1
Research Master’s Language & Cognition
How little is enough?
The question of intensity and resting-state EEG changes in
seniors through language learning interventions
Master’s Thesis
Louisa Sophie Richter
S3826880
Supervisors:
Prof. Dr. Merel Keijzer
Dr. Hanneke Loerts
Faculty of Arts
University of Groningen
The Netherlands
August 10, 2020
2
Declaration
I hereby declare that this thesis is my original work, with due reference to the literature and acknowledgements of collaboration and assistance.
The present study was conducted under the supervision of Prof. Dr. Merel Keijzer and Dr. Hanneke Loerts at the University of Groningen.
Louisa Richter
3
Acknowledgements
I still remember planning my thesis project - a project that, in fact, never ended up coming to life. Due to the COVID-19 pandemic, the research I am now presenting is different from my initial idea. So I want to start by saying thank you to Rhomé Busstra, who was a great help at the beginning of my originally planned MA project and helped me prepare a senior language learning study in Dutch – without your translations and modifications of my Dutch material, the project would have never had a chance. Thank you!
Furthermore, my gratitude goes out to Prof. Dr. Merel Keijzer. Without our talks at ISB and in your office, I would not have been able to shape and create such an (well, in fact: two) interesting and ambitious project. Thank you for your thoughts, support, feedback, and most of all for making sure this project stayed feasible. I will greatly miss our work and discussions and I could not be more grateful to have had such an inspirational mentor not only throughout this project, but since February 2019, with the beginning of my RAP.
This brings me to my next point. Thank you to the Graduate School of the Faculty of Arts for having selected me for my RAP position. Without my work in Saskia Nijmeijer’s project, my involvement in the BALAB (https://www.balab.nl/) and the incredible opportunity of going to ISB 2019 in Edmonton, Canada, I believe my path might have been a different one. I might have never learned about resting-state EEG in relation to linguistics and I might have never become fascinated by neuro-/psycholinguistics in the realm of healthy aging. All of these things are pure speculations, but the fact remains that I am happy where I am at and I hope to be able to pursue a life in academia filled with more research on the matter.
Thank you to the amazing individuals involved in the BALAB. Not only did you motivate me to dive deeper into various research topics, but you also inspired me to speak my mind, engage in scientific discussions and start finding my place in academia. A special thanks goes out to Mara van der Ploeg. Not only have you been an incredible coach and mentor through
4 my RAP and internship involvements in your project, but you have also become a friend that I really enjoy working with.
Saskia Nijmeijer – where do I even start? Let me start by saying how much I admire the set-up of your study and how grateful I am for not only having been involved in it for approximately one year, but also for having the opportunity to use your data for not just one but two research projects of mine. During my time as your research assistant and intern I learned more things about academic conduct, including struggles involved in large scale projects, than I can put into words. So, thank you for having me and letting me explore the crazy world of scientific research in the safe surroundings of your project. A big thank you also goes to you and Nena Lejko for the initial help with FieldTrip.
Many thanks go to Floor van der Berg, who not only tried to help me with technical issues, but also graciously shared the resting-state and behavioral data from her own MA thesis study. I admire your work and strive to be as helpful and kind to others after me as you were to me during these past months. Thank you.
Before I turn to more personal matters, I would like to shout a big thank you across the ocean to Dr. Kinsey Bice, who not only inspired me to consider doing resting-state analyses with her talks at ISB, but was also so kind as to respond to questions via email. You were a great help and never once made me feel stupid for needing assistance. I admire your attitude and work and hope to get to meet you again one day.
Surely, I could mention and thank more incredible people at this point, like Kees de Bot, who inspired me to follow my dreams and start the research master in the first place, but I do not want to miss the opportunity to thank the people without whom this master and especially this last semester would have been impossible. First and foremost, I want to thank my parents, who have always supported me in pursuing my dreams. Thank you for loving me and raising me to be an independent woman. Thank you for always believing in me, even in times I did not. Mum, thank you for telling me back in the winter of 2014 that linguistics is in fact not as terribly
5 boring as it seemed at first glance. As always – you were right. And dad, thank you for listening to all the gibberish about my project, for letting me talk and cry on the phone, but never doubt myself. Thank you for being there for all of it – you are a greater pillar of strength in my life than you could ever imagine. A huge thank you also goes to Niels. I know I am not always the easiest to be around, but without your unconditional love and support, this path would not have been an easy one to go. Thank you for being by my side and for believing that I am one of the smartest people you know. My utter gratitude also goes out to your lovely family. With my family farther away, your family has never been less than incredible, and they made me feel welcome and at home throughout this whole time. Last but not least, I want to thank my friends – whether you live close or far away, we have or have not talked recently: I appreciate and miss you and want to thank you for having been there for me all this time. My last thank you goes to my grandma: Thank you for being interested in what I do and thank you for our lovely talks on the phone. I hope you stay cognitively fit forever because I would not want to miss our calls. So, how about learning a new language?
6
Abstract
In this study, we investigated the effects of a 10-day online Spanish course and a 3-month blended learning English course on resting-state EEG spectral power. We found no significant changes, which we believe to be an indicator of cognitive maintenance, meaning that language learning appears to aid in postponing age-related cognitive decline.
Furthermore, we find that language learning outcomes can be predicted by baseline resting-state frequency power. Frequencies and regions of interest in these predictive correlations depend on the task used to measure language skills. Our results showed that beta and gamma power in the right frontotemporal ROI correlated with IELTS speaking score, which suggests that high frequency power in the right hemisphere predicts language learning outcomes in the realm of production. Furthermore, our data displayed a relation between lower theta power in the posterior ROIs and healthy aging in numerous tasks. We also discovered that lower cognitive performance and intelligence at baseline might lead to greater learning gains. Lastly, we encountered that beta power in the left frontotemporal ROI related to the Peabody score.
Moreover, working memory performance at baseline, measured by the Digit Span task, served as a predictor of delta and gamma power at post-intervention for the group of seniors without memory complaints. These findings, although of low power, show that resting-state EEG is a working method to investigate neuroplasticity, cognitive decline and second language learning in elderly.
