Behavioral Predictors of Right Hemisphere Language Dominance

BEHAVIORAL PREDICTORS

OF RIGHT HEMISPHERE

LANGUAGE DOMINANCE

Word Count: 16,574

Pieter De Clercq

Promotor: Prof. Dr. Guy Vingerhoets, Co-Promotor: PhD Candidate Robin Gerrits

A dissertation submitted to Ghent University in partial fulfilment of the requirements for the degree of Master of Science in Psychology (Theoretical and Experimental Psychology)

Acknowledgements

First and foremost, I would like to thank the co-promotor of this thesis, PhD student Robin Gerrits. Robin guided me through this project and gave useful feedback to improve my thesis. He further invited me to fMRI scan sessions at Ghent University Hospital which captivated me a lot. During these sessions, we had some interesting academic discussions and personal conversations. But Robin was more than a mentor for my thesis; he further prepared me for my job interview and personally assisted me with the presentation I was asked to give at the application. This is something I will never forget. So, for all of that, a very special thanks, Robin. Congratulations on the doctoral dissertation you are about to submit, and all the best in the United States as you embark on your post-doc. I am sure you will do great.

Second, a warm thanks to promotor Prof. Dr. Guy Vingerhoets for giving me the opportunity to write my thesis at your lab. Thanks for appointments to discuss progressions in my thesis and for brainstorming with Robin and myself on concrete elaborations of the project.

What’s an ‘acknowledgements’-section without mentioning the parents, of course. Thanks to you, mom and dad, for supporting me through thick and thin, providing me an emotionally stable environment to embark on my studies and to back my decision to switch to experimental psychology in third year organizational psychology. I am very grateful I was given the freedom to choose my Master’s study of interest and could not have done it without your support.

Finally, to my girlfriend Astrid, a special thanks (and also, a heartfelt sorry) for participating in almost every study I ran during my education. Thanks for your interest in my research projects and for the support you offered. Astrid was one of the participants we included in the fMRI sample, and for clarity, she was found typical left hemisphere language dominant. She was however one of the participants considered a “false positive”, which the reader will learn all about in this paper.

Preamble concerning COVID-19

Corona measures did not have an impact on the elaboration of the current thesis. Data collection was completed before the measures of Ghent University were implemented.

Abstract

One of the most intriguing observations in the study of human brain organization, is that language is strongly lateralized to the left hemisphere in the vast majority of the population. A rare group of (mainly) left-handers however deviates from this pattern and exhibit atypical, right hemisphere language dominance (RLD). We argue that research in large samples of RLD individuals is essential to address many outstanding research questions at present. However, the low prevalence of RLD in the human population (10-20% of left-handers only) constitutes a serious challenge for researchers to recruit these participants. Scanning a large group of left-handers would result in a relative small subgroup of RLD individuals. This is particularly problematic considering the high financial cost of the standard method to determine hemisphere language dominance, functional Magnetic Resonance Imaging (fMRI). The present study therefore assessed the predictive value of a low-cost behavioral screening method for determining language dominance: the Visual Half Field (VHF) paradigm. Laterality indices (LIs) calculated on the basis of latencies and accuracies in a VHF word naming task were compared to fMRI LIs in a large sample of left-handers (n=63, of which 24 were found RLD). Although we report no one-on-one relationship between VHF predictions and actual hemisphere language dominance, we demonstrated through various in- and out-sample analyses that the VHF task is a valid and useful screening tool. Inviting participants to an fMRI study on the basis of VHF predictions substantially increases the chance of recruiting actual RLD individuals. Finally, we provide recommendations for future researchers taking into account financial and logistical constraints.

Abstract in Dutch

Grootschalige studies in zowel gezonde als patiënten populaties de voorbije 2 eeuwen, hebben vastgesteld dat taal, als cognitieve functie, sterk gelateraliseerd is naar de linker hemisfeer bij de grote meerderheid. Een zeldzame subgroep, voornamelijk bestaande uit linkshandigen, devieert van dit patroon en vertoont atypisch, rechts-hemisferische taaldominantie (RLD). Onderzoek in grote samples van RLD individuen is essentieel om openstaande onderzoeksvragen omtrent de principes en grondslagen van de functionele en structurele organisatie van het menselijk brein te adresseren. De lage prevalentie van RLD (ongeveer 10% van linkshandigen) vormt helaas een serieus obstakel om grote samples bij elkaar te krijgen. Dit zeker in licht van de hoge gebruiks- en operationele kosten van de standaardmethode om taallateralisatie vast te stellen, functional Magnetic Resonance Imaging (fMRI). De huidige thesis heeft daarom de validiteit en predictie-accuraatheid van een goedkoop, gedragsmatig screeningsobject verder onderzocht: de Visual Half Field (VHF) taak. Lateraliteits-indicaties (LIs) van taaldominantie o.b.v. fMRI en LIs op basis van de VHF taak werden vergeleken en gecorreleerd in een grote sample linkshandigen (n=63, waarvan 24 RLD individuen werden gerekruteerd). Wij rapporteren geen één-op-één relatie tussen predicties o.b.v. de VHF taak en werkelijke taaldominantie, niettegenstaande we wel concluderen dat voorspellingen op basis van de VHF taak de kans om daadwerkelijk een groter aantal RLD individuen over te houden in de finale sample, drastisch verhoogt. Participanten op voorhand scannen op RLD met de VHF taak, is dus een zeer efficiënte en daadkrachtige techniek om een grotere sample RLD participanten te rekruteren.

TABLE OF CONTENTS

INTRODUCTION ... 1

LITERATURE OVERVIEW ... 2

1. GENERAL CONCEPTS ... 2

2. LANGUAGE DOMINANCE... 3

3. ATYPICAL LANGUAGE DOMINANCE ... 3

4. METHODS TO DETERMINE LANGUAGE DOMINANCE ... 8

5. PRESENT STUDY ... 14

METHOD ... 15

1. PARTICIPANTS ... 15

2. BEHAVIORAL SCREENING ... 16

3. FMRI SCANNING ... 19

4. PREDICTORS OF LANGUAGE DOMINANCE: DATA ANALYSIS ... 20

RESULTS ... 23

1. CORRELATIONS BETWEEN BEHAVIORAL PREDICTORS ... 24

2. DEGREE OF HANDEDNESS AND LANGUAGE DOMINANCE ... 25

3. VHF TASK AND LANGUAGE DOMINANCE ... 26

4. INTERACTION EFFECT VHF TASK AND DEGREE OF HANDEDNESS ... 32

5. OUTSIDE PREDICTION: CROSS-VALIDATION ... 33

DISCUSSION ... 35

1 Introduction

The human brain is a remarkably complex organ comprised of billions of neurons that intercommunicate by forming countless connections and synapses (Jawabri & Sharma, 2020). Roughly, the outer layer of the brain consists of gray matter (mainly composed of numerous cell bodies and relatively few myelinated axons), while the inner layer consists of white matter made up of myelinated axons which are organized into bundles (Blumenfield, 2010; Jawabri & Sharma, 2020). Large sulci or fissures divide the brain into lobes and also into the two cerebral hemispheres (left and right) as the deep longitudinal fissure. White matter tracts secure intra- and interhemispheric communication throughout these distinct regions.

