ON SPEAKING TERMS

(1)

ON SPEAKING TERMS

Examining the relationship between early speech-sound

production abilities and subsequent reading competence

By Marjolein Mues

1, 2

S2456567

Thesis submitted on August 8th_{2019 in partial fulfilment of the requirements for}

the degree of MASTER OF ARTS Research Master Language and Cognition

Supervised by prof. dr. B.A.M. Maassen1 & dr. J. Zuk2, 3

1_{Centre for Language and Cognition, University of Groningen}

2_{Labs of Cognitive Neuroscience, Division of Developmental Medicine, Boston}

Children’s Hospital

(2)

Abstract

In this study, the relationship between speech and reading was investigated. Two main research questions were posed. The first question focused on mediating factors into the relationship between early speech-sound production abilities and subsequent reading outcomes. Using multiple reading measures, we found that this relation can differ greatly depending on which operationalization is used. For most reading measures however, letter-sound knowledge and SES were found to mediate their relationship with speech-sound production abilities. RAN was found to be the most important predictor of reading abilities independent of speech. The second research question focused on group differences between children with speech-sound disorder, children with reading impairment, children with speech-sound disorder and reading impairment and typically developing children. Results showed a comorbidity of 40% between speech-sound disorder and reading impairment. Additionally, children with speech-sound disorder and reading impairment significantly differed from the typically developing children and the children with isolated speech-sound disorder on RAN and SES. Results indicate that children with SSD would benefit from an early literacy screener and that RAN should play an important role in screening children with SSD for reading impairment. Other implications, strengths and limitations and directions for future research are being discussed.

(3)

Introduction

The connections between spoken and written language are well established in that spoken language provides the foundation for the development of reading and writing. (ASHA, 2001, p.1)

As illustrated by this quote from the American Speech & Hearing Association, speech and reading are inherently linked to each other. For instance, children with poor speech-sound production abilities as characterized by a speech-sound disorder, have been found to develop below-average reading skills (Hayiou-Thomas, Carroll, Leavett, Hulm & Snowling, 2017; Cabbage, Farquharson, Iuzzini-Seigel, Zuk & Hogan, 2018). However, in other studies no such effect of speech on reading was found, or the effect was only found for some, but not all children with speech-sound disorder (Hesketh, 2004; Nathan, Stackhouse, Goulandris & Snowling, 2004; Pennington & Bishop, 2005).

It is clear that many unknowns still exist concerning the association between early speech-sound abilities and literacy, while a better understanding of this relationship could not only clarify mixed findings in the existing literature, it could also benefit the clinical practice. Once it is understood why some children with speech-sound disorder develop reading difficulties while others do not, this could stimulate future early literacy screening and intervention strategies for children with speech-sound disorder.

The aim of this thesis is to further investigate the relationship between speech production abilities in early childhood and subsequent reading outcomes to clarify mix results to inform the academic community and to potentially benefit the clinical practice.

In the next chapter a comprehensive theoretical overview is provided to inform the reader on the progress that has been made in research regarding the relationship between speech and reading. This framework will lead to the research question and hypotheses of the present study. I will then set out to explain how I aim to test these hypotheses in the method section. After the method section, the results will be discussed. In the discussion at the end of this thesis I will loop back to the research

(5)

questions and discuss the findings. I will also discuss strengths and limitations of this study and provide directions for future research.

In this thesis, I make use of the data of the Gaab Lab. Not only for the use of the data, but also for the incredible helpful insights and mentorship, I would like to thank dr. Nadine Gaab and her team, in particular dr. Jennifer Zuk, dr. Dana Sury Barot and Jade Dunstan.

(6)

Theoretical background

When children learn to talk, it is normal for them to make many mistakes. Sometimes, these mispronunciations result in adorable family anecdotes. However, at a certain age, speech-sound production errors can become worrisome rather than cute. Generally, children over the age of four are expected to produce almost all speech sounds correctly (ASHA, n.d.). Children who still struggle with the articulation of speech sounds by this age might have a speech-sound disorder (SSD).

SSD is characterized by omissions, substitutions, additions, distortions and/or syllable level errors in the speech of the child (ASHA, n.d.). Examples of errors are for instance ‘wabbit’ instead of ‘rabbit’ (substitution) or ‘poon’ instead of ‘spoon’ (omission). Speech-sounds that are pronounced atypically because of an accent or dialect are not seen as errors, but rather as varieties of the default language (ASHA, n.d.). Speech-sound disorder has typically been resolved by the time the child is nine, sometimes with the help of speech-language pathologists (Shriberg, Tomblin & McSweeney, 1999), although the problem can persist after this age (Wren, Miller, Peter, Emond & Roulstone, 2016).

Historically, speech-sound error patterns were called functional articulation disorders in early study, emphasizing motor-articulatory difficulties rather than involving linguistic aspects (e.g. Travis & Rasmus, 1931; Hall, 1938; McReynolds, Kohn & Williams, 1975). However, the work of Ingram (1976) marked a new way of thinking that led to a paradigm shift in the way how speech-sound disorders were being researched: by placing SSD into the broader context of linguistic and phonetic abilities, thus moving away from the idea that speech-sound error patterns were caused solely by poor motor skills. Today SSD is described in the DSM-5 as persisting difficulty with speech-sound production that cannot be explained because of sensory problems, motoric difficulties or other physical deficits, as inspired by Ingram.

Estimates of the prevalence of speech-sound disorder in school-aged children range from 2.3% to 24.4% (Law, Boyle, Harris, Harkness, & Nye, 2000; Wren, Miller, Peters, Emond, & Roulstone, 2016), with a slightly higher incidence in boys than girls (Shriberg, Tomblin & McSweeney 1999; Wren et al., 2016). The broad range of

(7)

prevalence rates might be caused by a difference in terminology, distinguishing between children with SSD that have delayed speech and those that have disordered speech. Children with delayed speech follow the typical speech acquisition process, making errors that are also observed in their typically developing peers, yet they persistently keep on making these errors after they should be able to produce the speech-sounds correctly. Children with disordered speech do not follow the typical speech acquisition process, but rather produce atypical speech-sound error-patterns. The prevalence of children with delayed speech is higher than those with disordered speech (Dodd, 2011), thus the classification might be causing different incidences. Although the two are distinct categories, as shown by for instance Dodd (2011), the terms are often used interchangeably in the literature, with ‘speech-sound disorder’ as an umbrella term for both (ASHA, n.d.). The current description of SSD in the DSM-5 also does not distinguish between delayed speech and disordered speech. Additionally, there are no clear diagnostic criteria to distinguish between the two. That is why here too we will use the label speech-sound disorder as an umbrella term. As with many developmental disorders, the etiology of SSD is currently unknown. However, some risk factors have been identified. Several studies report family history (e.g. Campbell et al., 2003) and pre- and perinatal complications (e.g. Byers Brown, Bendersky & Chapman, 1986; Dodd & Howard, 2002) as risk factors. Dodd & Howard (2002) found that complications during pregnancy and during birth were significantly more often reported among children with SSD compared to typically developing children. Additionally they found that 28% of the children with SSD in their sample had a positive familial history of SSD compared to only 6% of their typically developing peers. Similarly to developmental disorders such as dyslexia, family risk is an important risk factor for speech-sound disorder. Risk factors did not differ between delayed and disordered subgroups of SSD.