Keywords: third-age language learning, length and intensity of a language course, healthy
7 Contents List of Figures ... 9 List of Tables ... 9 List of Abbreviations ... 10 1. Introduction ... 11 2. Theoretical background ... 12
2.1. Bilingual experience and aging ... 12
2.1.1. Definitions: bilingualism ... 12
2.1.2. Bilingualism and executive functioning ... 13
2.1.3. Bilingualism and cognitive reserve ... 15
2.2. Bilingual experience as cognitive training to combat age-related cognitive decline ... 16
2.2.1. Changes in (behaviorally measured) cognitive functioning across the lifespan as a function of L2 learning ... 16
2.2.2. Changes in (late-life) neuroplasticity as a function of language learning ... 18
2.2.3. Effects of third-age L2 acquisition ... 19
2.3. Resting-state networks... 21
2.3.1. Frequency bands and oscillations ... 23
2.3.2. Age-related changes to the brain at rest ... 24
2.3.3. Resting-state changes induced by cognitive training regimes ... 25
3. This study ... 28
3.1. Method ... 29
3.1.1. Participants ... 29
3.1.2. Materials and Procedure ... 31
3.1.3. EEG acquisition and preprocessing ... 36
3.1.4. Statistical analysis ... 39
3.2. Results ... 39
(1) Resting-state changes as a function of language learning: ... 39
(2) Predictive power of frequency band power at baseline: ... 41
(3) Predictive power of executive functions at baseline: ... 43
3.3. Discussion ... 45
(1) Resting-state changes as a function of language learning: ... 45
(2) Predictive power of frequency band power at baseline: ... 47
Group 1: ... 47
Group 2: ... 50
(3) Predictive power of executive functions at baseline: ... 51
8 Sources:... 55 Appendix: FieldTrip script (EEG preprocessing) ... 73
9
List of Figures
Figure 1: Resting-state networks (taken from: Raichle, 2017) ... 22 Figure 2: Function of neural oscillations (taken from: Uhlhaas et al., 2008)... 24 Figure 3: Spectral power of younger and elderly participants (taken from: Scally et al., 2018) ... 24 Figure 4: Differences in correlational findings between the study by Rook et al. (2020) and Prat et al. (2016); [Blue = theta; red = low-beta; purple = upper-beta; orange = high-beta, green = gamma]; Figure taken from PowerPoint (J. Rook, personal communication, 16 June, 2020) ... 27 Figure 5: international 10-20 system (Brain Products GmbH, 2012) ... 36 Figure 6: Electrode clusters calculated on the basis of phase synchronization (taken from: Kepinska et al., 2017) ... 37 Figure 7: Alpha peak power changes ... 40 Figure 8: IELTS speaking and listening scores at baseline (yellow) and post-intervention (orange) of group 1 ... 41 Figure 9: Significant correlations between spectral power at baseline and language measures at post-intervention for group 1; Figure adapted from: Kepinska et al., 2017 ... 42 Figure 10: Picture Naming accuracy at baseline and post-intervention of group 2 ... 42 Figure 11: Significant correlations between spectral power at baseline and language measures at post-intervention for group 2; Figure adapted from: Kepinska et al., 2017 ... 43 Figure 12: Boxplot of Digit Span Scores for group 1 (pre = yellow, post = orange) and group 2 (pre = blue, post = purple) ... 44 Figure 13: Significant correlations between spectral power at post-intervention and Digit Span scores at baseline for group 2; Figure adapted from: Kepinska et al., 2017 ... 45
List of Tables
Table 1: Sociodemographic information of the participants ... 30 Table 2: Overview of the relevant set-up similarities and differences between the two studies ... 35 Table 3: Electrode clusters (ROIs) defined for our study ... 38
10
List of Abbreviations
DMN Default Mode Network
EEG Electroencephalography
fMRI functional Magnetic Resonance Imaging fNIRS functional near-infrared spectroscopy IAF Individual Alpha Frequency = alpha peak
L1 mother tongue
L2 second language / foreign language LOI Leidse Onderwijs Instelling
MCI Mild Cognitive Impairment
NL Netherlands
ROI Region Of Interest
11
1. Introduction
New problems and societal changes call for new perspectives, new research and innovative solutions. One of the big challenges of the 21st century is our aging society. As western medicine improves and life-expectancy increases, more people reach older adulthood (He et al., 2016). As older adulthood is frequently associated with declines in cognitive functioning due to structural brain changes (Persson et al., 2006) and loneliness (for a meta-analysis see Pinquart, & Sorensen, 2001), there is a strong demand for methods that aid the maintenance of cognitive functioning and mental wellbeing. Serious games and cognitive interventions are starting to emerge specifically for older adults. However, not much is known in terms of efficiency of such interventions and apps for seniors (for a review on intervention studies to prevent cognitive impairment see Kivipelto et al., 2018).
One specific intervention that has recently been coined as potentially neuroprotective in older adulthood deserves extra attention: early bilingualism has been shown to positively affect cognitive functioning and preserve mental fitness even at an older age. This has led to the postulation that late(r) second language acquisition aids the brain as the cerebral regions activated by language learning overlap with areas often affected by age-related cognitive decline (Antoniou et al., 2013). As a result, research into senior language learning to enhance neuronal activity and enhance mental flexibility even in the absence of lifelong bilingualism has increased in the last few years. Nonetheless, the literature on the matter is limited and hardly comparable. It has been suggested that neural activity changes at rest as a function of a cognitive intervention in seniors (such as a language intervention) can indicate changes in neuroplasticity (Styliadis et al., 2015). But other than suggestions, no empirical work exists to date. This study explores the matter and specifically evaluates pilot data that can contribute to the debate of language learning length and intensity needed to positively affect cognition in seniors. This is of great societal and scientific relevance as the outcomes can provide practical suggestions for
12 healthy aging, but – crucially – can also pave the way for future research into this topic. In this paper, we first summarize the scientific background on bilingualism, language learning as a cognitive training and resting-state changes induced by (language) learning. Then, the method, results and discussion of our study investigating spectral power changes as a function of language learning, including the modulating factor of length and intensity of the intervention, are presented.
2. Theoretical background
2.1. Bilingual experience and aging
Being exposed to two or multiple languages has been reported to have long-term effects on cognition. In this chapter, the term bilingualism and cognitive effects of lifelong bilingualism are explained and placed in a larger research tradition.
2.1.1. Definitions: bilingualism
When discussing bilingualism, many aspects should be taken into account. Societal
bilingualism, often referred to as a state of diglossia (Ferguson, 1959), in which two languages
are present in everyday life, is to be differentiated from individual bilingualism1. Definitions of the latter differ widely, making it hard to operationalize the term. Interpretations range from native-like proficiency in several languages to the ability to meet communicative needs (Hamers, & Blanc, 2000a). More broadly, these differences can be described in terms of
balanced vs. dominant bilingualism (Hamers, & Blanc, 2000b), which refer to the degrees of
ultimate attainment often mentioned in bilingualism research. This in turn is said to be determined by the age of acquisition of the second language (L2; Bylund et al., 2020). Taking such an approach, different ages of acquisition can be grouped as follows: childhood
1 Hamers, & Blanc (2000a) define individual bilingualism as bilinguality, in contrast to (societal) bilingualism.
13 bilingualism (before 11;0), which is often categorized in terms of simultaneous or consecutive acquisition, whereas the former refers to parallel acquisition of two or more languages and the latter describes staggered input and acquisition, adolescent bilingualism (11 – 17 years of age) and adult bilingualism (language acquired after 17;0) (Hamers, & Blanc, 2000b). In recent years, the term third-age learner has been introduced to the bilingualism literature. The term refers to (relatively) healthy retirees learning a foreign language (Oxford, & Gabrys-Barker, 2018).
Whilst definitions of bilingualism are not the focus of the current study, it is important to be aware of the differential interpretations and definitions, as they pertain to research findings of potential cognitive effects associated with bilingualism (for a review see Surrain, & Luk, 2019). Indeed, in the bilingualism literature, one often reads about advantages and disadvantages of growing up with two or multiple languages2. In the following section, increased executive functioning as a function of bilingualism is discussed. When reading and interpreting these results, it is important to take into account the individual experience that bilingualism is and how it can shape and modulate cognitive, neural and linguistic effects.