The left and right hemisphere of the brain are believed to function differently yet interdependently (Springer & Deutsch, 2000). Sensory information that enters one halve of the brain travels across the corpus callosum to the other halve (Friedrich et al., 2020; Tzourio-Mazoyer, 2016). Although both hemispheres have a role to play in the processing of information, a single hemisphere commonly dominates the execution of certain tasks or functions (Corballis, 2014; Mendoza, 2011). Epidemiologic research further indicated that the human brain favors a prototypical segregation pattern between left and right across these functions (Mendoza, 2011). Functional asymmetries between the left and right hemisphere has evoked pervasive fascination in academic discussions and remain the subject of principal research questions on human brain organization at present.

In this thesis, I will summarize the literature on functional asymmetries with a special focus on language. A minority of the population deviates from the typical observed leftward hemispheric language dominance (i.e. they exhibit atypical, rightward language dominance). I will emphasize why this is an interesting, and foremost, an important group to study. Research in sufficiently large samples of atypical language dominant individuals could help clarifying various issues in brain organization research. However, I will claim that this group of people are rare and difficult to recruit.

Functional Magnetic Resonance Imaging (fMRI) is currently considered the standard method in the assessment of hemispheric specialization. However, setting up and running an fMRI study is not all easy and hassle free. The availability of MRI scanners is limited in its availability in the wider research community, collecting data is

2 quite expensive and data analyses can be quite complex and time-consuming. The current

thesis assesses and validates a widely used behavioral alternative to fMRI: The Visual Half Field (VHF) paradigm. The VHF task is cheap and easy both in experiment setup and knowhow to handle the data afterwards. It is furthermore deployable on a large sample of participants. It could therefore be beneficial for future researchers first to test a large sample of individuals on atypical (rightward) language dominance using VHF tasks and later recruit those individuals for testing with the scanner. The current study therefore assessed the usefulness of the VHF paradigm as a first screening tool to detect right hemisphere language dominance. The results are discussed and placed in the broader literature on atypical language dominance and its assessment methods. Some useful recommendations for future researchers are made.

Literature Overview 1. General Concepts

The human brain favors a prototypical hemispheric specialization at the population level (Vingerhoets, 2019). Hemispheric specialization refers to the characteristic that each cerebral hemisphere mediates different aspects of behavior. This specialization is relative, because with a few exceptions, both hemispheres can process most types of information. However, they do so in fundamentally different manners and with significant trade-offs in efficiency (i.e. specialization) (Mendoza, 2011). Moreover, neuroimaging studies demonstrate very distinct patterns of brain activations between left and right in response to specific cognitive tasks. When one hemisphere is more heavily involved in a specific function, it is often referred to as being dominant (Bear, Connors, & Paradiso, 2007). A mental or cognitive function for which asymmetrical activation or specialization is found, is called a lateralized function. The prototypical pattern of all lateralized functions that is observed in the population, is what we refer to as typical

functional brain segregation.

Epidemiological research reported strong population biases in language (Carey & Johnstone, 2014) and praxis (Vingerhoets et al., 2013) favoring the left hemisphere, and spatial attention (Shulman et al., 2010), face recognition (Júnior, Marinho de Sousa, & Fukusima, 2014) and emotional prosody (Patel et al., 2018) favoring the right hemisphere. The division in labor between left and right is suggested to reduce interferences between

3 competing neural processes, allows for more efficient behavior and eliminates redundant

duplication (Lévy, 1977; Vallortigara, 2000).

2. Language Dominance

When in 1865, the French Physician Paul Broca declared that speech production was controlled by the left hemisphere, he broke major ground in the field of neuroscience (Broca, 1865). Broca’s conclusions were drawn from patients who suffered from aphasia, an acquired language disorder caused by brain damage in the left hemisphere. Later however, Broca reported some cases of aphasia due to right lateralized brain damage. Therefore, he slightly departed from his strong claim, and suggested that the left hemisphere is specialized in speech production in the vast majority of the population. In his doctoral dissertation, Carl Wernicke (1874) provided useful data and theory to explain previously reported symptoms of patients who could produce speech, but suffered from clear sensory problems (Lordat, 1843). Broca’s left hemispheric frontal region specialized in producing speech, and Wernicke’s left hemispheric temporo-parietal region specialized in language comprehension, are presently well-known as the Broca and Wernicke areas. Over the past years however, research indicated that 1) these areas are not unique to language, but play a role in other cognitive functions as well and that 2) language production and comprehension rely on a large-scale network of cortical and subcortical areas beyond Broca and Wernicke (Tremblay & Dick, 2016).

3. Atypical Language Dominance

Early studies on language lateralization relied on research in patients with brain damage (Critchley, 1954). Recent neuroimaging techniques however enabled the determination of hemisphere dominance in neurologically healthy individuals. It thereby brought a new impetus to the study of language lateralization, and brain organization more generally. FMRI is currently considered the standard method in the determination of hemisphere language dominance. The technique provides direct spatial localization of cortical areas involved in language processing by using changes in cerebral blood flow as a proxy measure for neural activation (Li, Newton, Anderson, Ding, & Gore, 2019)

Over the past decades, neuroimaging research further demonstrated that the degree of language lateralization differs from person to person (Goldie, 2016). FMRI

4 captures this variability through Laterality Indices (LIs), which are computed from

differential activation (blood flow) between the left and right hemisphere, and typically range from -100 to +100 (or -1 to +1). An index of -100 (-1) expresses complete lateralization to the left, +100 (or +1) to the right. A participant is then categorized as ‘Left Language Dominant’ (LLD) or ‘Right Language Dominant’ (RLD) when it exceeds a certain cut-off point. Bilateral representation of language (BLR) is assumed when no obvious hemispheric dominance is apparent. ‘Atypical language dominance’ is used as an overarching term for RLD and BLR individuals. Since the cut-off points are based on arbitrary choices of the researcher, prudence is recommended in the classification of BLR individuals as atypical language dominant. There is no common definition to separate BLR from LLD or RLD (Bradshaw, Bishop, & Woodhead, 2017).

3.1. Atypical language dominance and handedness.

Epidemiological studies report that approximately 90% of the general population expresses LLD (Carey & Johnstone, 2014; Groen, Whitehouse, Badcock, & Bishop, 2012; Knecht et al., 2000; Mazoyer et al., 2014; Pujol, Deus, Losilla & Capdevilla, 1999). Although LLD is observed in both right and left-handers, significant differences are reported in their incidence. In a highly influential study with large sample (N=326), Knecht et al. (2000) demonstrated that the prevalence of RLD increased almost linearly with the degree (the strength) of left-handedness: from 4% in right-handers, to 15% in ambidexters and 27% in strong left-handers. In a sample of 144 right-handers, Mazoyer et al. (2014) even reported not a single RLD individual as opposed to 10 cases with RLD out of the 153 left-handers they included. Important to mention is that the former study used a more liberate cut-off to classify individuals as RLD (0.2 vs 0.5 in the latter study). This underlines the comparison problems between studies that come with arbitrary cut-off points between LLD/RLD and BLR.

Mazoyer et al. (2014) therefore suggested that the better option is to present the raw distribution of LIs. The researchers included 297 subjects in an fMRI study and Gaussian modelled the language LI distribution (-100 to 100) for left and right-handers separately. The distributions helped identifying 3 clear distinct phenotypes in language dominance: Typical LLD individuals, RLD individuals and participants with no clear language dominant hemisphere (BLR). Strong RLD was found absent in right-handers.