Mixed results have been found regarding comorbid disorders in children with SSD. Some studies report that children with SSD are also at risk for literacy deficits (e.g. Rvachew, 2007; Pennington & Bishop, 2009; Peterson, Pennington, Shriberg & Boada, 2009). Other studies however found that this is only the case for children with SSD and co-occurring language impairment, but not for children with isolated speech-sound disorder (Nathan, Stackhouse, Goulandris & Snowling, 2004).

(8)

Pennington & Bishop (2009) describe the relationship between SSD, language impairment and literacy impairment as complex, and state that each disorder arises from a ‘specific constellation of underlying deficits’ (p. 301), that can be either overlapping or specific for one disorder. Peterson et al. (2009) found that children with SSD have a significantly elevated chance of reading disability, but that chance was even higher when children also had language impairment.

Insight into the nature of the relationship between SSD and reading impairment (RI) is important given the important role that reading abilities play in the modern, society. Children with reading disability can feel severe feelings of anxiety, failure and depression (Riddick, 2009). In schools there is a shift from learning to read to reading to learn. If children have not yet learnt how to read by the time this shift takes place, it will become difficult for them to keep track with the rest of the learning (Gaab, 2019). Additionally, Matthew effects of reading cause children with RI to progressively decline in other learning processes as well, widening the gap between those with and without reading impairment (Stanovich, 2009). This can cause children with RI to have lower academic success (Stanovich, 2009; Gaab, 2019). Research has shown that early identification and consequent adequate intervention of reading impairment can help children at risk for reading failure to achieve age-appropriate reading skills (e.g. Torgesen, 2000). This could in turn remediate the negative effects described above, allowing children to fulfill their full potential. Current identification of reading disabilities however does not take place until after the window of most effective intervention is already closed. This gap between the time point of ideal intervention and the current point at which intervention takes place is also called ‘the dyslexia paradox’ (Ozernov-Palchik & Gaab, 2016) (see also figure 1).

(9)

Similar to speech-sound disorder, the cognitive underpinnings of reading impairment are unknown, but it is thought to be a multi-deficit disorder (Pennington, 2006). The three key predictors of reading phonological processing (Vellutino, Fletcher, Snowling & Scanlon, 2004), letter-sound knowledge and rapid automatized naming (e.g. Georgiou, Manolitsis, Torppa & Lyytinen, 2012; Hulme & Snowling, 2015). Amongst additional known risk factors for reading impairment are family risk, low socioeconomic status and low language skills. (Thompson, Hulme, Nash, Gooch, Hayiou-Thomas & Snowling, 2015; Romeo et al., 2017)

Studies investigating reading ability and isolated SSD (i.e. without language impairment) have found mixed results regarding elevated chances of RI, possibly due to methodological differences and differing definitions (Peterson, Pennington, Shriberg & Boada, 2009). It appears however that literacy development in children with SSD is influenced by a number of linguistic factors, including general language abilities (Peterson, Pennington, Shriberg & Boada, 2009). Children with SSD and good language skills seem to develop better literacy outcomes than children with SSD and low language skills (Nathan, Stackhouse, Goulandris & Snowling, 2004).

In a longitudinal study of Peterson et al. (2009), the authors report that SSD history predicts literacy development, but that some children with speech-sound disorder did not develop any reading difficulties despite a long lasting phonological deficit. Considering the emphasis on phonological awareness in previous literature on reading impairment (e.g. Vellutino, Fletcher, Snowling & Scanlon, 2004), this is a surprising finding. Peterson et al. (2009) argue, however, that their finding is consistent with a multiple deficit approach of reading. Contrary to this finding,

(10)

Overby, Training, Bosma Smit, Bernthal & Nelson (2012) studied the longitudinal data of the Templin Archive (2004), and concluded that phonological awareness is one of the most important mediators for the relationship between speech and reading. Novel in this study compared to previous studies is that we employed a longitudinal design on a continuous sample to investigate literacy across the range of speech-sound production skill and literacy skill rather than focus only on those with SSD at one time point.

In accordance with Peterson et al. (2009), Hesket (2004) found that the majority of children with SSD appeared to have typical literacy development. Carroll & Snowling (2004) however concluded that children with SSD had similar reading outcomes as children with a familial risk of dyslexia. Overby et al. (2012) report that children on the higher ends of the speech production distribution had better literacy outcome than children on the lower ends.

Thus it seems clear that there are complex connections between speech-sound disorder and reading impairment, yet their nature is not yet fully understood. There is little consensus on the relationship between speech and reading and it remains unclear what causes some children with SSD to have reading deficits while others develop typical reading abilities. Could mixed findings be the result of methodological differences on how reading skill is defined?

Reading is a multi-faceted and complicated process that is characterized by different underlying cognitive skills (Scarborough, 2001). As shown in figure 2, many individual factors such as vocabulary and phonological awareness together form an important part of skilled reading.

(11)

Reading skills can be operationalized in many different ways. One could for instance focus on whole word recognition or rather emphasize decoding skills or perhaps prioritize reading fluency. The above discussed studies of Peterson et al. (2009), Overby et al. (2012), Hesket (2004) and Carroll & Snowling (2004) for instance each used different constructs to operationalize the term literacy skill. Note that these authors employed reading tests not with the main goal to assess reading impairment, but only to characterize reading outcomes as part of different research questions. Peterson et al. used the GORT-III (Wiederholdt & Bryant, 1992) to index reading accuracy, reading fluency and reading comprehension and additionally tested both single word reading and single word spelling. Hesketh, who found similar results to Peterson et al., also assessed single word reading and spelling. Contrary, Overby et al. (2012) used data from the Templin Archive, in which word reading was characterized by the Scott Foresman list and the word recognition subtest of the Gates Advanced Primary Reading test. Both these measures are severely outdated with the former one being published in 1956 (Robinson, Monroe & Artley, 1956) and the latter even earlier in 1939 (Gates, 1930). Yet this study is one of few studies specifically informing SLP’s and the clinical practice on the links between SSD and reading.