2.1.2. Bilingualism and executive functioning
Executive functioning, a term most commonly used to describe “three separate, but correlated, executive functions: updating of working memory, inhibition of distractors or responses, and shifting between mental sets” (Prior, & MacWhinney 2010, p. 253), is typically used to describe benefits of bilingualism, also labelled the so-called bilingual advantage (Schroeder, & Marian, 2012; Bialystok, & DePape, 2009). Such effects are mostly ascribed to balanced, simultaneous bilinguals (Bialystok, 2003). Though not always found (for a review see Antoniou, 2019; Kempe et al., 2015; Hernández et al., 2013), these positive non-linguistic
2 Since it is nowadays very hard to make a clear-cut distinction between bilinguals (people who speak exactly two
14 side-effects are said to be enhanced in bilinguals as opposed to monolinguals due to the constant need for inhibition of the other activated, yet not used language(s) in the brain of the bilingual (Van den Noort et al., 2019; Bialystok et al., 2012) and the cognitive effort that is needed to disambiguate meaning (Titone et al., 2017). According to Grant et al. (2014), inhibition is of even greater importance in bilingual environments that lead to frequent occurrences of language switching on an individual or societal level.
Enhanced cognitive flexibility (Kuipers, & Thierry, 2013; Kovács et al., 2009), memory generalization (Barr et al., 2020; Brito et al., 2014), selective attention (Comishen et al., 2019) and conflict inhibition (Poulin-Dubois et al., 2011) can already be found in bilingual toddlers below the age of 3 when compared to their monolingual peers, which reveals early effects of simultaneous L1 acquisition. Interestingly, Crivello et al. (2016) were able to show that the growth in bilingual proficiency over the course of 7 months at 31 months of age was correlated with enhanced executive functioning, especially in the realm of conflict tasks, which supports the claim of a bilingual advantage for early balanced bilinguals.
Enhanced executive functioning in lifelong bilinguals cannot only be found in toddlers, but extends to children (Bialystok et al., 2010), yet is rarely attested in adolescents (Chung-Fat-Yim et al., 2019) and (young) adults (for a review see Lehtonen et al., 2018). This is argued to be due to the accomplished maturation of executive functioning during that age and the fact that one is at peak efficiency of cognitive functioning skills, leveling out any differences in cognitive performance. In (elderly) adults, by contrast, the executive functioning advantage is once again found (Bialystok et al., 2005), including in recollection memory (Wodniecka et al., 2010), which involves retrieving contextual information alongside the information of a particular event, and episodic memory (Schroeder, & Marian 2012; Ljunberg et al., 2013). This finding has been interpreted to denote that lifelong bilinguals build up what has been termed cognitive reserve, which becomes especially noticeable at late adulthood, where cognitive flexibility in general decreases (Dajani, & Uddin, 2015).
15
2.1.3. Bilingualism and cognitive reserve
Cognitive reserve can be described as a compensatory mechanism of the brain against symptoms of clinical brain deterioration (Stern, 2009). It was first solely linked to brain size and neuronal count (Katzman et al., 1988), which is now often termed neural or brain reserve, whereas neural compensation is accumulated over the lifespan. Contributing factors include intelligence, education, leisure activities, socioeconomic status and, crucially, bilingualism (Radanovic, 2020; Scarmeas, & Stern, 2003). Summarizing, cognitive reserve is hypothesized to function through neural reserve, amounting to more efficient neural networks, and/or neural
compensation, meaning that alternate brain networks are employed once others are weakened
through brain pathology (Duncan, & Phillips, 2016; Stern, 2009). Generally, it can be assumed that neural reserve and compensation are interconnected, because “many of the factors associated with increased cognitive reserve, such as stimulating experiences, have a direct effect on the brain” (Stern, 2009, p. 2016; for a review see Guzmán-Vélez, & Tranel, 2015; for a review more focused on neural reserve see Perani, & Abutalebi, 2015).
Alongside the review by Stern (2009), which gives a structured overview on studies looking into these two functioning mechanisms, more neuroimaging research into the matter, especially with a focus on the variable bilingualism, has been done in recent years to shed more light on the matter (for a review see Baum, & Titone, 2014; Abutalebi et al., 2015; Abutalebi et al., 2014; Del Maschio et al., 2018; Anderson et al., 2018; Luk et al., 2011). Apart from neurological changes, studies investigating the influence of lifelong bilingualism on cognitive reserve often report positive effects, including delayed symptoms of memory decline as part of the diagnosis for single-domain amnestic mild cognitive impairment, which is often regarded as a precursor stage of dementia (Ossher et al., 2013), and the onset delay of visible symptoms of brain pathology, dementia specifically, by approximately 4 years (Bialystok et al., 2007; Schweizer et al., 2012; for a review see Bialystok et al., 2012; for another more updated review
16 see Bialystok et al., 2016). In contrast to that stand studies like Zahodne et al. (2014), who did not find bilingualism to be a modulating factor of age-related cognitive decline and dementia in over 1000 Spanish-speaking immigrants. This exhibits a need for further studies on the matter taking into consideration the individual differences affecting the bilingual experience.
Research in recent years suggests that late(r) second language acquisition also aids in building up cognitive reserve (Antoniou et al., 2013) and, therefore, might result in cognitive maintenance or improvement at an older age even in the absence of lifelong bilingualism. At the same time, given the mixed findings of the scantly available research to date on cognitive benefits to ensue from late-life language learning, more research is needed before we can state with increased certainty that third-age language learning can serve as an effective cognitive training with the power to enhance cognitive reserve.
2.2. Bilingual experience as cognitive training to combat age-related
cognitive decline
Due to the positive effects of lifelong bilingualism, researchers have set out to investigate if and to what extent second language learning at a later point in life could induce similar cognitive benefits. In other words, can taking up a new language later in life induce similar cognitive effects as lifelong bilingualism and could it perhaps even be used as a form of cognitive therapy?
2.2.1. Changes in (behaviorally measured) cognitive functioning across the lifespan as a function of L2 learning
Due to the executive functioning benefits of simultaneous bilingualism, possibly similar effects in consecutive L2 acquisition across the lifespan have been investigated. In consecutive childhood bilinguals, some studies report enhanced attentional control and cognitive flexibility in comparison to monolinguals (Nicolay, & Poncelet, 2013; Kalashnikova, & Mattock 2014). However, Poarch, & van Hell (2012) report no statistically significant difference between
17 attentional control in simultaneous, consecutive bilinguals and monolinguals, but interpret their findings as an emergence of enhanced attentional control that is expected to increase with further input/immersion. Unfortunately, child research on the topic of L2 learning effects is rarely comparable because the input that the children receive in their L2 varies greatly in quantity, quality, and environment (see Unsworth, 2016). In other words, the operationalization of the term bilingualism is too vastly different in this context for unambiguous evidence.
