Procedural artificial grammar learning and visual sequence learning in adults with developmental dyslexia: Testing the procedural deficit hypothesis.

(1)

Procedural artificial grammar learning and visual sequence

learning in adults with developmental dyslexia

Testing the procedural deficit hypothesis

(2)

(3)

(4)

Colofon

Author Iris Rebecca Lucina Broedelet

Title ‘Procedural artificial grammar learning and visual sequence learning in adults with developmental dyslexia: Testing the procedural deficit hypothesis.’

Place Amsterdam

Date 25th_{of July, 2016}

Progamme Research Master Linguistics

University University of Amsterdam

Supervisor dr. Judith Rispens

Second reader prof. Jeannette Schaeffer

(5)

Acknowledgements

I would like to take this opportunity to thank dr. Judith Rispens, Imme Lammertink, Merel van Witteloostuijn and prof. Jeanette Schaeffer for their supervision and guidance. Moreover, many thanks to Kyle and Sascha who took the time to read my thesis and gave me some very helpful feedback. I am also really thankful for the support of my friends and family, especially Tom, Delia and Arnoud, oma Lida and Sascha. Finally, I want to thank everyone that took the time to participate in my experiment. Without you all, this thesis would have never been finished!

Iris Broedelet

(6)

Abstract

The procedural deficit hypothesis assumes that developmental dyslexia is (in part) caused by an impairment in procedural memory. The current study has tested this claim by investigating 15 Dutch adults diagnosed with dyslexia and 15 Dutch adults without dyslexia on two types of procedural learning tasks. Moreover, timed word/pseudo-word reading, spelling of Dutch and pseudo-words and Rapid Automatized Naming were tested as well as visual short-term memory and visual selective attention. The aim of the study was to investigate whether procedural memory is impaired in dyslexia, and if procedural memory is connected to literacy ability.

In the auditory AGL task, it was tested whether participants were able to learn the dependency relation between two non-adjacent elements (tep + lut and sot + jik) that were intervened by freely varying elements (X), without explicit instruction (similar to Gómez, 2002). They were asked to respond to one stimulus (either lut or jik) during the training phase of the experiment. The training phase consisted of auditory strings in the form of tep X lut and sot X jik. Thus, if the dependency relation between the two non-adjacent elements was learned, lut and jik became predictable. A random block was added to the training phase, during which the reaction times were expected to increase. After the training phase it was tested whether participants could discriminate between grammatical and ungrammatical strings (tep

X jik or sot X lut). In the SRT task, learning of a pattern of visual-motor sequences without explicit instruction was investigated. A smiley-face could appear in four different positions on the computer screen, and the participants were asked to press the corresponding button as fast as possible on a game controller. The same sequence of 10 elements was repeated during the training blocks, while in the random block the elements were randomized.

Results show that the adults with a diagnosis of dyslexia still experience difficulties on all tasks measuring literacy, while visual short-term memory and visual selective attention are intact. Unexpectedly, procedural learning in the SRT task is intact as well: both participant groups showed a clear learning effect. In the AGL task however, neither of the groups exhibited a learning effect, indicating that the method has to be improved. No significant correlations between literacy variables and procedural learning variables indicating an important role for procedural memory in reading and spelling were found. Thus, the current study does not provide convincing evidence in favour of the procedural deficit hypothesis. Methodological issues could partly explain these findings, but it should be concluded that procedural learning on an SRT task is not impaired in all individuals with dyslexia. More research into non-adjacent dependency learning in dyslexia is necessary.

(7)

1. Introduction

Dyslexia is a common and well-known developmental disorder characterised by persistent difficulty with reading and spelling. Because literacy plays such an important role in our society, this disadvantage can have a large negative impact on someone’s life. Having dyslexia can hinder development in school, enjoyment of activities like reading a book, and may cause low self-esteem and anxiety (Ferrer, Shaywitz, Holahan, Marchione and Shaywitz (2010). In fact, the low self-esteem and anxiety can persist into adulthood or may even be life-long (Ferrer et al., 2010). Children with dyslexia are often still wrongly perceived as lacking motivation to work or being less intelligent than peers (Pavlidou, 2010). Talking to the participants in this study taught me that this was the case for a notable group of them too. These misconceptions can result in a late diagnosis of the disorder, while early intervention is very important (van der Leij, 2013). Moreover, having dyslexia often results in reading less, which could have a negative influence on language and IQ development. Recent research indicates that more general deficits like impairments in motor and executive functions are also associated with dyslexia (Vellutino, Fletcher, Snowling and Scanlon, 2004, amongst others). Considering the vast impact dyslexia can have on multiple aspects of one’s life, research on the subject is of great importance.

Although a lot of studies have been dedicated to reading disorders, uncertainties remain about the core deficit that underlies developmental dyslexia. A promising theory is the procedural deficit hypothesis (e.g. Ullman and Pierpont, 2005; Nicolson and Fawcett, 1990). This hypothesis assumes that dyslexia is (in part) caused by suboptimal functioning of procedural memory. Procedural memory is important for implicit learning of habits and skills and extracting regularities from the environment. The procedural deficit hypothesis is an attractive theory because it has the potential to explain both the specific reading problems of dyslexia, as well as the more general difficulties that are associated with the reading disorder.