(12)

Lastly, Carroll & Snowling operationalized literacy skills with a letter knowledge test and a word-reading list consisting of 42 early acquired words.

Additionally, pseudoword reading is often employed in research on reading (e.g. Dietz, Jones, Zeffiro & Eden, 2000) and has even successfully been used as a dyslexia screener (e.g. the Dutch Klepel test by Van den Bos, Spelberg, Scheepstra, De Vries & Swets, 1999). By using pseudowords, the researcher ensures that the child needs to use phonological decoding skills rather than use other strategies such as sight word reading. Investigating the differences between word and pseudoword reading, Rvachew (2007) found that children with differing speech-sound abilities score differently on tasks of pseudoword reading but not of word reading. This consequently implies that there are differences present between these reading tasks. Considering the different skills and cognitive capacities needed for skilled reading, it seems evident that each different reading task (e.g. pseudoword reading compared to word reading or timed versus untimed reading) and each different characterization of reading skills (e.g. accuracy, fluency and comprehension) has an at least slightly different composition of underlying skills and abilities compared to the other reading measures (see also Storch & Whiteburst, 2002; Fiorello, Hale & Snyder, 2006). Campbell and Butterworth (1985) stated: ‘pseudoword reading requires skills that are not needed for word reading’ (p.437). In their case study, they described a girl who was highly literate, but had severe difficulty in reading pseudowords. More recently, Pattamadilok et al. (2015) used transcranial magnetic stimulation (TMS) to investigate processing differences between word and pseudoword reading in the brain. The authors found that starting from an early age, word and pseudowords are distinctively phonetically and semantically processed in the brain. Additionally, fMRI-based research has shown that pseudowords place higher cognitive demands on brain areas related to reading compared to real words (Mechelli, Gorni-Tempini & Price, 2003).

Perhaps in light of these findings, Coltheart & Ulicheva (2018) stated that any theory of reading should include an account of how pseudoword reading is accomplished, given the differences that are present between word and pseudoword reading. After

(13)

all, a skilled reader can identify words, but not pseudowords by sight recognition. This infers that word reading might be more heavily reliant on vocabulary (which would make the reader recognize more words by sight) while pseudoword reading is more reliant on decoding skills. In accordance, Duncan & Johnston (1999) found that phonological awareness was significantly correlated with participants’ pseudoword reading ability. Additionally, it had been reported that rapid automatized naming (RAN) is more related to word reading compared to pseudoword reading (Araújo, Reis, Petersson & Faísca, 2015).

Investigations into timed versus untimed conditions of reading have mostly focused on test validity and the added value of additional time for students with reading disabilities (e.g. Brooks, Case & Young, 2004; Lesaux, Pearson & Siegel, 2006). However, it seems intuitive to expect differences in automatization level that is needed here, given the time constraint. After all: a skill that is automatized takes less time than executing a non-automatized skill because every step has to be consciously taken. Simultaneously, a skill that is not automatized does need conscious execution, taking up not only more time, but also cognitive resources. For example: Shinar, Meier & Ben-Shalom (1998) found that novice drivers in a manually geared car are much poorer in detecting road signs than experienced drivers, because shifting hear is an automatic task for experiences drivers, but not for novice drivers. For reading a similar line of thought can be employed. Skilled readers can rely on their automatization of the reading process, but unskilled readers have to consciously take all the different steps (decoding, letter to sound mapping, etc.) to achieve reading, thus taking up more time. This difference between skilled and unskilled readers is especially highlighted in timed constraint reading, because it is then that the automaticity of the skilled reader is especially needed.

As far as accuracy, fluency and comprehension are concerned, several differing underlying abilities have been found to be of importance. Fluency has been reported to be characterized by a shift from conscious decoding to rapid word recognition (Elhassan, Crewther & Bavin, 2017) and has been strongly associated with rapid automatized naming (Norton & Wold, 2012; Landel et al., 2018). Contrary, reading comprehension has long been associated with vocabulary and language abilities (Davis, 1968; Hayiou-Thomas, Harlaar, Dale & Plomin, 2010; Petscher, Justice &

(14)

Hogan, 2018) and socioeconomic status (SES) (Ransdell, 2012). Reading accuracy is not often separately discussed, but it seems fair to expect different influences here too, such as letter-sound knowledge and phonological awareness, which are also key predictors of reading in general (e.g. Hulme & Snowling, 2015).

While reading alone is predicted by several behavioral measures, most importantly by phonological awareness, RAN and letter-sound knowledge, it remains unclear if these factors remain equally important in predicting reading skill for children with SSD. Literature on the impact of these factors on reading in children with differing speech-sound abilities is still scarce. More research into the relationship of speech and different measures of reading is needed to investigate mixed results and to inform the field about the influence of speech-sound abilities on reading. Additionally, it remains unclear why some children with SSD develop typical reading skills while others lag behind their peers.

Insights from neuroimaging

Apart from behavioral measures, group level differences for children with and without reading impairment have also been found using neuroimaging methods. Using fMRI, Raschle, Zuk & Gaab (2012) found that children with a familial history of dyslexia show less functional activity in temporal-parietal and occipital-temporal regions during a phonological awareness tasks (Raschle, Zuk & Gaab, 2012). Utilizing diffusion tensor imaging, Langer et al. (2017) found that children familial risk of dyslexia displayed lower fractional anisotropy in parts of the left arcuate fasciculus compared to children without a familial risk (Langer et al., 2017). Fractional anisotropy is a number between zero and one indexing the strength of white matter pathways, with zero corresponding to weak and one to strong connections. The arcuate fasciculus is the white matter pathway between Broca’s area and Wernicke’s area. Thus Langer et al (2017) found that children with a familial risk of dyslexia have less strong white matter connections in the pathway between two known language areas in the brain compared to typically developing children. Yeatman et al. (2011) even found that properties of the arcuate fasciculus predict phonological and reading skills in children.

(15)

Especially of interest for reading in the context of speech-sound disorder is the superior longitudinal fasciculus (SLF). Not only has this white matter pathway been found to have reduced fractional anisotropy in dyslexia (Carter et al., 2009; Vandermosten, Boets, Sunaert, Wouters, Ghesquière & Poelmans, 2011), but the SLF has also been found to be associated to play a part in sound-to-articulation mapping and phonological processing during speech production (Dick & Tremblay, 2012; Andreatta, 2018). Children with below average reading skills were also found to have increased fractional anisotropy in the inferior longitudinal fasciculus, possible due to different organization of the reading network (Yeatman, Dougherty, Ben-Shachar & Wandell, 2012). Research into the neural correlates of children with SSD only has found that children with SSD have greater white matter volume in the corpus callosum than their typically developing peers (Preston et al., 2013). The corpus callosum is the largest white matter pathway in the brain, connecting the left and right hemispheres.