Though less often attested in bilingual adolescents (see chapter 2.1.2.), patterns of enhanced executive functioning have been found in highly proficient adolescent L2 learners. Javan, & Ghonsooly (2018) state that adolescents (age mean = 16.4, SD = 0.6) with a good command of their L2 show enhanced cognitive flexibility and working memory, yet not greater inhibition, in comparison to their less proficient peers. In young adult learners, inhibition appears to be trainable through L2 immersion (Linck et al., 2009; Heidlmayr et al., 2013). However, according to the review by Hilchey, & Klein (2011), the finding that bilinguals show increased inhibitory control, which in turn leads to enhanced conflict resolution, is rarely found. The effects of middle aged L2 acquisition are not often targeted in studies, which is why possible cognitive functioning benefits remain unclear.
Some recent studies looking into cognitive benefits of late (third-age) language learning have linked this specific cognitive training to an increase in attention switching (Bak et al., 2016), enhanced inhibition (Pfenninger, & Polz, 2018), better reading scores, greater verbal fluency and intelligence (Bak et al., 2014). However, other studies show no cognitive training effect of L2 learning in seniors that relate to variables like cognitive control and attention (Pfenninger, & Polz, 2018) or switching ability (Ramos et al, 2017). Hypotheses as to why a null effect was found include course-induced cognitive maintenance (Ware et al., 2017; Valis et al., 2019) as well as positive effects solely occurring in cognitively under-stimulated elderly adults, meaning that seniors who get little cognitively demanding input in their daily lives may respond positively to an intervention, whereas seniors with cognitively stimulating input do not
18 benefit further from a language course (Berggren et al., 2018). It also needs to be pointed out that the scant research to date has tended to be in the form of pilot-type studies comprising relatively small groups. Generally, more comparable research with a bigger sample size is needed to determine variables that influence these mixed results (Pot et al., 2019). Until then the conclusion remains tentative and hypothetical: “learning a foreign language may generate benefits for older individuals, such as enhancement of cognitive function” (Klimova, 2018, p. 5).
Generally, we see that different life stages appear to incur different effects. Based on the presented background, we hypothesize that this is due to various reasons, including cerebral aging, enhancement and decline of executive and cognitive functioning and increasing (world) knowledge. Furthermore, differing results might arise due to differences in L2 acquisition approaches. Modulating factors encompass quality and quantity of L2 input, teaching methods, materials, student motivation and so forth. Underlying the enhancement of cognitive functions as a function of language learning at all life stages, however, is an increase of neuroplasticity.
2.2.2. Changes in (late-life) neuroplasticity as a function of language learning
“Neuroplasticity is a continuous processing allowing short-term, medium-term, and long-term remodeling of the neurosynaptic organization, with the aim of optimizing the functioning of neural networks during phylogenesis, ontogeny, and physiologic learning, and following brain injury” (Duffau, 2017, p. 260). Though aging is usually accompanied by a decrease of gray and white matter density3 (Farokhian et al., 2017) and, therefore, the decline of certain cognitive abilities and processing speed, there is evidence that neural activity can be increased through training even at an old(er) age (Park, & Bischof, 2013).
3 “White matter tracts mediate the essential connectivity by which human behavior is organized, working in concert
with gray matter to enable the extraordinary repertoire of human cognitive capacities” (Filley, & Fields, 2016, p. 2093).
19 Specifically, L2 learning-induced brain changes can occur in people of all ages (for a review see Li et al., 2014; Rossi et al., 2017) and are usually found in the left inferior frontal and parietal regions in the realms of cortical gray matter and in the anterior parts of the corpus callosum, witnessed in increased white matter, independent of the language learning environment (for a review see Stein et al., 2014). However, in interpreting these findings it is important to acknowledge earlier found differences in gray matter between early and late bilinguals, as successive L2 learning generally induces less gray matter volume in brain regions related to language than simultaneous bilingualism (Wattendorf, & Festman, 2008; Kaiser et al., 2015; Mechelli et al., 2004). Furthermore, the gray and white matter network’s involvement during L2 processing appears to be modulated by the age of acquisition and the language proficiency of the speaker (Nichols, & Joanisse, 2016), which suggests greater neuroplasticity in highly proficient and typically early L2 learners. However, an increase of neuroplasticity even at an old age is feasible, making third-age L2 acquisition a potential cognitive training tool for the elderly.
2.2.3. Effects of third-age L2 acquisition
Baum, & Titone (2014) state in their review that “[n]ormal aging is an inevitable race between increasing knowledge and decreasing cognitive capacity” (Baum, & Titone, 2014, p.1). However, before looking at the advantages and disadvantages that third-age language learners may or may not experience, it is of primary importance to create a profile of the third-age learner, while at the same time acknowledging individual differences. Seniors are often faced with a decline in visual and auditory abilities, might find sitting and/or moving physically challenging and frequently experience feelings of loneliness and/or depression (Bosisio, 2019). Seniors also often struggle with concentration and error-induced anxiety (Klimczak-Pawlak, & Kossakowska-Pisarek, 2018). Furthermore, their general cognitive functioning may decline, and short-term memory is often impaired (Bosisio, 2019). This has consequences for the
third-20 age classroom, making a language learning needs analysis at the beginning of a course essential (see Grognet, 1997; Djoub, 2013).
Despite possible disadvantages hindering elderly students in their learning process, seniors have access to learning strategies (Garcia, 2017), meaning that they are aware of different learning methods and have deduced which one has proven most effective to them, and world knowledge (Gabryś-Barker, 2017), which may benefit them. Moreover, it has been argued that facilitatory effects are at play when a third (or more) language is learned due to increased metalinguistic awareness (Jessner, 1999) and possible linguistic transfer between typologically close languages (Cenoz, 2001), making additional language learning at a later stage in life particularly feasible.
Furthermore, it has been suggested that re-learning rather than learning a new language has great potential, as Van der Hoeven, & De Bot (2012) show that third-age learners re-learn previously acquired vocabulary as quickly as young learners, even though the acquisition of new words is of greater difficulty for them. This effect, which relates to the savings paradigm
effect, also applies to languages that were learned a long time ago, as retained lexical knowledge
is found even after a time span as large as 30 years (De Bot, & Stoessel, 2000).
However, close to nothing is known about the cognitive effects of late-life language learning. Language learning has been proposed to be a particularly effective mental training, as the activated brain regions overlap with areas often affected by age-related cognitive decline (Antoniou et al. 2013; Antoniou, & Wright, 2017). A meta-study by Pot et al. (2019) gives an overview of factors that have the potential to explain the differences in findings of senior language intervention studies, like discrepancies in sample size, course length and intensity, modality (in person, blended learning or online) and the used tests to assess linguistic and cognitive changes.
Especially concerning the effect of length and intensity of a third-age language intervention, little is known. Bak et al. (2016) found that a one-week intensive course with
21 approximately 14h of language instruction significantly improved attentional switching across all ages. This enhancement persisted until the follow-up measurement approximately 9 months after the intensive course for students who continued studying the language (Gaelic) with an intensity of 5h+ per week. This suggests that 5h of language training per week are what is minimally needed to cause cognitive benefits, especially following a short, intensive introductory course. In line with this thought, it makes sense that the study by Pfenninger, & Plotz (2018) also reported significant improvement of executive functioning, as their 4-week English course included 6h of classes per week, broken down into three 2h sessions. Reasons as to why Cox (2017) with around 10h per week of language input for a week and Berggren et al. (2018) totaling 5h per week for 11 weeks found no cognitive improvements can only be speculated upon without further research.