5 Since only +- 10% of the human population is left-handed, Mazoyer et al. (2014)

estimated that the group of individuals with RLD is less than 1% of total population.

In summary, RLD is present in a subgroup of left-handers, with its prevalence ranging from 6 to 27% depending on the source. By contrast, RLD appears far less frequent in right-handers (Carey & Johnstone, 2014; Knecht et al., 2000; Mazoyer et al., 2014).

Mechanisms underlying the association between language dominance and handedness remain poorly understood. Traditional theories assume that language lateralization and handedness are determined by the same gene (Annett, 1998; Crow, 2010). More specifically, these theorists tried to link a single, unspecified gene, i.e. the Right Shift or RS gene, to both asymmetries. According to the RS theory, one RS+ allele increases the probability of an individual to be right-handed and LLD, while the RS- allele leaves the direction of both asymmetries up to chance. Carriers of two RS- alleles have the highest probability of being left-handed and RLD. A lot of variance observed in the population was initially explained by the RS theory. However, later genome-wide association studies were unable to identify such gene (Armour, Davison, & McManus, 2014; Ocklenburg, Beste, & Güntürkün, 2013). Ontogenetic studies that tried to associate existing genes to language lateralization and handedness, and studies exploring the phylogeneses of these two traits, have obtained mixed results as well (for an overview, see Ocklenburg, Beste, Arning, Peterburs, & Güntürkün, 2014). Despite the work done so far, research still has a lot of ground to cover in this discussion.

3.2. Structure-function coupling.

MRI further enables structural investigations of the human brain. Both hemispheres are seemingly identical, however, some clear anatomical asymmetries are reported between left and right. Gray matter regions in the human brain such as the planum temporale and the anterior insula, and the arcuate fasciculus - a large white matter tract traditionally believed to connect Broca’s and Wernicke’s areas - are typically larger in the left hemisphere (Ocklenburg & Güntürkün, 2018). Many theorists suggest that these structural asymmetries underly left hemispheric specialization of language (Geschwind & Galaburda, 1987). However, attempts to link these structural asymmetries

6 to language dominance varied in success (for overviews, see Greve et al., 2013; Keller et

al., 2011; Ocklenburg, Hugdahl, & Westerhausen, 2013).

Studies investigating structural asymmetries in RLD individuals are conflicting as well. Some studies concluded that RLD individuals are more likely structurally reversed (Bidula & Kroliczak, 2015; Keller et al., 2018), while others report typical leftward asymmetries for language-related areas in RLD individuals (Vennooij, Smits, Wielopolski, Houston, Krestin, & van der Lugt, 2007; Greve et al., 2013). In a recent literature overview, Vingerhoets (2019) claimed that, if structure-function couplings do exist, it would take very large samples to demonstrate this as they are likely subtle. This is particularly problematic in atypical samples because of the low prevalence of RLD in the human population (Mazoyer et al., 2014).

3.3. Relationships between cognitive functions.

Research on the lateralization of single cognitive functions exists in abundance, however, only few studies addressed the relationship between different functions in the same individuals. Very divergent opinions emerge as to the lateralization of other functions when one is found atypical (language for example). According to the ‘multiple independent source’ or ‘statistical’ hypothesis, atypical lateralization of one function does not imply atypical lateralization of other functions, as they will lateralize according to their own specific bias (Bryden, Hécaen, & DeAgostini, 1983). Studies that reported no relationship between the lateralization of language and spatial attention claimed evidence in support of the statistical hypothesis (Rosch, Bishop, & Badcock, 2012).

A second, alternative hypothesis however proposes that there is a causal relationship in the lateralization of cognitive functions (Kosslyn, 1987). It suggests that activities involving the coordination of ordered operations require unilateral control, because a single set of commands for both halves of the body is more effective. Kosslyn postulated two innate unilateral control systems; (1) spatial control, controlled by the right hemisphere, and (2) speech control, commonly lateralized to the left hemisphere. Both systems would perform best if they are controlled by different hemispheres. According to the causal hypothesis, laterality of one control system enhances the probability of crossed asymmetry of the other system. It thus predicts a significant relationship between the lateralization of complementary functions (such as speech comprehension and production

7 for example). Supporting evidence has been reported for this hypothesis (Cai, Van der

Haegen, & Brysbaert, 2013; Gerrits, Van der Haegen, Brysbaert, & Vingerhoets, 2019). Gerrits and colleagues (2019) recently ran a study with individuals previously found LLD (n=12) and RLD (n=12). The researchers reported a significant relationship between language and recognizing written words (clear complementary functions, r=0.65), and a negative relationship between language and face recognition (a typical right lateralized function, r=-0.62). Although these results do not imply a one-on-one relationship between asymmetries for language and face recognition, they also argue against a complete independence of their lateralization. This study is a clear example of how the recruitment of atypical language dominant individuals opens perspectives to the study of brain functional organization (i.e. relationships between cognitive functions).

Another interesting line of studies suggest co-lateralization of two functions (Artamenko, Sitnikova, Soltanlou, Dresler, & Nuerk, 2020; Vingerhoets et al., 2013). Vingerhoets and colleagues (2013) for instance, provided evidence of co-lateralization of language and praxis. The researchers demonstrated that all 10 RLD individuals also had (atypical) right hemisphere dominance for praxis. They suggested the existence of shared hubs in neural networks underlying lateralized cognitive functions. In summary, possible mechanisms underlying the relationship between the lateralization of cognitive functions are contradictory and therefore remain largely unknown. The literature is in desperate need of studies investigating a complete pattern of lateralized functions within subjects. Most studies namely explored asymmetries of a single cognitive function at a time.

3.4. Behavioral relevance of atypical language dominance.

Many theorists believe that the typical leftward asymmetry for language has an evolutionary advantage. Various studies therefore explored behavioral consequences of atypical language dominance. In a highly controversial paper published in 1987, Geschwind and Galaburda claimed that many developmental disorders, including stuttering, autism and dyslexia, resulted from right hemispheric language dominance. Over the years, some studies reported higher prevalence of atypical language dominance in individuals with developmental language disorders compared to the healthy population (Annett, 2002; Bishop, 2013), while other studies did not (Berl et al., 2014). Wilson and Bishop (2018) rightly noticed the small sample size in previous studies. The researchers

8 ran a large replication study with high statistical power in an attempt to create clarity in

this discussion. They failed to report a significant relationship between atypical language dominance and developmental language disorders. Bishop (2013) further suggested that, should there exist a subtle relationship, atypical language dominance may be a consequence rather than a cause of language impairments. Research investigating a possible link between cognitive performance and language dominance in healthy individuals is contradictory as well. Some studies found a negative relationship between the degree of atypical language dominance and performance (Everts et al., 2009; Groen et al., 2013), while others found no link (Mellet et al., 2014), and even some studies reported a positive relationship between them (Van Ettinger-Veenstra et al., 2010). Future studies including larger samples of RLD individuals could shed light on the true relationship between language dominance and its behavioral implications.