Insight into the factors that differ between children with SSD with and without reading impairment can potentially inform the clinical field on the need for early literacy screening amongst children with SSD. Additionally, research comparing children that have both SSD and reading impairment with children that only have reading deficits can give insight into the similarities and differences of impaired underlying cognitive abilities. If both groups appear to have similar underlying deficits, remediation and diagnosis are expected to also be the same for both groups. However, if underlying mechanisms differ, alternative approaches are needed for children with SSD.

Summarizing, there is a need for research investigating the relationship between speech and different measures of reading, and the mediating factors into this relationship that might differ between children with varying speech-sound abilities. Uunderstanding the basic mechanisms behind this relationship would benefit clinical practice, as would a comparison between children with speech-sound disorder and typical reading skills, children with speech-sound disorder and reading deficits, and children with isolated reading deficits.

(16)

1. Is there a predictive relationship between early speech-sound production abilities and subsequent reading skills and if there is, which factors mediate this relationship?

2. Which factors differ between children with speech-sound disorder without reading deficits; children with speech-sound disorder with reading deficits; and children with reading deficits only?

With regard to the first research question we hypothesize there to be a predictive relationship between early speech-sound production abilities and reading outcomes considering previous findings regarding of Peterson et al. (2009) and Overby et al. (2012).

In accordance with Overby et al. (2012), it is hypothesized that phonological awareness mediates the relationship between speech-sound production abilities and word reading. Additionally, it is hypothesized that the other two key-predictors of reading, i.e. letter-sound knowledge and RAN mediate the relationship between speech and reading. Moreover, language abilities are expected to moderate the relationship between speech and reading because it has been found that children with SSD and language impairment developed reading deficits while children with SSD without language impairment developed typical reading abilities (Nathan, Stackhouse, Goulandris & Snowling, 2004).

Furthermore, we hypothesize that there is a difference in mediating factors between speech-sound abilities and reading depending on reading is operationalized Here, we use different conditions of reading (e.g. timed versus untimed and word versus pseudoword) and different reading skills (e.g. accuracy, fluency and comprehension). It is the expectation that there is an influence of RAN in a timed condition of reading compared to an untimed condition. In addition, RAN is also expected to play a role in word reading but not pseudoword reading, as described by Torgesen et al. (1997) and Araújo et al. (2015). Phonological awareness is expected to play a role in pseudoword reading, in accordance with effects found by Duncan & Johnston (1999).

(17)

It is expected that letter-sound knowledge and phonological awareness will be important mediators for the relationship between speech-sound production abilities and reading accuracy. For speech-sound abilities and reading fluency, RAN is expected to be the most important mediator (Norton & Wold, 2012; Landel et al., 2018) and for reading comprehension it is expected that vocabulary, language abilities and SES will be of importance (Ransdell, 2012; Petscher, Justice & Hogan, 2018).

The second research question concerns group differences between three groups of interest (group one, two and three) and a control group of typically developing children:

1. Children with speech-sound disorder in kindergarten who subsequently developed below-average reading skills in second grade.

2. Children with speech-sound disorder in kindergarten who subsequently developed typical reading in second grade;

3. Children with typical speech in kindergarten who subsequently developed typical reading skills in second grade;

4. Children with typical speech in kindergarten who subsequently developed below-average reading skills in second grade;

Comparisons between these groups can potentially provide insights for the scientific field and the clinical practice. It is hypothesized that there are differences between the groups. Behaviorally, it is expected that there will be differences between group scores on the three key predictors of reading (i.e. phonological awareness, letter-sound knowledge and RAN), with the groups with reading deficits scoring less well than those without. In addition it is expected to find lower functional activity during phonological processing for the groups with reading deficits (Raschle et al, 2012). For these groups, it is also expected that there will lower fractional anisotropy in the arcuate fasciculus and the superior longitudinal fasciculus (Carter et al., 2009; Langer, 2017). Increased fractional anisotropy for children with reading deficits is expected in the inferior longitudinal fasciculus (Yeatman et al., 2012). For the groups with SSD, we expect to find higher numbers of fractional anisotropy in the corpus callosum (Preston et al., 2013). Of especial interest is the contrast between the groups with SSD with and without reading deficits, since differing factors between these

(18)

groups might indicate potential risk or protective factors for RD in children with SSD.

Novel contributions

Compared to the current literature, this study brings three novel contributions. First, previous research on the relationship between speech and reading has previously only looked at group-level differences (e.g. Peterson et al., 2009) or a continuous sample (e.g. Overby et al., 2012). This study provides a new perspective by combining both approaches and investigating reading both in a continuous sample with varying levels of speech-sound abilities and comparing different groups classified on their speech and reading abilities.

Secondly, this study incorporates different measures of reading to investigate possible mixed results in previous literature as a result of differing reading outcomes. Lastly, this study is, to the best of our knowledge, the only study combining behavioral methods and neuroimaging techniques to investigate reading abilities among children with speech-sound disorder.

Significance

There are mixed findings with respect to the relationship between speech and reading. By investigating multiple reading measures, I aim to disentangle these findings and contribute to a better understanding of reading in children with speech-sound disorder. Moreover, a better understanding of the relationship between speech and reading might benefit early screening and intervention strategies for children with speech-sound disorder.

(19)

Method

Participants

Data for this study were derived from the larger longitudinal investigation

Researching Early Attributes of Dyslexia (i.e., the READ study). Participants for the

READ study were recruited in pre-kindergarten and kindergarten classrooms in the New England area in the Northeast of the United States. In the United States, children typically start kindergarten at the age of five years old. To ensure an economically, ethnically and religiously diverse sample, participants were recruited from different school types (public district, public charter, private, and religious). Additionally schools were located in diverse areas (e.g. urban and suburban areas). Since the READ study was set up to study dyslexia, children with a familial risk of dyslexia were oversampled in the selection of longitudinal enrolment to ensure a sample with large enough groups for statistical analyses.