2.3. Resting-state networks
The influence of cognitive training programs like (online) language learning on cognition in seniors can be investigated using different methods. Apart from behavioral tests that can be administered before and after the intervention, one can use neuroimaging methods to capture more precisely what happens in the brain. In recent studies on third-age L2 acquisition, researchers have turned towards neuroimaging methods to investigate neural effects. Very recently in this regard is the focus on the brain in the absence of a task and how such resting-state networks change as people engage in language learning.
22 At rest, the so-called default mode network (DMN),
one of several resting-state networks (Damoiseaux et al., 2006), shows a particular activation pattern in the brain.The DMN “instantiates processes that support emotional processing […], self-referential mental activity […], and the recollection of prior experiences […] [and is therefore] never turned off but, rather, carefully enhanced or attenuated” (Raichle, 2017, p. 440). The DMN has been shown to be susceptible to age-related changes in
connectivity, which is further expedited in patients with Alzheimer’s disease (Jones et al., 2011). The (functional) Magnetic Resonance Imaging ((f)MRI) literature on aging gives us some insights into how the brain changes as a result of aging. Salami and colleagues (2014) found that declining memory results from disrupted functional connectivity in the hippocampus, in turn leading to a functional isolation of the region at rest. Generally, functional disruptions correlate with a decrease in white matter integrity and, therefore, cognitive decline (Andrews-Hanna et al., 2007). These disruptions are usually attested in connections between systems, which leads to the functional isolation of networks (Sala-Llonch et al., 2014). According to Andrews-Hanna et al. (2007), the fronto-parietal disruption particularly affects the DMN. A visualization of the activated brain regions of resting-state networks, including the DMN, can be found in Figure 1. Resting-state networks are mostly captured using fMRI because of the good spatial resolution of the tool, which results in precise connectivity coherence analyses. Electroencephalography (EEG) recordings are not known for their good spatial resolution, so detailing resting-state networks using EEG methods is harder to do. What is done instead is a superficial connectivity analysis, which provides information about the connectivity across the scalp, yet remains uninformative about the source of the activity, or spectral power analyses. Power spectra are calculated on the basis of brain wave frequency
Figure 1: Resting-state networks (taken from: Raichle, 2017)
23 power across all electrodes, individual electrodes or within specific regions of interest (ROIs) and are associated with gray and white matter densities (see Smit et al., 2012).
2.3.1. Frequency bands and oscillations
Brain waves are the continuous firing of neurons in the brain that result in so-called neural oscillations at different paces and frequency bands (Mantini et al., 2007). These oscillations are susceptible to long- and short-term changes, including - crucially - aging (Scally et al., 2018), but also states of vigilance (awake, asleep or sleep-deprived) (Van Diessen et al., 2015) and a preceding task (Klimesch et al., 1999). This means that, ideally, all participants of a study have a similar background, are in a comparable state of mind and the resting-state recording is done before any other task. Furthermore, all participants should be asked to either do the measurement with their eyes open or closed. Both variants are accepted, but the modulations of the data that occur under these two conditions are to be taken into account. Generally, the activity across all frequency bands is reduced in the eyes-open condition. This finding can be found across all ages (Barry, & De Blasio, 2017).
Frequency bands are more or less standardized definitions of neuronal wave lengths, with each band’s values being situated within a certain Hertz range. The frequencies are usually defined as follows: delta (0.5-4 Hz), theta (4-8 Hz), alpha (8-13 Hz), beta (13-30 Hz) and gamma (30-90 Hz) (Van Diessen et al., 2015). Subdivisions are sometimes made for analyses focused on the function of a certain Hz range. An example for this is the alpha band, sometimes divided into upper and lower alpha, which consists of the Hertz frequencies situated to the left and right of the alpha peak, also known as individual alpha frequency (IAF; Klimesch, 1999), as upper alpha relates to memory demands and lower alpha is sensitive to attention (Klimesch, 1997). IAF varies between participants and declines across the lifespan; it tends to be situated between 8-14Hz (Klimesch, 1997), but is usually between 9.5 and 11.5 for healthy young adults (Klimesch, 1999). Due to the personal and age-related variability of IAF and possible
mis-24 calculations of alpha power through the usage of standardized frequency bands (see above), recent studies, like Bice et al. (2020), take the approach of defining each participant’s frequency bands on the basis of their IAF. This is done in order to avoid mistakes in alpha power calculations. Alpha power calculations are often crucial to the set-up of a study, because it has been shown to positively correlate with intelligence (Doppelmayr et al., 2002).
Since different frequencies serve specific functions, as indicated in Figure 2, power spectra can serve as indicators of functioning and neuroplasticity.
Figure 2: Function of neural oscillations (taken from: Uhlhaas et al., 2008)
An example of a prototypical power spectrum at rest in an eyes-closed condition and the age-related changes in frequency power that are often observed, can be found in Figure 3.
2.3.2. Age-related changes to the brain at rest
As suggested in Figure 3, neuronal frequencies change rather drastically with increased age. (Healthy) seniors generally experience a slowing in their individual alpha peak frequency, which affects the whole power spectrum (Scally et al., 2018), creating a need for individually
Figure 3: Spectral power of younger and elderly participants (taken from: Scally et al., 2018)
25 calculated frequency bands on the basis of alpha peak to ensure adequate power calculations (see above).
Slow wave (< 7 Hz) power, which is associated with cognitive performance and healthy aging, is usually accompanied by a decrease of delta power. However, it remains unclear if higher or lower delta power relates to better cognitive performance in healthy elderly (see Vlahou et al., 2015). Lower theta power, especially in the posterior areas of the brain, has been linked to healthy aging (for a review see Ishii et al., 2018). Furthermore, relative theta power relates to verbal memory, attention and executive functioning in healthy elderly (Finnigan, & Robertson, 2011).
In EEG research, cognitive decline is associated with a decrease in median frequency, peak frequency and alpha-to-theta ratio in the prefrontal regions during eyes-closed resting-state recordings (Choi et al., 2019). When the signs of cognitive decline are clinical and in fact indicate an early stage of dementia, specifically Alzheimer’s disease, the EEG power spectrum tends to show an absence of an alpha peak (Pucci et al., 1999). This reduction in alpha power and slowing in alpha peak correlates with an observable decline in cognitive functioning in patients with mild cognitive decline (MCI; López-Sanz et al., 2016) and can, therefore, be used as an indicator of cerebral deterioration.