Many researchers however argue that we should investigate the complete pattern of lateralization in the human brain when assuming potential evolutionary advantages. Functional brain segregation might have contributed to the natural selection procedure by reducing interference between competing neural processes (Lévy, 1977; Vallortigara, 2000). Crowding of functions in one hemisphere due to unconventional lateralization of at least one function would especially be detrimental for human behavior. According to this view, performance of one function would be crowded out if another function involves regions in the same hemisphere (Levy, 1977). Recently, Vingerhoets, Gerrits, & Bogaert (2018) found converging evidence for this “crowding hypothesis” in situs inversus patients (a condition in which one or more visceral functions are lateralized to the other side of the body). More specifically, patients with one crowded hemisphere were outperformed by typical individuals, and even by those with reversed typical functional brain segregation, another distinct phenotype in which all functions are atypically lateralized, resulting in a “functionally mirrored brain”. Although these results need to be generalized to the healthy population, it opens new perspectives in the evolutionary principles underlying functional brain organization.

4. Methods to Determine Language Dominance.

In summary, we can conclude that the research on (atypical) language dominance faces various challenges and perspectives. The origin of and mechanisms underlying the

9 remarkable link between handedness and language dominance remain poorly understood

at present. Current views on the relationship between lateralization of different functions are contradictory. It further remains unclear whether anatomical brain asymmetries underly language specialization. Finally, the behavioral relevance of atypical language dominance, and more generally of atypical functional segregation, is largely unknown. Solving these questions requires larger samples of participants with RLD. Traditional and modern methods to determine language lateralization are now discussed.

4.1. Traditional methods.

Back in the early days, language lateralization research relied on patients with aphasia (Critchley, 1954). Clinical studies are still quite informative. Brain lesions are probably the most dramatic display of cerebral asymmetries and can map loss of mental functions to damaged regions. However, brain damage is rarely focal. Many possible interaction effects and confounds due to overlapped brain regions and neural circuits may account for the deficit pattern of cognitive functions. Another disadvantage arises when one assumes that a damaged brain functions like a normal, healthy brain, except for the damaged regions. Alternative explanations may account for the deficits caused by brain lesions. Therefore, generalizing results obtained from clinical studies to the healthy population is difficult and should be done with precaution. Finally, clinical studies typically rely on small sample sizes (for an overview of advantages and disadvantages of studying patients with brain lesions, see Adolphs, 2016).

Another traditional study method to determine language lateralization, is the Wada-test (Wada & Rasmussen, 1960). It was used most intensively in patients undergoing brain surgery. For those patients, it is important to know where the language-related areas are, so they can be spared (Benjamin et al., 2018). The Wada procedure involves the injection of sodium amobarbital in the left or right internal carotid artery via a catheter. The injected sodium effectively blocks or shuts down functions of the hemisphere where the drug is injected and hence makes testing of the function partaken in the targeted hemisphere by virtue of deficiency. The Language Lateralization Index (LI) is then calculated by comparing the performance after left and right hemisphere injection. The Wada-test however does not provide intrahemispheric information; the technique is unable to localize the investigated cognitive function in more specific

10 subregions of one hemisphere or the other. Moreover, research found that the procedure

increases risk of carotid artery dissection and cerebral infarction (Loddenkemper, Morris, & Perl, 2002). As a result, researchers have looked for noninvasive alternatives.

4.2. Functional Magnetic Resonance Imaging (fMRI).

In recent years, functional Magnetic Resonance Imaging (fMRI) has been widely used in the study of language lateralization. This noninvasive alternative to the Wada-test can measure real-time blood flow while participants perform language tasks (such as a silent word generation task) in the scanner. The fMRI images obtained from the scanner show the brain according to changes in blood oxygen level and are a proxy for degree of mental activity. Later, statistical mappings are used to compare activation in certain distinct brain regions. In laterality research, differences in activation between the right and left hemisphere are computed and expressed in LIs, which typically range from -100 (left hemisphere dominance) to +100 (right hemisphere dominance).

The LIs can be obtained in different ways. Researchers typically define Regions Of Interest (ROIs) a priori. Areas of the brain that were previously found important in the implementation of the cognitive function in which the researcher is interested, are included in the ROIs. Then, techniques for measuring brain activity are either based on the number of active voxels (3D rectangle cuboids) or on the magnitude of the fMRI signal change in the prespecified ROIs. The number of active voxels can be determined by exceeding a fixed threshold, or a variable threshold (subject-specific). On the other hand, if researchers decide to use the magnitude of the signal change, they can choose to include all the voxels within a ROI, or only those with a prespecified activation level. Research identified that LIs based on signal change are more robust and reproducible than LIs based on the number of active voxels. Moreover, individual adapted thresholds are superior because they increase intrasubject variability (Jansen et al., 2006).

MRI is furthermore useful in establishing structural asymmetries of the brain. Morpho- and volumetric studies enable investigating gray matter differences between the left and right hemisphere. Moreover, white matter nerve tracts can be visualized through tractography, a 3D modelling technique, using diffusion weighted imaging data collected by MRI scanners. These modern techniques are instrumental for the study of structure-function coupling which faces high controversy at present.

11 One major disadvantage however, is that the purchase and maintenance costs of

an MRI scanner requires huge investments. The availability of MRI scanners is therefore limited in the wider research community. Moreover, collecting data from participants is quite expensive. This is perhaps reflected in smaller sample sizes in fMRI studies compared to behavioral studies.

4.3. The Visual Half Field (VHF) paradigm.

While fMRI is currently considered the standard method to determine language dominance, the low prevalence of atypical language dominance in the healthy population constitutes a considerable challenge in recruiting participants for researchers. Less than 1% of total population exhibits strong RLD, with this group seemingly found absent in right-handers (Mazoyer et al., 2014). Excluding right-handers in the initial sample is therefore a valuable first step for researchers investigating atypical language dominance. Nonetheless, scanning a large sample of left-handers with no a priori knowledge of actual language dominance would result in a relative small subsample of RLD individuals (approximately 10-20% of left-handers). This is problematic considering the high financial cost of fMRI tests. Low-cost behavioral indications of atypical language dominance could increase the probability of including actual RLD individuals.

Over the years, the Visual Half Field (VHF) paradigm was proven a valuable and well-validated tool to screen for hemisphere language dominance (Van der Haegen, Cai, Seurinck, & Brysbaert, 2011). It is a widely used method in the investigation of the lateralization of language – and other cognitive functions – at the behavioral level. The VHF task involves tachistoscopic presentation of stimuli (e.g. words) in the left or right parafovea. Stimuli displayed unilaterally are initially projected to and arrive in the contralateral visual cortex. The crossover of the optical fibers occurs in the optic chiasm, in the frontal regions of the human brain. At a later stage, interhemispheric communication is secured through transmission over the corpus callosum should task execution require processing in the contralateral hemisphere. The transmission however comes with a delay that can take up to 25-30ms (Aboitiz, López, & Montiel, 2003).

In the VHF paradigm, words are presented to the left and to the right of fixation to which participants have to respond (name out loud for example) as fast and accurate as possible. Words presented in the left visual half field (LVF) are initially projected to and

12 processed by the right hemisphere and words presented in the right visual half field (RVF)

are handled by the left hemisphere of the brain. Exploiting the anatomical cross-over principle, researchers argued that LLD participants would show a RVF advantage (in both reaction time and accuracy, which are expressed as the LIs), while the reverse would hold true for RLD individuals. Concordance between the initial processing hemisphere and the language dominant hemisphere would improve task performance because it can directly target language-related areas. Conversely, information arriving in the subdominant hemisphere leads to an efficiency loss, because it either requires interhemispheric transfer or it is processed by the less specialized hemisphere in which it arrives (Hunter & Brysbaert, 2008).