Over a thousand children were screened to participate the READ study. Screening took place to ensure that children had received only minimal formal reading instruction prior to the investigation and that they fit the requirements for the study, which were mostly assessed using parental questionnaires, such as the Macarthur-Bates Communicative Development Inventories questionnaire (Fenson, Marchman, Thal, Dale, Reznick & Bates, 2007). This questionnaire is used to assess the child’s vocabulary. Parents see a list of words or combination of words and had to indicate if their child used this word. On the list would for instance be ‘another cookie’ and parents could score this with ‘not yet’, ‘sometimes’ or ‘often’. On different, non-standardized, questionnaires parents can fill in information regarding their child’s handedness, family history, possible language or learning disorders and general background.

Of all screened children, 186 were eligible to participate in the READ study. Participants eligible to participate were those who:

- who were right-handed*;

(20)

- who were reported to have no history of neurological or psychiatric disorders, head or brain injuries;

- who were reported to have no poor vision or hearing in developmental questionnaires;

- who had no contraindications for MRI examination;

- who did not read yet, as indicated by the parents in a developmental questionnaire and their result on a short screener for reading (i.e. a raw score below 5 on the word identification subtest of the WRMT-T (Woodcock, 1998), during which children were asked to read words from a list without time constraints)

- who demonstrated nonverbal cognitive abilities within the typical range (e.g. a standard score over 80) as indicated by the Kaufman Brief Intelligence Test. *Language is strongly lateralized in the left hemisphere of the brain for most right-handed people, but not for a substantial portion of left-right-handed people (eg. Szaflarski, Binder, Possing, McKiernan, Ward & Hammeke, 2002; Szaflarski et al, 2012). At the time of participant selection for the READ study, it was decided that left-handed participants should be taken out of consideration as to not introduce noise to the neuroimaging data. However, in more recent work, researchers have made a compelling case to stop excluding left-handers as they represent a substantial part of the population and the exclusion of left-handers severely impoverishes the data (Willems, Van der Haegen, Fisher & Francks, 2014).

At the time of the READ study, it was decided that many typically developing children that were screened and were found eligible to participate eventually were not enrolled in the longitudinal aspect of the study. This was decided because the difference in sample size between the group with an FR of dyslexia and the group typically developing children would otherwise become very large and pre-processing and analysis of all that data would be very time-consuming and costly. This consideration could possibly have introduced bias in the sample.

For the present study, 61 children in the longitudinal selection were additionally excluded from the present sample because they did not complete the speech assessment that was necessary for this investigation. Of the remaining 125, 11 were

(21)

excluded because they did not complete any tests at the second time point. The present study is thus based on the data of 114 children (60 girls, mean age at the start of the investigation 5;6 (years;months)). For a visualization of the participant selection, see figure 3.

Participants involved in the longitudinal aspect of the READ study were assessed with behavioral and neuroimaging tasks at three different time points until the end second grade. For this study I make use of the data collected at the first and the last time point of the READ study. The first time point took place before or at early kindergarten, i.e. before the start of formal reading instruction, and the second time point took place at the end of second grade three years later, when formal reading instruction was well established. From now onwards, these time points will be called time point 1 (T1) (prior to formal reading instruction) and time point 2 (T2) (after formal reading instruction), see also figure 4.

Figure 4. Visualization of the time points for the present study

Time point 1 Time point 2

Kindergarten Grade 1 Grade 2

- Completed speech assessment

- Completed tests at first and second time point - No formal reading instruction

- Fitting study criteria - Diverse sample

- Oversampling for children with FR of dyslexia Screening READ (N = 1000+) _Longitudinal aspect READ (N = 186) The present study (N = 114) Figure 3. Visualization of the participant selection process

(22)

Participants were retrospectively classified in one of four independent groups:

1. Children with speech-sound disorder but without reading deficits (N = 30) 2. Children without speech-sound disorder but with reading deficits (N = 10) 3. Children with speech-sound disorder and with reading deficits (N = 19) 4. Children without speech-sound disorder and without reading deficits (N = 55) Participants were classified in each of the four groups depending on their speech-sound production scores at time point 1 and on their scores on the reading outcome measures at time point 2 of this study. This retrospective classification was done by the researcher for the purpose of this study and is not a formal diagnosis. Children were classified with speech-sound disorder if their percentage of correctly produced consonants scored below established cut-off points as described by Campbell, Janosky, Dollaghan and Adelson (2007). These cut-off points are based on age and differ per month of age. This procedure is discussed in more length in the next paragraph. Children were classified with reading deficits if they scored at least one standard deviation below the mean on at least one of four word reading outcome measures. Of the 29 children classified with reading deficits, one child scored one standard score below the mean on only one of the reading outcome measures, the remaining 28 children all scored below the cut-off point on two, three or all reading measures. The speech-sound production abilities measure and the reading measures are discussed in more length in the upcoming paragraphs.

Informed consent was obtained by all participants, such that verbal assent was obtained from each child and written consent from each legal guardian. All experimental protocols and procedures were approved by The Institutional Review Boards at Boston Children's Hospital and Massachusetts Institute of Technology and all research was performed in accordance with respective guidelines and regulations. Materials for first research question

The first research question was:

1. Is there a relationship between early speech-sound production abilities and subsequent reading skills and if so, which factors mediate this relationship?

(23)

In order to answer this question, early speech-sound production abilities were indexed by evaluating speech accuracy during a speech assessment at time point 1. This speech-sound production accuracy was eventually linked to subsequent word reading skills that were assessed at time point 2 by using different literacy measures. Possible influencing factors were measured at time point 1 using behavioral measures and self-report measures.

Speech assessment. In order to be able to assess the relationship between early speech-production abilities and literacy as described in the first research question speech production accuracy was determined. Accuracy was evaluated through audio-recorded speech samples derived from two different behavioral tests administered at the first time point (T1): sentence repetition from the Development of the Grammar and Phonology Screening (GAPS) (Van der Lely, Gardner, Froud & McClelland, 2006) and elision from the Comprehensive Test of Phonological Processing (CTOPP) (Wagner, Torgesen & Rashotte, 1999). These tests were chosen because they involved the most verbal responses.

The GAPS serves as a short standardized screening to identify children with a suspected language impairment by addressing the main language skills with which these children are known to have difficulties (Van der Lely et al., 2006). The sentence repetition task within the GAPS addresses a variety of syntactic and morphological structures. To minimize failure by not knowing the words, all words in the sentences have an early age of acquisition (e.g. cat, dog and house), which all children are supposed to be familiar with, regardless of their socioeconomic status (Van der Lely et al., 2006).

The CTOPP specifically addresses phonological processing skills and is used to identify children who score poorly on important phonological abilities and to determine children’s strengths and weaknesses considering phonological processes (Wagner et al., 1999). The subtest elision measures the ability the child to omit or add sound from words. The child was for instanced asked: “what is the word plant without ‘t’?”