2.3.3. Resting-state changes induced by cognitive training regimes
The brain at rest may show different patterns for different people as a function of factors like age and health. Furthermore, it has been shown that brain changes can occur due to cognitive training programs, like music interventions (Fachner et al., 2013) and meditation (Taren et al., 2017). In healthy elderly, a three-month multi-domain cognitive training program with a focus on memory, reasoning and problem-solving strategies prompted changes in functional connectivity within and between the DMN, as well as the so-called salience network
26 and the central executive network, all visualized above in Figure 1, which were attested in a follow-up measurement, one year after the intervention (Cao et al., 2016). The three resting-state networks are of particular importance as they “have been considered the ‘core’ neurocognitive networks for understanding higher cognitive functions […] [and] are strongly affected by normal aging and Alzheimer’s disease“ (Cao et al., 2016, p.2). The lasting functional connectivity changes within and between these networks suggests that long-lasting changes in neuroplasticity through cognitive training programs in elderly are achievable.
Similarly, positive findings were reported using resting-state EEG. Styliadis and colleagues (2015) investigated the effect of a combined 8-week physical and cognitive training intervention in elderly with mild cognitive impairment (MCI) and found an increase in neuroplasticity, visible through a decrease in delta, theta and beta frequencies and a reduction in complexity and coherence of the resting-state data.
In relation to second language acquisition and resting-state EEG, a seminal study by Prat et al. (2016) is of great importance. They found that resting-state EEG power at baseline serves as a predictor of language learning rate in healthy young adults. Their analyses show that particularly beta and low gamma frequencies in the right hemisphere strongly correlate with second language learning rate and explained approximately 60% of language level variability at post-intervention. Rook et al. (2020) tried to replicate this effect for senior language learners. The preliminary results suggest that the frequencies and electrodes that significantly correlate with language learning outcomes differ in seniors when compared to the young adult population targeted in Prat et al. (2016). This might be due to various reasons, most pertinently the age difference of the learners, but also the length and intensity of the training programs, differences in behavioral tasks, the used headsets and so forth. Moreover, it has to be taken into account that the language learning rate, as used in the study by Prat and colleagues (2016), does not equal language learning outcomes. This might further modulate the effects attested. The differences in findings are visualized in Figure 4.
27
Figure 4: Differences in correlational findings between the study by Rook et al. (2020) and Prat et al. (2016); [Blue = theta; red = low-beta; purple = upper-beta; orange = high-beta, green = gamma]; Figure taken from PowerPoint (J. Rook, personal communication, 16 June, 2020)
A study by Rossi et al. (in prep.) shows that changes in the resting-state power spectrum can occur even after a two-day language intervention in healthy monolingual and bilingual adults learning Finnish. Most importantly, they found a significant alpha power decrease in monolinguals in some ROIs, but an alpha power increase in the bilingual learners. Since a decrease in alpha power is related to growing memory demands (Klimesch, 1997) and increased alpha power is an indicator of enhanced inhibition (for a review see Klimesch, 2012), these findings appear to be a result of the different cognitive demands of growing up with one versus two/multiple languages. This finding goes hand in hand with the study of Bice et al. (2020), who state that bilinguals tend to show greater alpha power and coherence in the alpha and beta frequencies than monolinguals, which suggests that the bilingual brain not only responds differently to learning new languages but also differs in oscillatory power at baseline.
Rossi and colleagues (in prep.) also find that course-induced power changes are age-dependent, as they find less significant changes when controlling for age in their middle-aged
28 sample (age mean monolinguals = 20.8, bilinguals = 19.9). Despite this finding, other studies mentioned above have shown that the elderly brain is still capable of change, which is why it can be assumed that language-course-induced resting-state changes are possible in elderly. However, to our knowledge, no research has been done on the matter. Moreover, it remains unclear to which extent the length and intensity of a third-age language training might modulate these possible effects.
3. This study
From the work that has been done follows that language learning has potential to serve as a cognitive training regime for elderly and might, subsequently, induce brain changes. However, it remains unclear how long and intense these training programs ought to be to cause such neurological modifications and to which extent language learning brings about changes measurable with resting-state EEG. The main aim of this study is to ascertain the effect of a 10-day versus a 3-month language learning intervention on resting-state patterns in elderly. The underlying intent in doing so is to examine in more detail the effectiveness of language learning in relation to healthy aging. The research questions that form the basis of this study are as follows:
1. Does a foreign language intervention lead to changes in resting-state EEG in seniors? If so, in which frequency bands and regions does this change occur most prominently and to which extent does the length and intensity of the language training modulate this effect?
2. Does frequency band power during resting-state EEG at baseline predict short-term language learning outcomes (see Prat et al., 2016)?
3. Do executive functions at baseline predict resting-state network changes in elderly? Is this manifested differently depending on the length and intensity of the training?
29 Although this research is highly exploratory, we hypothesize resting-state changes as a function of foreign language learning in seniors, most prominently in the prefrontal and parietal areas, similar to the changes observed in Rossi et al. (in prep.). We expect these changes to mostly occur in the alpha band, as Rossi et al. (in prep.) observed alpha changes and this is the frequency most affected by age-related neuroplasticity changes. It is unclear to what extent the length and intensity of the language training will modulate this effect, but greater input might causally relate to greater effects. Beta and low-gamma frequencies might predict short-term language learning rate in seniors, similar to the effects found in Prat et al. (2016), but it is unclear if these correlate with learning outcomes. This is caused by the different set-ups of the studies and a focus on learning outcomes, rather than learning rate. Furthermore, it is unclear if we can replicate the findings of Rook et al. (2020) visualized above, due to set-up differences and potential differences in (subjective) cognitive decline between the elderly. It is assumed that executive functioning at baseline influences resting-state network changes in elderly, but the direction of the effect is unclear and will be explored in this study.
3.1. Method
For the purpose of this study, the data sets of two different studies (see Nijmeijer et al., submitted and Van der Berg et al., 2019) were used, referred to as group 1 and group 2 respectively. In what follows, both groups are described in more detail. We use terms like the
study to refer to the data comparison done in this paper.
3.1.1. Participants
The study included a total of eight participants in group 1 and five participants in group 2. Some data sets had to be excluded due to noisy EEG data, which are not included in this count. The steps taken to evaluate and preprocess the data are explained in section 3.1.3. All
30 seniors were functionally monolingual elderly (65 – 85) with Dutch as a mother tongue, and who all lived in/around the state of Groningen (NL) at the time of testing. The mean age in group 1 was 69.87 and 70.60 in group 2. There was no significant age difference between the two groups (w = 17.50, p = .76). Functionally monolingual in this case means that the participants grew up with one language and solely used it in daily life. However, the participants differed significantly in the quantity of learned L2s across the lifespan (w = 0, p < .01), as the participants of group 2 had greater previous foreign language learning experience. Unfortunately, nothing can be said about the level of proficiency of these languages or their typological distance to Dutch and English/Spanish, as we do not possess this information for both groups. The relevant sociodemographic data is summarized in Table 1.
Table 1: Sociodemographic information of the participants GROUP_
NUMBER
SEX AGE COGNITIVE
RESERVE LEARNED L2S 1_1 m 72 185.7 0 1_2 f 67 156 1 1_3 f 73 131.3 1 1_4 m 67 145.7 2 1_5 f 73 131.3 1 1_6 m 71 NA 0 1_7 f 68 126.3 1 1_8 f 67 151.3 1 2_1 m 69 167 5 2_2 m 75 182.3 4 2_3 m 70 145 6 2_4 f 67 147.7 5 2_5 f 72 186.7 4
The participants in neither group had a diagnosed form of mild cognitive impairment (MCI) or dementia, but all participants in group 1 reported to have experienced subjective signs of cognitive decline, memory complaints in this case, for over 5 years. However, no participant demonstrated signs of pathological aging, which is why we considered them healthy elderly.