The feasible and low-cost setup of the VHF paradigm has led it to become a very popular method in laterality research. However, despite numerous promising findings of lateralized cognitive functions, there have been doubts about the validity and reliability of conclusions drawn from VHF studies.

Voyer (1998) assessed the reliability of the VHF task. A meta-analytic approach was used with 88 significance levels pertaining to test-retest reliability data. Results indicated that the reliability was only at a moderate level, namely 0.56 for verbal tasks. In the past 2 decades, new opportunities arose to further validate the VHF paradigm. Modern techniques such as fMRI or Electroencephalogram (EEG) measure brain activity more directly and its LIs can be feasibly correlated with VHF advantages suggesting left or right language dominance. Using blood flow as LIs, Krach, Chen and Hartje (2006) found that correlations with VHF LIs was remarkably low, r=0.18. Later research further reported low validation levels. Using electrical activity recorded from the scalp (i.e. EEG) while left-handed participants were performing a language-related VHF task, Cai, Lavidor, Brysbaert, Paulignan, & Nazir (2008) argued that validation of the VHF task was particularly low for RLD individuals. While LLD individuals showed the expected RVF advantage, no LVF advantage was reported in the RLD group (94% correct in the RVF, 91% in the LVF). Important to mention is that these conclusions were drawn from only 4 RLD individuals. Furthermore, a completely different method was used, i.e. EEG, and no subcategory for BLR individuals was made. This (again) underlines the challenge of comparison problems laterality research faces at present.

13 Hunter & Brysbaert (2008) claimed however, that many of the VHF tasks used in

the validation studies were suboptimal. Moreover, they suggested that some of the low correlations reported were probably due to bad testing rather than to the fact that the VHF paradigm is not fit for assessment of language dominance. Combining previously reported recommendations for a good VHF study, Hunter and Brysbaert prescribed several steps to design a VHF task.

First of all, the researchers emphasized the importance of including a sufficient amount of trials (>150 trials). Furthermore, bilateral presentation is preferred over unilateral presentation. Bilateral presentation creates competition between the right and left hemisphere and better counteract attentional biases. An arrow to the left or right of central fixation indicating the target word to be processed, should be included right before target presentation. A presentation time of around 200ms is preferred, as research indicated that participants cannot initiate an eye movement within this period when they first have to pay attention to a stimulus at the fixation location (Walker & McSorley, 2006). Shorter presentation time (<200ms) however would degrade the stimuli too much. To control for iconic memory of the target and any possibility of afterglow effects on the screen, including a backward mask is strongly recommended. Stimuli should also be matched in the RVF and LVF. Ideally, each word appears once in the RVF and once in the LVF in random sequence. Finally, an optimal VHF task involves word naming and not lexical decision (in a lexical decision task, participants manually decide whether or not the target word is a real or pseudo word). The reason for this, is that speech production is the most clearly lateralized function in the human brain (Kosslyn, 1987). For a visualization of a properly designed VHF task, see Figure 1 in the method section. The arrow at central fixation points to the RVF. The participants’ task is to name the target word as fast and as accurate as possible. On average, LLD individuals are expected to process words in this RVF more efficiently, while the reverse is true for RLD individuals.

Hunter and Brysbaert (2008) implemented their own recommendations to design a VHF word naming task and compared reaction time as a function of presentation location to the LIs obtained in the same 10 individuals during a mental word generation task (participants generate words they think of silently) in the fMRI scanner. The researchers reported strong positive correlations. All LLD participants showed a RVF advantage. The reverse was true for RLD participants.

14 Further fMRI validation with the same behavioral tasks was provided by Van der

Haegen et al. (2011) in a sample of 50 left-handed participants. Individuals with a VHF advantage in reaction time >60ms were invited for a follow-up fMRI study. All 22 participants with a RVF advantage in the VHF task were found LLD, while 80% of those with a consistent LVF advantage showed RLD in the scanner. Although no one-on-one relationship between VHF advantage and language dominance in the scanner was reported, the VHF paradigm was found a valid screening object, if used properly. Van der Haegen and colleagues further recommended future researchers limited in financial resources to use a strict criterium as a basis to invite individuals with a LVF advantage on, in order to keep the number of false positives as low as possible (60ms or even more). Researchers not restricted to certain financial limits could use a more liberate criterium (<60ms). This would decrease the chance of missing out on actual RLD individuals.

Remarkably, prediction errors in the latter study only occurred in the group that exhibited a LVF advantage. A similar story was reported by Cai et al. (2008). Considering the low prevalence of RLD however, the VHF paradigm is still very useful in the recruitment of RLD individuals (80% of individuals with a LVF advantage were RLD versus 0% with a RVF advantage in the study of Van der Haegen and colleagues).

Test-retest reliability of a properly designed VHF task has been assessed as well (Van der Haegen & Brysbaert, 2018). One hundred participants completed a word naming test. Test-retest correlations were provided for reaction time and accuracy as LIs separately. Strongest correlations between the first and second session were reported for reaction time as a LI of language dominance (r=0.83 versus r=0.77 for accuracy). Moreover, a significant correlation between reaction time and accuracy was found (r=0.59).

5. Present Study

Many research questions on atypical language dominance remaining unanswered to date could be solved by investigating sufficiently large samples of participants with RLD. FMRI is an indispensable research tool in the assessment of language dominance. However, researchers are often limited in financial resources and in access to scanners. Therefore, it is highly recommended first to screen an initial sample of left-handers on behavioral level via the VHF paradigm. Participants with a LVF advantage could then be

15 included in an initial sample of possible RLD individuals. The potential benefits of this

strategy relies however on the predictive value of the VHF task.

The current thesis is part of a large project investigating brain organization in atypical language dominant individuals at Ghent University. Various behavioral measures were recorded in a large sample of 315 participants. Participants with a LVF advantage (indicating RLD) of < -20ms in the VHF paradigm were invited for a follow-up fMRI study. Matched controls with a RVF advantage of >+20ms were included in the sample as well. Hemispheric language dominance in the scanner was determined according to the strength of fMRI signal change in a preselected ROI. A positive LI indicated RLD, a negative LI indicated LLD. Herewith, a third ‘BLR’ group was avoided because it is merely based on an arbitrary cut-off point.

The present study identifies the best behavioral predictors of atypical language dominance. LIs obtained from the VHF task were assessed against fMRI LIs. We predicted a strong relationship between these LIs. We further investigated the previously reported correlation between the strength of left-handedness and atypical language dominance (Knecht et al., 2000). Strong left-handedness, assessed through questionnaires and a hand performance task, might be an extra indication of atypical language dominance.

Method 1. Participants

A total of 315 left-handed and ambidexter participants were behaviorally screened at Ghent University. Potential RLD individuals, i.e. those with a mean LVF advantage of ≥20ms in reaction time, were invited for a follow-up fMRI study in which the lateralization of language, praxis, spatial attention, face recognition and prosody were assessed. A rather liberate criterium was used (Van der Haegen and colleagues, 2011, used a criterium of 60ms), in order to keep the number of missed RLD individuals as small as possible. Matched controls and additional LLD individuals were included as well. The larger project embarks on the perspectives and challenges laterality research faces at present. The current thesis on the other hand, identifies behavioral predictors of atypical language dominance.

