Popal, H., Wang, Y., & Olson, I. R. (2019). A Guide o Rep esen a ional Simila i y Analysis
o Social Neu oscience. Social Cogni i e and A ec i e Neu oscience, 14(11), 1243–
1253. h ps://doi.o g/10.1093/scan/nsz099
P ice, C. J., & De lin, J. T. (2003). The my h o he isual wo d o m a ea. Neu oImage,
19(3), 473–481. h ps://doi.o g/10.1016/S1053-8119(03)00084-3
P ice, C. J., & De lin, J. T. (2011). The In e ac i e Accoun o en al occipi o empo al
con ibu ions o eading. T ends in Cogni i e Sciences, 15(6), 246–253.
h ps://doi.o g/10.1016/j. ics.2011.04.001
P o opapas, A., O anidou, E., Taylo , J. S. H., Ka a asilis, E., Kapnoula, E. C.,
Panagio a opoulou, G., Velonakis, G., Poulou, L. S., Smy nis, N., & Kelekis, D.
(2016). E alua ing cogni i e models o isual wo d ecogni ion using MRI: E ec s o
lexical and sublexical a iables. Neu oImage, 128, 328–341.
h ps://doi.o g/10.1016/j.neu oimage.2016.01.013
Pugh, K. R., Mencl, W. E., Jenne , A. R., Ka z, L., F os , S. J., Lee, J. R., Shaywi z, S. E., &
Shaywi z, B. A. (2001). Neu obiological s udies o eading and eading disabili y.
Jou nal o Communica ion Diso de s, 34(6), 479–492. h ps://doi.o g/10.1016/S0021-
9924(01)00060-0
Ralph, M. A. L., Je e ies, E., Pa e son, K., & Roge s, T. T. (2017). The neu al and
compu a ional bases o seman ic cogni ion. Na u e Re iews Neu oscience, 18(1),
42–55. h ps://doi.o g/10.1038/n n.2016.150
Ra cli , R. (1978). A heo y o memo y e ie al. Psychological Re iew, 85(2), 59–108.
h ps://doi.o g/10.1037/0033-295X.85.2.59
Ra cli , R., & McKoon, G. (2008). The Di usion Decision Model: Theo y and Da a o Two-
142 OF 150
Choice Decision Tasks. Neu al Compu a ion, 20(4), 873–922.
h ps://doi.o g/10.1162/neco.2008.12-06-420
Ra cli , R., Smi h, P. L., B own, S. D., & McKoon, G. (2016). Di usion Decision Model:
Cu en Issues and His o y. T ends in Cogni i e Sciences, 20(4), 260–281.
h ps://doi.o g/10.1016/j. ics.2016.01.007
Rauschecke , J. P. (1998). Pa allel P ocessing in he Audi o y Co ex o P ima es. Audiology
and Neu o ology, 3(2–3), 86–103. h ps://doi.o g/10.1159/000013784
Richa dson, F. M., Seghie , M. L., Le , A. P., Thomas, M. S. C., & P ice, C. J. (2011).
Mul iple Rou es om Occipi al o Tempo al Co ices du ing Reading. Jou nal o
Neu oscience, 31(22), 8239–8247. h ps://doi.o g/10.1523/JNEUROSCI.6519-
10.2011
Riesenhube , M., & Gleze , L. S. (2017). E idence o apid localis plas ici y in he en al
isual s eam: The example o wo ds. Language, Cogni ion and Neu oscience, 32(3),
286–294. h ps://doi.o g/10.1080/23273798.2016.1210178
Rissman, J., Gazzaley, A., & D’Esposi o, M. (2004). Measu ing unc ional connec i i y du ing
dis inc s ages o a cogni i e ask. Neu oImage, 23(2), 752–763.
h ps://doi.o g/10.1016/j.neu oimage.2004.06.035
Ri chie, J. B., Zeman, A. A., Bosmans, J., Sun, S., Ve haegen, K., & Op De Beeck, H. P.