(24)

The recordings made of each participant during these tests were analyzed through the Percentage Consonants Correct Revised (PCC-R). The PCC-R is a measure of phonemic accuracy. The PCC-R is derived from the Percentage Consonant Correct (PCC) as proposed by Shriberg and Kwiatkowski (1982) and is calculated in a similar manner. The PCC and PCC-R are calculated by dividing the number of correctly pronounced consonants in a sample by the total number of consonants and multiplying that number by 100 (correct consonants/total consonants*100). For example, in a sample in which the child utters 80 consonants of which 65 are correct, the PCC-R would be 65/80*100 = 81.25%. The PCC and the PCC-R differ from each other in what is considered incorrect. There are three possible classes of incorrect pronunciation, as indicated by Shriberg (1993):

1. Common clinical distortions, which include: labialized /l/ or /r/, velarized /I/ or /r/, lateralized voiced or voiceless sibilant fricatives or affricates, derhotacized /r/ or /e/ and dentalized voiced or voiceless sibilant fricatives. 2. Uncommon clinical distortions, which include: weak consonants, imprecise

consonants, failure to maintain oral/nasal contrasts and failure to maintain appropriate voicing.

3. Omissions and substitutions

Using the regular PCC analysis, consonants in all three categories are marked as incorrect. In the PCC-R however, only omissions and substitutions are marked as incorrect. For example: in a sample with 120 consonants, the child makes 10 clinical distortions and 15 errors of the third category (i.e. omissions and substitutions). The PCC would be 95/120*100 = 79.2% whereas the PCC-R would be 105/120*100 = 87.5%.

Because clinical distortions are included in PCC, this measure can give an indication of the severeness of the disorder, making is an appropriate measure to compare children that have (Shriberg, Austin, Lewis, McSweeny & Wilson, 1997). PCC-R however does not take into account clinical distortions, making this a more appropriate measure to compare children with SSD to children without SSD (Shriberg et al., 1997). For this reason, PCC-R is used as a measure of speech production accuracy in this study.

(25)

Transcriptions of the speech samples were made by researchers with training in speech-language pathology and linguistics. Inter-rater reliability was conducted between two raters with 15% of the sample to verify consistency across raters. A correlation coefficient of > 0.9 was achieved between raters (Zuk et al, submitted). Consonants produced per characteristics of the New England regional dialect were not counted as errors.

According their PCC-R, children were classified as either having typical speech or speech-sound disorder based on the classification established by Campbell, Janosky, Dollaghan and Adelson (2007). Campbell et al. used curve fitting analyses and model selection to establish a model that estimates the expected PCC-R score and the standard deviation at each monthly age. Because error-patterns of children were not classified as either typical (yet delayed) or disordered, it was decided to wield the commonly accepted term speech-sound disorder as an all-encompassing umbrella term rather than differentiating between the two.

The classification of Campell et al. (2007) shows a linear upward monotonic trend for speech-sound production abilities with increasing age. Allowing for precise cut-off points per month of age, the classification makes it possible to distinguish between children with and without speech-sound disorder. The CTOPP and GAPS speech data in this study were analyzed to obtain a PCC-R score per participant. For each participant, PCC-R score was compared to the age-appropriate cut-off points as established Campbell et al. (2007). Children scoring below this age-equivalent cut-off point were classified with SSD, which resulted in 49 children with SSD in our sample.

Most children that struggled with speech-sound production at the first time point and were consequently classified in the sound disorder group had received speech-language pathology remediating their articulation difficulties. This was not documented throughout this study.

Reading outcome measures. Reading abilities were assessed at time point two using several different measures. There was no reading test at the starting point

(26)

since children were too young to be able to read and also did not receive reading instruction yet. Children were assessed with three different literacy tests: a test measuring both word and pseudoword reading in a timed condition (TOWRE-2, Torgesen, Wagner & Rashotte, 2011); a test measuring both word and pseudoword reading in an untimed condition (WRMT-R, Woodcock, 1998); and a test indexing the child’s reading accuracy, reading fluency and reading comprehension (GORT-5, Bryant &Wiederholt, 2011). Of all tests, standard scores were derived with a mean of 100 and a standard deviation of 15. Additionally, a composite score for word reading was computed to assess the relationship between speech and an all-encompassing average word reading score. This composite score was derived for the two word reading tests by calculating the average score on these tests per participant, giving each test equal weight.

Test of word reading efficiency. The TOWRE-2 can be administered within the

age range of six to twenty-four years old and consists out of two subtests: Sight Word Efficiency (TSWE) and Phonemic Decoding Efficiency (TPDE). Both subtests were administered to our participants. For these tasks, the child is asked to rapidly and accurately read aloud as many words (Sight Word Efficiency) or pseudowords (Phonemic Decoding Efficiency) as he or she can within 45 seconds from a list that increases in difficulty. Each correctly read word is worth one point. Thus for instance ten correctly read words would result in a raw score of ten. Raw scores were then converted into standard scores with a mean score of 100 and a standard deviation of 15 based on age equivalents.

Woodcock reading mastery tests - revised. The WRMT-R can be assessed from

the age of five years until the age of 75 years old. Two subtests were administered to assess reading ability: Word Identification (WI) and Word Attack (WA). These subtests are very similar to the ones in the TOWRE: the child is asked to accurately read aloud as many words (word identification) or pseudowords (word attack) of increasing difficulty from a list. Contrary to the TOWRE, the child can read at his or her own pace. Scoring was done exactly like the TOWRE subtest: correctly read words result in one point each. These points are accumulated and make up the raw score. This raw score was converted into a standard score (mean 100, standard deviation 15) based on age appropriate norms.

(27)

Gray oral reading tests. The GORT-5 is appropriate for participants within the age range of six years old until 24 years old. With this test, reading accuracy, fluency and comprehension can be measured. The test consists out of sixteen reading passages with each five open-ended comprehension questions. While the child reads the passage, the administrator times how long the child needs to read the passage and marks the errors the child makes. Children do not receive credit for self-corrections and lose points for repetitions. After each passage, the administrator asks the student the five open-ended comprehension questions. When answering the comprehension questions, the child is not allowed to look back at the text.