16 The final fMRI sample comprised 63 participants (N female=56, mean age=21

years). All participants gave informed consent and subsequently completed both the behavioral and the fMRI session.

2. Behavioral Screening

2.1. Handedness assessment.

Strength of hand preference was measured by the Edinburgh Handedness Inventory (Oldfield, 1971) and the Tapley-Bryden dot-filling task (Tapley & Bryden, 1985). The inventory questions hand use in everyday activities by means of self-report. Participants have to indicate which hand they use for the act in question, and how strongly (“++” means strong preference for using one hand and is equal to 2 points, “+” means weak preference, 1 point) they do so. In the Tapley-Bryden task, participants are asked to mark as much circles as possible within 20 seconds. They have to do so with both hands separately.

Points from the inventory are added up and then the Laterality Index (LI) for handedness is calculated: LI=[(R – L)/(R+L) * 100]. The range of scores varies between -100 (preference of “strong left”) to +100 (preference of “strong right”). The same formula is applied to the number of dots marked by the participants in the Tapley-Bryden task.

2.2. VHF task.

2.2.1. Stimuli, design and apparatus. Participants completed a word naming task.

A total of 218 words were included in the stimulus list. Half of these words (109) served as target words to which participants responded, the other half as filler words to create matched bilateral word pairs. Words consisted of three or four letters (each 109 words). Target and fillers of each pair in the same trial had an equal number of letters, belonged to the same word class, never started with the same letter and were pairwise controlled for confounding variables as determined by the CELEX database (Baayen, Piepenbrock, & Van Rijn, 1993). The words were selected with the Word-gen software (Duyck, Desmet, Verbeke, & Brysbaert, 2004).

The VHF task comprised 218 word trials. 22 Fixation-control trials (numbers 1-9) were included as well, to ensure central fixation at the beginning of the trial. The

17 fixation-control trials were excluded from all analyses because they are centrally

presented and therefore no measure of language lateralization. The experimental thus consisted of 240 trials, divided in 4 blocks with short breaks in between. Before the start of the experimental block, all participants received 16 practice trials which were excluded from analyses as well.

We implemented Hunter & Brysbaert’s (2008) recommendations for the design of a proper VHF task. All targets appeared once in the LVF and once in the RVF, with an optimal presentation time of 200ms (Walker & McSorley, 2006). Bilateral presentation was assured (confer the bilateral word pairs). The target word was indicated by an arrow to the right or the left of central fixation. This ensures that participants are motivated to pay attention to the fixation location rather than look around (Schmuller & Goodman, 1980). Target presentation order and target location were randomized. A backward mask was included as well, to control for potential iconic memory or afterglow effects.

Trials were presented on a 15.6 inch monitor with 60 Hz refresh rate. Naming response latency were recorded with a voice key, and accuracy was determined by the experimenter using a key press.

2.2.2. Task and procedure. Participants were welcomed by the experimenter and

guided to a soundproof experimental room. They were instructed to fixate the center of the screen placed at a reading distance of 60cm at the beginning of each trial. Subsequently, they were instructed to shift attention to the side the arrow points to. Participants were asked to name the target as fast and as accurate as possible. On each trial, a fixation cross appeared first in the centre of the screen for 500ms. After that, either a central arrow (visual angle = 1°) together with the bilateral word pair, or a central number (fixation-control trial) appeared on the screen for 200ms. The arrow pointed to the word the participants had to respond to. Words were presented in Courier New font size 15. The outer edge of both words was fixed at 3.39°, the inner edge varied according to the length of the word (three letter word 2.07°, four letter word 1.6°). Stimulus presentation was followed by a backward mask. At the onset of the mask, the arrow was replaced by the fixation cross which remained visible until the voice key registered a verbal response. The next trial started when the experimenter determined accuracy. For an example of a trial, see Figure 1.

18 Figure 1. Trial example. Target word is presented in the RVF. “Tulp” is the Dutch word for

“Tulip”.

2.2.3. VHF analyses. Researchers typically run two separate analyses; one for

reaction time and one for accuracy. The mean difference between the RVF and the LVF is used as a LI of language dominance. The reaction time analysis is based on correct trials only; the accuracy analysis takes all trials into account. Unfortunately, research has indicated that it is highly unpredictable whether participants focus more on doing the task right or doing it fast, which results in speed-accuracy tradeoffs (Heitz, 2014). Many researchers therefore attempted to combine both measures. The most often suggested combined measure is the Inverse Efficiency Score (IES), which is typically defined as mean correct reaction time divided by percentage correct responses. However, Bruyer & Brysbaert (2011) advised against the use of this measure because the variance is often inflated. Recently, a promising new measure is developed by Liesefeld & Janczyk (2019) that tackles these variance issues: The Balanced Integration Score (BIS). It is calculated by first standardizing reaction time (of the correct trials) and accuracy to bring them to the same scale and then subtracting one standardized score from the other. In this way, 1 value per condition (e.g. RVF vs LVF) per participant is returned. The method was found relatively insensitive to speed-accuracy trade-offs and is therefore positively received (for

19 an overview, see Liesefeld & Janczyk, 2019). To our knowledge, the balanced integration

score has never been implemented in laterality research before.

2.2.4. Laterality indices. The LIs were based on three measures: mean reaction

time, mean accuracy, and mean balanced integration score differences between the RVF and LVF. A positive value indicates a RVF advantage, and thus LLD is expected; a negative value indicates a LVF advantage, and thus RLD is expected. The reaction time analysis is based on correct trials only; accuracy analysis takes all trials into account. Moreover, data was cleared from outliers (reaction time >1500ms).

3. FMRI Scanning

3.1. Word generation task.

Hemispheric language dominance in the scanner was determined using a covert letter verbal fluency paradigm (Cai, et al., 2013; Vingerhoets, et al., 2013). This paradigm consists of seven cycles that each includes a word generation and a control block. Both blocks are separated by a rest block. All blocks last for 15s each, and the task took 7 minutes in total to complete. During the word generation blocks, subjects were instructed to silently produce as many words as possible starting with a certain letter presented on a screen (b, d, k, m, p, r, s). The control block comprised silent production of the meaningless string “baba” projected on the screen. In the rest block (indicated by a horizontal line on the screen), participants were asked to relax and not to think of anything in particular. At the start of the fMRI session, the importance of avoiding head movement was emphasized.

3.2. FMRI session.

The fMRI data was collected on a 3.0 T PRISMA scanner using a 64-channel head coil. High-resolution T1 anatomical images of the participants’ brain were obtained using an MPRAGE sequence with 1mm isotropic voxel size and 176 sagittal slices (Repetition Time, TR=2250ms, Echo Time, TE=4.18ms, Inversion Time, IT=900ms, flip angle=9°). Functional imaging comprised T2* weighted echo planar images (total of 401 volumes) acquired from the scanner (TR=1070ms, TE=31ms, IT=17ms, flip angle=52°).

20 3.3. FMRI data analysis.