31
3.1.2. Materials and Procedure
The participants for both groups were recruited through several channels, like newspapers, short presentations at elderly homes and events for seniors, personal networks, and flyers. Everyone provided informed consent after having read an information booklet. The participants were able to withdraw at any given moment. The procedures were approved by the ethics committee of the University Medical Center Groningen (MeTC – group 1) and University of Groningen (CeTO – group 2) respectively.
Participants in both groups were invited to the cognitive neuroscience center (CNC) at the University Medical Center Groningen (UMCG) for baseline and post-intervention testing. The measurements were planned within a one-week time window before/after the intervention. Both testing sessions included linguistic tasks and the Digit Span task as a behavioral measurement of (verbal) short term memory (Cullum, & Larrabee, 2010). At baseline, the participants furthermore filled out a background questionnaire and responded to the questions of the cognitive reserve index questionnaire (CRIq; Nucci et al., 2012). The tasks are explained in greater detail below.
Verbal Fluency: For this task, seniors were asked to name as many words as possible
starting with a given letter (phonemic verbal fluency task) or belonging to a certain category (semantic verbal fluency task) within one minute. The score comprises the number of correct items. The Dutch verbal fluency group difference at baseline was not statistically significant (w = 9, p = .12), with an overall mean of 40.31 correct items (SD = 9.60). For the purpose of this study, we used the overall verbal fluency score, a combination of the phonemic and semantic verbal fluency score, as a measure of updating capacity (Shao et al., 2014). The verbal fluency task was also used in the language measurements explained below.
Cognitive Reserve Index questionnaire: The questionnaire was administered in
32 individual has spent doing certain things, like reading newspapers and playing instruments, and is used to estimate the lifestyle contributions one has made to increase cognitive reserve (Nucci et al., 2012). The participants of the two groups had accumulated a comparable cognitive reserve across their lifespans, such as there was no statistically significant group difference in their cognitive reserve index (w = 26, p = .19). The cognitive reserve index score of each participant is included in Table 1.
Language tests: In group 1, all participants had an English proficiency level of A1 or
A2 according to the Common European Framework of Reference (CEFR) scale (Council of Europe, 2001) at baseline, which was measured with an IELTS listening task (British Council, 2019), an IELTS-inspired prepared free speech task based on the online IELTS practice tasks (Nijmeijer et al, 2019), the Peabody picture vocabulary test4 (Dunn, & Dunn, 2007) and English verbal fluency (mean = 21.67, SD = 5.75). Unfortunately, we do not possess the complete language data for two participants, so most calculations were done with just six seniors. All tasks, except the listening task, which was administered at the beginning of the first conversation class, were completed alone with two research assistants who slowly explained the tasks in English. The same procedure was repeated after the 3-month training program.
Group 2 performed a Spanish picture naming task based on the vocabulary the students learned with the Rosetta Stone program (van der Berg, 2019) at baseline (mean = 0.60, SD = 1.34) to ensure minimal to no previous knowledge of the language. Another version of the task was completed after the learning period. After the 10-day training, they also completed a free verbal fluency task, where they named as many Spanish words as they could within one minute. Furthermore, the Rosetta Stone program calculated a score containing the percentage of correctly answered exercises across the duration of the course, which we refer to as the Rosetta Stone score.
4 The Peabody Picture Vocabulary Test is a measure of receptive vocabulary. Participants are asked to identify
33
Digit Span task: Both groups performed the Dutch version of the Digit Span task
(forward, backward and letter-number-sequencing) at baseline, which is part of the Wechsler Adult Intelligence Scale (WAIS-IV; Wechsler, 2008). During the task, participants were presented with a sequence of digits they were asked to recall. Each trial consisted of an increasing number of digits. In the backwards task, participants repeated the learned sequence in a backwards order. During the letter-number sequencing task, participants were presented with a sequence of numbers and letters, which they recited in ascending (numbers) and alphabetical (letters) order. For the forward and backward subtask, testing was terminated after the participants could not adequately recall two series within a trial, three in the letter-number sequencing task. The maximum score for the first two subtasks was 12, 21 for the letter-number sequencing.
The forward task measures attention, whereas the backwards task taps more into working memory. The letter-number sequencing task is an expansion of the digit sequencing task, which also relies on working memory, yet was created for individuals below 70 (Cullum, & Larrabee, 2010). This means that the results might not be meaningful for our age group. The group difference for the backward Digit Span task at baseline was statistically significant (w = 34, p = .05), which suggests similar attention capacities, as measured with the forward task, yet greater working memory performance in group 2. The complete task with all subtasks was repeated during the post-intervention measurement.
Language intervention: In between the testing sessions, the language interventions
took place. The length and intensity of the two groups differed greatly. For group 1, the intervention period was 3 months and consisted of a blended learning approach. The participants were asked to study five times a week for 45 minutes each, using the online platform of the Leidse Onderwijs Instelling (LOI; https://www.loi.nl/), which focusses on meaning, rather than explicit grammar instruction. Furthermore, they participated in a conversation class for 1,5 hours every second week, which was communication-based and also
34 followed implicit teaching methods (see Andringa & Rebuschat, 2015 for an overview of implicit vs. explicit language teaching methods). Group 2 used the commercial online learning software Rosetta Stone to learn Spanish for 10 consecutive days. The program offers vocabulary and phrase acquisition in a real-life context through visual and auditory cues (Rosetta Stone Ltd., 2020). The students were urged to spend one hour per day studying. There were no further means of practice or input.
Electrophysiological measurements: After the behavioral part of the session and more
crucially for the present study, no-task neuroimaging data, also called resting-state, was recorded using electroencephalography (EEG). In this respect, the two groups differed greatly. One reason for this is that they were recorded in two different labs, which differ in regard to used caps, quantity of channels and recording software. Further differences between the two recordings were (1) their length, (2) when they were administered and (3) if the eyes were opened or closed during the measurement. Firstly, the resting-state recording of group 1 was longer (10 minutes) than the recording of group 2 (5 minutes). Furthermore, resting-state recordings for group 1 were done after the completion of another EEG task, a color-shape-switch task in this case, while the resting-state in group 2 was recorded immediately after the capping of the participant. Lastly, while group 1 looked at a fixation cross (eyes-open condition) during the resting-state measurement, group 2 was recorded in an eyes-closed condition. The latter is of particularly great importance, as studies have shown that global activity in all frequency bands increases when eyes are closed, due to the reduction in external arousal. This finding has been shown to apply across age groups (Barry, & De Blasio, 2017). Moreover, the data of group 1 might have been influenced by the parallel fNIRS recording. fNIRS stands for functional near-infrared spectroscopy and, similar to fMRI, used infrared light to measure oxygen changes in the blood, which serves as an indicator of cognitive demand (Fishburn et al., 2014). When EEG and fNIRS are on one cap, the fNIRS cables can disturb the EEG signal
35 when not placed correctly. Although we believe this not to be the case in the recordings used for this study, we cannot completely rule out artifacts caused by this double measurement.