The raw score for reading accuracy is the number of words that the child read accurately. The raw score for reading fluency is the sum of the child’s raw score on accuracy and the time in seconds it took the child to read the passages. For reading comprehension, the raw score is calculated by the number of correctly answered questions (each worth 1 point). Raw scores for each of the assessments were transformed to index scores with a mean of ten and a standard deviation of three. Influencing factors for the relationship between speech and reading

Possible influencing factors for the relationship between early speech-sound production abilities and subsequent reading skills were assessed at the first time point. Overby et al. (2012) are replicated by testing the same factors that were evaluated as possible mediators in this relationship. Additionally, accordingly the hypotheses, RAN, SES and sentence repetition, which were not included in Overby et al. (2012), were also included. Thus influencing factors investigated in this study were: phonological awareness, RAN, letter-sound knowledge, language abilities (vocabulary and sentence repetition), non-verbal IQ and SES. The first of these measures was assessed using standardized behavioral tests. SES was indexed using parental questionnaires.

Behavioral assessment. At T1 a test battery was administered to assess children’s cognitive and language abilities. During the assessments, children were not being penalized for articulation error patterns to avoid that children with a speech-sound disorder would receive lower scores due to their speech. All behavioral tests

(28)

(including the reading tests at the second time point) were administered by speech-language pathologists or research-assistants with a relevant bachelor’s degree that received extensive training on each test. An overview of the behavioral assessment can be found in Table 1.

Table 1.

Operationalization behavioral testing T1

Subtest(s) Derived from

Phonological awareness Elision Blending Nonword repetition CTOPP RAN Objects Colors RAN/RAS

Letter-sound knowledge Letter identification WRMT

Language abilities Sentence structure

Complete test

CELF-IV PPVT-4

Nonverbal IQ Matrices KBIT

Phonological awareness. Phonological awareness was measured by the

subtests elision, blending words, and non-word repetition from the Comprehensive Test of Phonological Processing (CTOPP) (Wagner, Torgesen & Rashotte, 1999). The CTOPP is developed for the ages four to twenty-four years old and is used to assess phonological abilities.

In the elision task, the ability to remove phonological segments from spoken words to form other words is tested. The participant first listens to an orally presented word and repeats the word. He or she then hears a sound that is in that word and repeats this sound. Lastly, the participant has to remove the sound from the originally presented word and say the new word. For example: the child is presented with the word tree and is asked to leave out the sound r. The correct answer is then tee. For each correct word, the child receives one point. There are 20 items in this subtest that become progressively more difficult.

(29)

The subtest blending assesses the ability to synthesize sounds to form new words and requires the child to merge a string of individually presented sounds to form a word. The child listens to audio-recorded speech-sounds and is asked to put the sounds together to make a whole word. One repetition of the speech-sounds per item is permitted. An example would be hearing the sounds c – a – t, which makes up the word cat. There are 20 items in the subtest blending that become increasingly difficult, with the most difficult items consisting out of eight speech-sounds for the child to blend. Each correct answer is worth one point.

Non-word repetition is the ability to repeat words without meaning, but which are pronounceable because they follow the phonetic rules of English and sound like real words. For this subtest, the child is orally presented with a nonword, for instance the pseudoword scarbs. The child is then asked to repeat the word exactly the same. There are 18 items in this subtest that become increasingly more difficult. The child is first asked to repeat pseudoword with only three sound, but this can go up to pseudowords with fifteen speech-sounds.

For each subtest, the participant started with the first item. After three incorrectly answered items in a row, the subtest was ended. The raw scores on each subtest were transformed to age-equivalent norm scores with a mean of 100 and a standard deviation of 15. A composit score for phonological awareness was derived per participant by adding up the three norm scores and dividing the total by three, giving each subtest equal weight.

RAN. Rapid Automatized Naming (RAN) has been found to index the

automaticity needed in the accurate and fluent reading process. We used the RAN/RAS test from Wolf & Denckla (2005) to assess RAN. This test is appropriate for the ages of five years old through eighteen years old. Two subtests were used: rapid naming of pictures of objects and rapid naming of colors. For each category, five high frequency stimuli were randomly repeated in an array of five rows with each ten stimuli, resulting in a total of fifty stimulus items. The raw score consisted out of the amount of time in seconds required to correctly name all of the stimulus items. Raw scores were transformed into age equivalent norm scores with a mean of 100 and a standard deviation of 15.

(30)

Letter-sound knowledge. Letter-sound knowledge means knowing which

phonetic sound corresponds each orthographic letter in the (English) alphabet. This was assessed by the subtest letter identification from the Woodcock Reading Mastery Tests – Revised (WRMT-R) (Woodcock, 1998). The WRMT-R can be administered to people within the age range of 5 and 75+ years old. In the letter identification subtest, participants are asked to read alphabetic characters (including diphthongs like

ou as in sound) in different fonts (including bold and cursive), in uppercase and

lowercase. Correct answers are the name of the letter and the letter’s most common associated sound with the letter. Each correct answer is worth one point. Raw scores of correctly read characters were transformed into standard scores with a mean of 100 and a standard deviation of 15.

Language abilities. Two aspects of language were assessed: sentence comprehension

and vocabulary. Sentence comprehension was characterized by the subtest of sentence structure of the Clinical Evaluation of Language Fundamentals-IV (CELF-IV) (Semel, Wijg & Secord, 2003). The CELF IV can be used to assess language abilities of people from the ages of 5 to 21 years old. The subtest sentence structure is used to evaluate children’s understanding of grammatical rules by asking the child to select which of four pictures best represents the sentence read by the researcher. The child for example hears the orally presented phrase the boy has a ball and sees four pictures as shown in figure 5. When answering correctly, the child can either point towards the picture in the top right corner, or say ‘picture B’ or ‘the second picture’, if the pictures were numbered during the explanation of the subtest. Every correct response is worth one point. After four consecutive incorrect responses, the test is ended. Raw scores on this subtest are transformed into age-equivalent scaled scores with a mean of 10 and a standard deviation of 3.

(31)

Vocabulary was characterized using the Peabody Picture Vocabulary Test (PPVT-4) (Dunn & Dunn, 2007). In this test, the child is orally presented with a stimuli word and is then asked to choose which of four pictures best describes that word. The test is developed and normed for people from 2 years old up to 90+ years of age. Participants did not start with the first set of items of the test, but at different entry levels. Each set consists out of twelve items. In total, the test has 19 items with together 228 items. The entry-level set for each participant was established on age and all items in the set had to be answered correctly. If a mistake was made, the examiner assessed one set prior to the set with the mistake until the entry-level set was found. When a participant made eight or more errors in one set, the test was discontinued. Each correctly chosen picture resulted in one point. These raw scores were then transformed into age-appropriate standard scores with a mean of 100 and a standard deviation of 15.