Data was processed and analyzed in Brain Voyager version 20.3. First, preprocessing was applied, which consisted of slice timing correction, motion correction, temporal filtering and co-registration of the fMRI images to the T1-weighted anatomical scan. A Gaussian smoothing filter was then applied to the functional data. Next, an Independent Component Analysis (ICA) identified noise components in the data for use in the General Linear Model (GLM). The condition onsets (onset of the rest/control/word generation block) were then convolved with the hemodynamic response function and were used as predictors. The GLM included the task-related predictors and nuisance predictors (identified by the ICA) to fit the functional data. For each task contrast, t-maps were generated. These maps were used to calculate the LIs from the fMRI session. The LIs were calculated within Brodmann areas 44 and 45 (i.e. the Regions of Interest – ROI), which were previously found to disturb word generation when damaged and match with Broca’s speech production area (Baldo, Schwartz, Wilkins, & Dronkers, 2006). The Brodmann areas were defined in a participant-specific way using BrainVoyager’s surface-based alignment tool.

3.3.1. Laterality indices. The LIs were computed based on the magnitude of the

signal change (Fernandez et al., 2001; Jansen et al., 2006). The statistical threshold was defined by taking the mean t-value of the 5% most active voxels over the left and right ROI separately. All t-values of the active voxels for each hemisphere were then summed and divided by the number of active voxels within the ROI. LIs were then obtained by using the following formula: LI= [(TsumR – TsumL)/(TsumR + TsumL) * 100]. The LIs were calculated using an in-house script programmed Matlab, version 2016b. A negative LI indicates LLD, whereas a positive value indicates RLD. By using 0 as a cut-off, we avoided creating a third BLR category because it is merely based on arbitral choices of the researcher.

4. Predictors of Language Dominance: Data Analysis

The above-mentioned fMRI LI calculations are considered preprocessing steps. The current thesis investigates how well the behavioral LIs predicts actual language dominance.

21 4.1. Correlations between behavioral predictors.

First, correlations among behavioral predictors are reported. We expected strong correlations between LIs within both measures (VHF task: reaction time, accuracy and BIS; degree of handedness: Edinburgh Handedness and Tapley-Bryden). We further controlled for (multi)collinearity between all LIs by means of the variance inflation factor. A factor larger than 10 indicates a problematic amount of collinearity (James, Witten, Hastie, & Tibshirani, 2014). It is then recommended to remove the concerned variable, or run separate analyses, because the presence of multicollinearity implies that the information that this variable provides about the response is redundant in the presence of the other variables (Bruce & Bruce, 2017). Multicollinearity would particularly be problematic for the analyses described in section 4.5 . A model with highly correlated variables used for outside prediction is inappropriate.

4.2. Degree of Handedness and language dominance.

We further looked at differences in the degree of handedness between RLD and LLD participants. Should there be a significant difference, we expect RLD participants to have stronger left hand preference by means of self-report (Edinburgh Handedness Inventory). We further expect a negative correlation between strength of self-reported hand preference and degree of language dominance in the scanner, in line with Knecht et al. (2000). Furthermore, we have no knowledge of published papers that established hand performance differences between the left and right hand in RLD and LLD individuals. We therefore assessed the usefulness of the Tapley-Bryden task for language lateralization research.

4.3. The VHF task and language dominance.

Subsequently, we assessed the predictive value of the VHF task. Prediction accuracy of our 3 LIs (reaction time, accuracy and BIS) from the VHF task are reported. Then, we divided individuals with a consistent VHF advantage from those with an inconsistent VHF advantage in reaction time and accuracy. Participants with a consistent VHF advantage thus respond faster and more accurate to one VHF than the other. In participants with an inconsistent VHF advantage however, there is a speed-accuracy tradeoff between the RVF and the LVF. This could potentially be problematic for our predictions. Moreover, only 1 of these measures could thus predict actual language

22 dominance correctly. We therefore expected that prediction accuracy would be higher in

individuals with a consistent VHF advantage.

Then, we correlated LIs obtained from the scanner with the behavioral LIs. We expected strong correlations between them. Since its use in laterality research has never been investigated before, we were particularly interested in the BIS. The previous standard method to combine reaction time and accuracy, i.e. the IES (reaction time divided by accuracy), was found inappropriate for laterality research because its variance is commonly inflated (Bruyer & Brysbaert, 2011). The variance explained by the IES (R²) was considerably lower compared to reaction time and accuracy as LIs. Therefore, we directly compared this new measure, the BIS, with reaction time, accuracy and the IES.

We further ran logistic regressions with language dominance identified by the scanner (LLD/RLD, 0/1) as dependent variable, and our behavioral LIs as predictors. This analysis was performed to investigate how well the behavioral predictors could categorize participants as LLD or RLD (considering our zero cut-off).

We also expected that estimation errors would be smaller in strongly lateralized individuals. Larger mean differences between the LVF and RVF in reaction time, accuracy and BIS would decrease estimation errors. We ran logistic regressions with prediction (incorrect/correct, 0/1) as dependent variable, and lateralization strength as predictors to determine this.

Finally, we demonstrated that the number of false positives in RLD individuals decreases and the number of missed RLD individuals increases as the cut-off criteria in reaction time, accuracy and the BIS to invite participants gets stricter. We made a direct comparison between the criteria used in the present study (i.e. 20ms) and the criteria used in our reference study (60ms, Van der Haegen et al., 2011). This analysis is particularly useful for future researchers considering their own financial resources.

4.4. Interaction effect VHF task and degree of handedness.

Since strong correlations between fMRI LIs and (1) LIs obtained from the VHF task (Hunter & Brysbaert, 2008; Van der Haegen et al., 2011) and (2) strength of hand preference (Knecht et al., 2000) have been reported before, we investigated potential interaction effects between them.

23 4.5. Outside prediction: cross-validation.

We finally investigated how well a model trained on all relevant behavioral predictors predicts new data. This was done with k-fold cross-validation. We ran analyses with language dominance as dependent variable (LLD/RLD). The goal of this analysis is thus to predict actual language dominance on new, “unknown” or “untrained” data. It is widely used in classification settings and is therefore valuable for future researchers in the recruitment of RLD individuals (it provides a means for model estimation in new data).

In cross-validation, the model is given a part of the dataset on which training is run, and later (after training), the model is tested against unknown test data. In k-fold cross-validation, the original dataset is randomly partitioned in k subsamples (k is most commonly 10; Refaeilzadeh, Tang L, & Liu, 2009). Then, a single subsample is retained to validate the model against. The remaining k-1 samples are used as the training dataset. After training part, the model estimates the outcome, and a prediction error is returned. This cross-validation process is then repeated k times, with each of the k subsamples used exactly once as the validation data. The k results (prediction errors) are then averaged to produce a single estimation. The model automatically identifies the best model to predict the outcome.

Results

The total sample for behavioral screening comprised 315 left-handed participants. All participants with a LVF >20ms, i.e. expected RLD individuals, were invited for the follow-up fMRI study at Ghent University Hospital. 38 participants agreed, 2 additional participants with a LVF advantage <20ms were included as well (one due to an administrative error; the other because of demographic variables). 23 Of these 40 participants with a LVF advantage were found actual RLD in the scanner. 24 matched controls with a consistent RVF advantage, indicating LLD, were invited as well. 23 Of them were actual LLD, the remaining 1 participant was found RLD. Thus, the final sample comprised 24 RLD, and 39 LLD participants. The LLD group included matched controls (N=24) and additional LLD participants. See Figure 2 below for a flowchart of the recruitment process.

24 Figure 2. Flowchart of the recruitment process. Total sample includes 24 RLD, and 39 LLD

participants.