Due to the fact that group 1 had a preceding (switch) task and did an eyes-open resting-state recording, we expect less activity across the power spectrum, but increased alpha and theta power is possible due to the preceding switch task, which leads to greater prefrontal and posterior theta coupling and increased alpha power (Sauseng et al., 2006). However, an eyes-open condition has been shown to lead to decreased alpha power due to increased arousal, which leads to the neuronal activation of visual processing (Barry et al., 2007). For this study, this means that the data of the two groups is not usable for a direct comparison, especially concerning spectral power. However, that does not mean that for the sake of exploring the potential of the method in relation to older adults, the different intensities of the two studies and groups are not interesting to examine. An overview of the set-up and procedure of the two studies can be found in Table 2 below.
Table 2: Overview of the relevant set-up similarities and differences between the two studies
Time Group 1 Group 2
pre Baseline interview
Cognitive Reserve Index questionnaire Verbal Fluency task: Dutch
Digit Span Task
pre Capping in the lab
pre Color-Shape-Switch Task
pre Resting-state recording: eyes-open Resting-state recording: eyes-closed pre Verbal Fluency task: English
IELTS (listening, speaking) Peabody vocabulary naming
Picture Naming task: Spanish
course English course: 3 months blended learning (online: LOI + in person: biweekly communication class)
Spanish course: 10 days online (Rosetta Stone)
post Postintervention interview
Verbal Fluency task: Dutch Digit Span Task
post Capping in the lab
post Color-Shape-Switch Task
post Resting-state recording: eyes-open Resting-state recording: eyes-closed post Verbal Fluency task: English
IELTS (listening, speaking) Peabody vocabulary naming
Open Verbal Fluency task: Spanish Picture Naming task: Spanish
36
3.1.3. EEG acquisition and preprocessing
For group 1, the EEG signal was collected from a 32-channel EEG cap, four reference electrodes around the eyes (EOGH, EOGV) and two mastoid electrodes at a sampling rate of 512. Group 2 used a 64-channel EEG cap with four reference electrodes around the eyes (EOGH, EOGV), two mastoid electrodes and a ground electrode on the sternum. The sampling rate was 1000 Hz. Both caps followed the 10-20 system (see Figure 5).
To ensure good data quality, impedances were kept below 7 k in group 1 and below 10 k in group 2 (for a review on the effects of impedance on ERP data quality see Kappenman, & Luck, 2010).Since John and colleagues (2009) state that a minimum of 32 seconds of artifact-free data has been shown to be required to obtain reliable power values for resting-state analyses, participants with more than 175 rejected trials were excluded from further analyses.
The preprocessing was carried out offline using the MATLAB (2020a) toolbox FieldTrip (Oostenveld et al., 2011). Before re-referencing the data to the average of the two mastoids, the first and last 3 minutes (group 1) / 0.5 minutes (group 2) were deleted from the data, which is similar to practice trials in ERP research, which are not included in the final analysis either. This was done to dispose of the beginning, due to possible interference with previous interactions/tasks, and the end, because we often noticed signs of drowsiness (alpha waves) towards the end of recordings. Furthermore, this deletion ensured a similar data length in both groups. After the removal of the re-referenced mastoids, we implemented a bandpass
Figure 5: international 10-20 system (Brain Products GmbH, 2012)
37 filter with a high pass of 1 Hz5 and a low pass of 50 Hz to avoid line noise, which typically occurs at 50-60Hz. To further clean the data, we performed visual channel rejection and independent component analysis (ICA) to manually discard eye artifacts, following FieldTrip tutorials (Oostenveld et al., 2011). We only considered components 1-25. The data was then segmented into 2 second intervals with 50% (see Prat et al., 2016). Afterwards, we performed manual trial rejection on the created epochs (see Fieldtrip tutorial). The power spectrum was computed using the Fourier transform with a Hanning taper, which calculates the average frequency power across segments (see Prat et al., 2016). The log-transformed power of each frequency (from 1 to 45 Hz) was used in further analyses. The preprocessing script created for this study can be found in the Appendix.
The frequency bands were defined as follows: delta (1-4 Hz), theta (4-8 Hz), alpha (8-12 Hz), beta ((8-12-30 Hz) and gamma (30-45Hz). Since no participant had an alpha peak below 9 at baseline, we decided against creating individual power spectra based on IAF (see Bice et al., 2020).
Regions of interest (ROIs) were created on the basis of previous work by Kepinska and colleagues (2017), who encountered electrode clusters on the basis of functional connectivity analyses. These electrode clusters also served as basis of
Bice et al.’s (2020) resting-state analysis and are visualized in Figure 6. The ROIs and the associated electrodes for each group can be found in Table 3. The log-transformed frequency band power was calculated for each ROI and used in further analyses.
5 A high pass filter of 1 had to be implemented to restrictions within FieldTrip, which do not allow for a high
pass filter below 1.
Figure 6: Electrode clusters calculated on the basis of phase synchronization (taken from: Kepinska et al., 2017)
38
Table 3: Electrode clusters (ROIs) defined for our study
Name Group 1 Group 2
medial frontal Fp1, Fp2, F3, Fz, F4, FC3, FCz, FC4 Fp1, Fpz, Fp2, AFz, AF3, AF4, F3, Fz, F4, FCz, F1, F2 left frontotemporal F7, FT7, T7, C3 F7, FT7, T7, C3, C5 right frontotemporal F8, FT8, T8 F8, FT8, T8 central Cz, CPz, Pz, P4, CP4, C4 Cz, CPz, Pz, P4, CP4, C4, CPz, CP2
left posterior TP7, CP3, P7, PO7,
O1, P3
TP7, CP5, CP3, CP1, P7, PO7, O1, PO3, P3, P5
right posterior Oz, O2, PO2, PO8, P8 Oz, O2, PO4, PO8, P8
For completeness, we add a quick overview of functions associated with these brain areas. However, this information can only be tentatively included in the discussion, as EEG is known for its poor spatial resolution. In relation to cognitive functioning, but also language skills, the medial frontal area of the brain has been linked to cognitive control (for a review see Miller, & Cohen, 2001). The frontotemporal ROIs have been related to language processing, specifically semantic processing in the left frontal cortex (Gabrieli et al., 1998), including Broca’s area, an area of the brain involved in language production and comprehension (Davis et al., 2008) and language comprehension in the right hemisphere (Federmeier et al., 2008). Furthermore, the right tempo-parietal area has been linked to second language learning rate with resting-state EEG (Prat et al., 2016). The central ROI, which is situated in the area of the parietal cortex, is mostly responsible for sensorimotor processing (for a review see Freund, 2001). Of interest for language studies is also Wernicke’s area, an area of the brain that is believed to be linked to comprehension and phoneme retrieval (for a critical review see Binder, 2017), which is situated in the left posterior superior temporal gyrus. Moreover, the posterior area of the brain has been connected to attention (Behrmann et al., 2004).