Non-verbal IQ. Non-verbal intelligence was measured using the matrices

subtest from the Kaufman Brief Intelligence Test (KBIT-2) (Kaufman & Kaufman, 2004). The KBIT-2 is normed for people between 4 and 90 years old. In the matrices subtest, the child sees several pictures (matrices) and has to choose which picture matches with a target picture by pointing at the right picture. Naming the picture or giving the number in which the picture appears is also correct. In total, there are 46 items and each correctly answered items is worth one point. The start item is determined based on age in years. After four consecutive incorrectly answered items, Figure 5. Example of sentence structure subtest CELF-IV

(32)

the test is discontinued. Raw scores for correct responses were transformed to standard scores with a mean of 100 and a standard deviation of 15. An example of a question can be found in Figure 6. In this example, the target picture is a chair and the child has to choose which of the five other pictures goes best with the chair. The correct answer would be the picture with a bed on it, because they are both furniture. A test of nonverbal IQ is included in the present investigation to replicate earlier studies such as that of Overby et al. (2012).

Figure 6. Example question of the Matrices subtest of the KBIT-2

Self-report measures: SES. Information regarding socioeconomic status was collected at the first time point using the Barratt Simplified Measure of Social status (Barratt, 2006). In this measure, a composite score is formed using parents’ education and their current (or, in the case of retirement, former) occupation. Parents were asked their education levels and their occupation (both the mother and father were asked to provide this information) using a questionnaire. Additionally to occupation and education, families were asked to provide information regarding their income to calculate a composite for socioeconomic status. This way, more weight is given to families current situation compared to their former education level, which is in accordance with Barratt’s notion that families’ current identity is the most important and characterizes the social mobility that is part of the American culture (Barratt, 2012). The score for SES is not normed: only raw scores are used.

(33)

Materials for second research question Our second research question was:

Which factors differ between children with speech-sound disorder without reading deficits; children with speech-sound disorder with reading deficits; and children with reading deficits only?

In order to be able to answer the second research question, behavioral/environmental

factors and neuroimaging assessments were investigated. The

behavioral/environmental factors investigated are the same ones as discussed in the previous paragraph (i.e. phonological awareness, RAN, letter-sound knowledge, non-verbal IQ, language abilities and SES). The neuroimaging assessments were done using fMRI and DTI and will be discussed in the next paragraph.

Neuroimaging assessment. An MRI scanner was used to conduct neuroimaging assessment at the first time point. Brain imaging using an MRI is a research technique that enables researchers to assess brain anatomy (structural MRI) and brain functions (functional MRI). MRI data acquisition is completely safe for participants, but due to the use of a strong magnet in the scanner, MRI cannot be used on people that haves pieces of metal (e.g. pace makers) in their bodies or pregnant women.

Before the scan sessions were conducted, children were introduced to the MRI scanner with mock scanner training. A mock scanner is a realistic approximation of an actual MRI scanner, but without the magnetic field. This ensures that children and their caregivers can get accustomed to the scanner without any impediments, see also figures 7 and 8.

Imaging data were collected using a 3Tesla Siemens Trio Tim MRI scanner (for comparison: the earth’s magnetic field is approximately .00003 Tesla). Each scanning session lasted approximately 40 minutes, which included breaks as individually requested. During the scan sessions, one parent and a researcher were always present in the MRI room near the child. If necessary, the researcher would remind the child to stay still throughout the scan session because any motion could potentially lead to unusable data. Slight motion artifacts in scans could be digitally corrected using MRI

(34)

software. In figure 9 the testing setting can be seen: the picture is taken from the technician room from which you can see the MRI room and the scanner through the window.

For this study, two types of imaging data were analyzed: diffusion tensor imaging (DTI) data, which is a type of structural MRI that measures white matter properties, and functional magnetic resonance (fMRI) data, which measures task-based brain activity. Structural magnetic resonance (MRI) data measuring grey matter properties were also acquired within the larger READ study but were not analyzed for this study.

Figure 7. Mock scanner Figure 8. Child practising MRI session in

a mock scanner

(35)

DTI. Diffusion tensor imaging can be used to investigate white matter tracts,

or pathways. This is done by measuring the ease with which water molecules diffuse in the brain. White matter tracts can be seen as the ‘highways’ of the brain, because they connect different parts of the brain. These parts are called cortical areas, which are areas that are known to have specific functions, such as motor areas, sensory areas and vision areas. The highways, or neural pathways, make up the structural (or anatomical) connectivity of the brain. On DTI scans, white matter connections are made visible with different colors, depending on the different directions of the neural pathways. This generally results in brain scans that are not only informative, but also look like sheer pieces of art. For two examples, see figure 10.

Isotropy versus anisotropy. DTI measures the diffusion of water molecules. Any

diffusion of liquid, so also water, can be either isotropic or anisotropic. Whether the Figure 10. Pieces of DTI artwork. Left: DTI reconstruction of whole brain white

matter tracts in a healthy adult. Right: DTI reconstruction of the corticospinal tract of a healthy adult. The colored lines represent the different directions of major neural pathways (blue means top to bottom of the brain (or: superior-inferior), red/pink means left to right and green means front to back of the brain (or: anterior-posterior)). Derived from Roceanu, Antochi & Bajenaru (2012).

(36)

diffusion is isotropic or anisotropic depends on the uniformity of the direction of the water. If the water diffuses the same in every direction, it is isotropic (technically, when the diffusion is unrestricted or equally restricted in all directions). This is for instance the case in a glass of water. The diffusion can be characterized by a single diffusion coefficient. However, in biological tissue, such as brain tissue, the diffusion is different for different directions, it is anisotropic. Anisotropic diffusion cannot be described with one coefficient, but with a 3x3 table, or rather a 3D model, which is called the diffusion tensor. These numbers represent coefficients along the x-, y- and z-axis (O’Donnell & Westin, 2011).

Fractional anisotropy. The isotropy/anisotropy distinction is not binary. Instead it is

characterized with a number on a scale between zero and one. This number represents the degree of anisotropy in the process and is called fractional anisotropy. This number tells us how strong the white matter connections between specific brain areas are: zero means that they are weak, and one means they are strong. Research questions generally focus on the association between stronger white matter connections and better skills, such as reading.

White matter tracts in this study. In this study, several well-described white matter

tracts that have been previously associated with reading and language were investigated. Fractional anisotropy along the different tracts was correlated with the reading outcome measures and additionally was compared between the four groups. I investigated the anisotropy of the following white matter tracts:

- The Arcuate Fasciculus (AF): the AF is part of the superior longitudinal fasciculus and connects Broca’s area and Wernicke’s area, which are both

recognized language areas. In

isolation, the AF white matter pathway looks like figure 11. In this figure, the fiber bundles can be well

distinguished.

ON SPEAKING TERMS