(2021). Un angling he Animacy O ganiza ion o Occipi o empo al Co ex. The
Jou nal o Neu oscience, 41(33), 7103–7119.
h ps://doi.o g/10.1523/JNEUROSCI.2628-20.2021
Roe, A. W., & Ts’o, D. Y. (2015). Speci ici y o V1–V2 o ien a ion ne wo ks in he p ima e
isual co ex. Co ex, 72, 168–178. h ps://doi.o g/10.1016/j.co ex.2015.07.007
Roge , E., Banjac, S., Thiebau de Scho en, M., & Baciu, M. (2022). Missing links: The
unc ional uni ica ion o language and memo y (L∪M). Neu oscience & Biobeha io al
Re iews, 133, 104489. h ps://doi.o g/10.1016/j.neubio e .2021.12.012
Roge , E., Rod igues De Almeida, L., Loe enb uck, H., Pe one-Be olo i, M., Cousin, E.,
Schwa z, J. L., Pe ie , P., Dohen, M., Vilain, A., Ba aduc, P., Acha d, S., & Baciu,
143 OF 150
M. (2022). Un a eling he unc ional a ibu es o he language connec ome: C ucial
subne wo ks, lexibili y and a iabili y. Neu oImage, 263.
h ps://doi.o g/10.1016/j.neu oimage.2022.119672
Rosenbe g, M. D., Finn, E. S., Cons able, R. T., & Chun, M. M. (2015). P edic ing momen -
o-momen a en ional s a e. Neu oImage, 114, 249–256.
h ps://doi.o g/10.1016/j.neu oimage.2015.03.032
Rueckl, J. G., Paz-Alonso, P. M., Mol ese, P. J., Kuo, W.-J., Bick, A., F os , S. J., Hancock,
R., Wu, D. H., Mencl, W. E., Duñabei ia, J. A., Lee, J.-R., Oli e , M., Ze in, J. D.,
Hoe , F., Ca ei as, M., Tzeng, O. J. L., Pugh, K. R., & F os , R. (2015). Uni e sal
b ain signa u e o p o icien eading: E idence om ou con as ing languages.
P oceedings o he Na ional Academy o Sciences, 112(50), 15510–15515.
h ps://doi.o g/10.1073/pnas.1509321112
Rundle, M. M., Coch, D., Connolly, A. C., & G ange , R. H. (2018). Dissocia ing equency
and animacy e ec s in isual wo d p ocessing: An MRI s udy. B ain and Language,
183, 54–63. h ps://doi.o g/10.1016/j.bandl.2018.05.005
Saenz, M., & Lange s, D. R. M. (2014). Tono opic mapping o human audi o y co ex.
Hea ing Resea ch, 307, 42–52. h ps://doi.o g/10.1016/j.hea es.2013.07.016
Salminen-Vapa an a, N., Koi is o, M., No eika, V., Vanni, S., & Re onsuo, A. (2012).
Neu ona iga ed ansc anial magne ic s imula ion sugges s ha a ea V2 is necessa y
o isual awa eness. Neu opsychologia, 50(7), 1621–1627.
h ps://doi.o g/10.1016/j.neu opsychologia.2012.03.015
Sánchez, A., Ca ei as, M., & Paz-Alonso, P. M. (2023). Wo d equency and eading
demands modula e b ain ac i a ion in he in e io on al gy us. Scien i ic Repo s,
13(1), 17217. h ps://doi.o g/10.1038/s41598-023-44420-z
Sandak, R., Mencl, W. E., F os , S. J., & Pugh, K. R. (2004). The Neu obiological Basis o
Skilled and Impai ed Reading: Recen Findings and New Di ec ions. Scien i ic
S udies o Reading, 8(3), 273–292. h ps://doi.o g/10.1207/s1532799xss 0803_6
Sapoun zis, P., Schluppeck, D., Bow ell, R., & Pei ce, J. W. (2010). A compa ison o MRI