In total, we have 63 participants for our analyses. Unfortunately, accuracy and BIS data of 2 participants was lost due to keyboard failure to administer accuracy (1 RLD and 1 LLD participant).

We first ran Shapiro-Wilk tests to check normality assumptions (Shapiro & Wilk, 1965). The null hypothesis was rejected for the distribution in fMRI LIs, in reaction time (VHF task) and in the Edinburgh Handedness Inventory (EHI). As a consequence, correlations with at least one of these variables were performed with Spearman’s Rho non-parametric test, others with Pearson Product-moment correlation. Furthermore, we ran Fisher’s F-tests (Fisher, 1922) to check the homogeneity assumptions in two sample (post-hoc) t tests. All analyses were performed in RStudio 3.6.3. (RStudio Team, 2015), visualizations were made with the “ggplot2” package (Wickham, 2016).

1. Correlations Between Behavioral Predictors

Table 1 displays the correlation matrix. As expected, strong correlations were observed between the LIs (reaction time=RT, accuracy=ACC, BIS) from the VHF task (RT and ACC: r=0.83, p<0.001; RT and BIS: r=0.92, p<0.001; ACC and BIS: r=0.98, p<0.001). Remarkably, there was no correlation between the measures of hand preference (r=0.01, p=0.91). A (marginally) significant correlation was reported for EHI with RT (r=0.27, p=0.03) and BIS (r=0.26, p=0.045), and close to significant with ACC (r=0.24,

25 p=0.06). The Tapley-Bryden (TB) hand performance task did not show any relationship

with other variables.

Table 1. Correlation matrix between all behavioral predictors.

Multicollinearity was then assessed for the further analyses. A Variance Inflation Factor (VIF) >10 was found for RT, ACC and BIS. These variables will therefore not be combined in a single model. The VIF for the EHI (1.11) and TB (1.06) was smaller than 10. We can thus safely set up models that include degree of handedness measures and VHF measures (although separately).

2. Degree of Handedness and Language Dominance

Mean LI for hand preference, measured with the EHI in the LLD group was -84.87, and -93.71 in the RLD group. A post-hoc two sample t-test indicated that this difference is not significant at 0.05 significance level (t=1.92, p=0.06, d=0.43). Spearman’s correlation between the LI from the scanner and the EHI variable was close to significant as well (r=-0.24, p=0.06). However, we noticed 2 outliers (Z score > 2.5) in our scatterplot. A second analysis was performed without these outliers, and the correlation level dropped to r=-0.21 (p=0.11). We thus failed to replicate previous reports of a significant linear relationship between hand preference, assessed through self-report, and language dominance. Figure 3 displays scatterplots of the correlation degree of handedness, assessed with the inventory, and the fMRI LIs.

We found no significant differences in the TB dot-filling task between RLD (-17.84) and LLD (-16.81) individuals (t=0.49, p=0.62, d=0.15). We further report a relationship far from significant with the LI obtained from the scanner (r=-0.06, p=0.66).

26 Figure 3. a.) Spearman’s Rho correlation between LI hand preference, as measured by the EHI,

and LI language dominance. b.) Same analysis, however cleared from outliers (Z > 2.5). Blue lines around the regression line represent the confidence interval, which is larger in panel a due to outliers.

3. VHF Task and Language Dominance 3.1. Prediction accuracy.

First, we analyzed the prediction accuracy of our 3 VHF LIs. A correct estimation is when a participant exhibits a LVF advantage and RLD in the scanner, or when it has a RVF advantage and is found LLD in the scanner. Mean prediction accuracy was 70.49% (n=43/61) for RT, 77.49% (n=46/61) for ACC and 75.41% (n=47/61) for the BIS. In practice, this means that 4 more individuals were correctly categorized as RLD/LLD according to ACC prediction compared to RT prediction. This difference is however negligible: a Kruskal-Wallis test revealed no differences between the predictors (χ2=8.06, p=0.66). Thus, all LIs predicted actual language dominance equally well.

3.3.1. Consistency of VHF advantage. We further investigated potential

differences between individuals with a consistent VHF advantage (n=41) for RT and ACC versus those with an inconsistent VHF advantage (n=20). The BIS automatically points to the same VHF in consistent individuals because it is a combination method of RT and ACC. Concordance between RT and ACC prediction might increase prediction accuracy. Furthermore, only one could have predicted actual language dominance correct when there is an inconsistency in estimation.

A Kruskal-Wallis test was first performed on mean prediction accuracy in participants with a consistent VHF advantage (n=41) versus prediction accuracy in

27 inconsistent individuals according to RT, and according to ACC, and was found

significant (χ2 =13.29, p=0.001). To establish the group differences, we ran post-hoc t tests.1 There was a non-significant difference between prediction accuracy according to RT and ACC in inconsistent individuals (RT=40%, ACC=60%, t=1.26, p=0.22, d=0.40). The differences between prediction in consistent individuals (85.37% correct) and prediction according to RT (t=4.06, p<0.001, d=1.11) and ACC (t=2.27, p=0.03, d=0.62) were significant. We conclude that estimation errors are lower in participants with a consistent VHF advantage compared to participants with an inconsistent VHF advantage. Figure 4a displays the mean accuracy differences, Figure 4b displays the statistical summary table.

Furthermore, participants with a consistent VHF advantage were more clearly lateralized to either the left or right hemisphere than those with an inconsistent VHF advantage (mean LI consistent=69.08, mean LI inconsistent=47.80, t=2.84, p=0.008, d=0.88).2 The absolute value of language lateralization was taken to determine lateralization strength (LI ranges from -100 to +100). Finally, a similar story is reported for the strength of lateralization in the VHF task (i.e. strength of VHF advantage). For a visualization of the difference in language lateralization strength, determined by the scanner, in consistent versus inconsistent individuals, see Figure 4d. A statistical summary table of all lateralization strength differences is displayed in Figure 4c.

3.2. Correlations between VHF and fMRI-based LIs.

Then, we correlated the LI obtained from the scanner against VHF LIs. As expected, significant negative relationships for all our predictors were found (RT: r=-0.53, p<0.001; ACC: r=-r=-0.53, p<0.001; BIS: r=-0.56, p<0.001). (Stronger) LVF advantages are thus related to (stronger) RLD in the scanner and vice versa. See Figure 5a, b and c for correlation plots.

1 _{Homogeneity assumption was accepted for all comparisons.}

28 Figure 4. a.) Visualization of prediction accuracy in consistent vs inconsistent individuals. b.)

Post-hoc comparisons prediction accuracy. c.) Post-hoc comparisons in strength of LIs for consistent and inconsistent individuals. d.) Visualization LI in the scanner. ***= p<0.001, **= p<0.01, *= p<0.05.

We further assessed the usefulness of the BIS as a new method to combine RT and ACC. The previous standard method to combine both methods, the IES, was found not suitable for laterality research: the variance is commonly inflated and therefore the explained variance (R²) in a linear model is substantially smaller than RT or ACC (Bruyer & Brysbaert, 2011). We made a direct comparison between all LIs, including the IES, by performing separate linear regressions. R² values were similar for RT, ACC and the BIS (range 0.31-0.35), however, R² was considerably smaller for the IES (0.21). We conclude that, while the results with regard to the IES are in line with previous research and therefore not well suited for the study of language lateralization, the BIS opens new opportunities for future laterality studies that aim to combine RT and ACC. Figure 5d displays the statistical summary table.