In this study, I will focus on the comparison of Dutch adults with dyslexia to adults without dyslexia. The aim of the study is to find out how these groups perform on tasks measuring two types of procedural learning: non-adjacent dependency learning in an auditory artificial grammar learning task and visual sequence learning on an SRT task. Importantly, I will investigate whether these abilities connect to reading and spelling proficiency. This study will shed more light on the question whether procedural learning is impaired in dyslexia, and how literacy abilities are connected to the more general cognitive ability of procedural learning. In the bigger scheme of things, this topic also touches the fundamental question of how general cognitive systems underlie language and literacy (acquisition).

(11)

2. Background

2.1 Developmental dyslexia

Developmental dyslexia (also known as specific reading disability) is a learning disability with a strong genetic component, characterised by severe difficulties in the acquisition of reading and spelling. It is estimated to occur in 10% to 15% of the school age children (Vellutino et al., 2004), although estimations differ across researchers. Children with dyslexia have at least average intelligence, but still experience huge, unexpected problems in acquiring reading and spelling skills. These difficulties are not due to factors like sensory deficits, major neurological damage or socioeconomic disadvantages (Lyon, Shaywitz and Shaywitz, 2003; Vellutino et al., 2004; Ramus and Ahissar, 2012). The reading and spelling problems that characterise dyslexia often persist well into adulthood (Pavlidou, 2010). Young adults with a childhood diagnosis of dyslexia show impaired single-word recognition, non-word reading, phonological processing, naming speed, verbal working memory and have a smaller vocabulary size compared to age-matched adults (Ransby and Swanson, 2003). Despite being defined as an impairment specific to reading and spelling, dyslexia is increasingly often associated with (subtle) impairments in other domains (Joanisse, Manis, Keating and Seidenberg, 2000), leading to hypotheses of a more general underlying deficit.

Both the severity of the reading disorder as well as its exact appearance varies strongly between individuals. Indeed, several subtypes of dyslexia have been proposed, for example ‘surface dyslexia’ and ‘phonological dyslexia’ (Coltheart and Jackson, 1998; Chaix et al., 2007; Ramus and Ahissar, 2012). The heterogeneity of the disorder contributes to the wide array of underlying deficits that have been proposed (Ramus and Ahissar, 2012). The transparency of the native language (the consistency of letter-sound relations) also influences the occurrence and manifestation of dyslexia: ‘deep’ orthographies like English pose more problems than more ‘shallow’ orthographies like Dutch (Pavlidou, 2010). Consistent sound-letter relations facilitate literacy acquisition, and research has shown that children growing up learning English become accurate and fluent readers later than children learning a more transparent language like German (Ise and Schulte-Körne, 2010). Dutch, the language that is studied in the current paper, has a semi-transparent spelling: 26 letters can express approximately 40 phonemes, while in English (a non-transparent language), 44 phonemes can be spelled in 561 different ways (Tops and Boons, 2013).

Although both reading and spelling are affected by dyslexia, research has mostly focussed on the problems in reading that individuals with dyslexia experience. Reading is a complex ability and requires a number of subskills (for a more thorough description and a comprehensive model of the cognitive processes and knowledge involved in reading, see Vellutino et al., 2004). It is assumed there are two routes to written word identification: the orthographic route and the phonological route. In the orthographical route, a word is recognized directly from print, while the phonological route is mediated

(12)

by phonology. Often it was assumed that reading starts out phonological and gradually becomes mainly orthographic. However, research suggests the phonological route is still the dominant route for experienced readers, although both routes contribute to reading (Shaywitz et al., 1999). When learning how to read, the most important task is to establish links between letters (graphemes) and sounds (phonemes). These grapheme-phoneme mappings are a prerequisite for the phonological decoding of text (Pavlidou, 2010). Important for the establishment of these mappings is phonological awareness, the concept that spoken words consist of individual sounds. Moreover, orthographical knowledge, that is to say knowledge about rules and constraints on letters and letter combinations, is important for fluent reading as well. Phonological awareness and orthographic knowledge are reciprocally related and together are important for the acquisition of literacy skills (Vellutino et al., 2004).

The problem for individuals with dyslexia lies mainly with difficulties decoding print, but reading comprehension often suffers as a result from dysfluencies in word identification. To understand the meaning of a sentence or a paragraph, it is necessary that the words are fluently identified, as working memory is limited. A more fundamental deficit in language comprehension is not the case in dyslexia (Vellutino et al., 2004; Hedenius, 2013; Pavlidou, 2010). Severe difficulties with spelling also characterise developmental dyslexia. Spelling requires active production of sound-letter-relations and orthographic knowledge about rules and constraints on letters and letter combinations. Especially in orthographically ‘deep’ languages, ‘orthographic spelling’ often remains difficult for people with dyslexia, even when ‘phonological spelling’ has improved (Ise and Schulte-Körne, 2010). Still, most dyslexia-related research focusses on the phonological route of reading and spelling. Proposed as the underlying deficit of developmental dyslexia is an impairment in phonological processing.

2.2 The core phonological deficit hypothesis

2.2.1 Description of the hypothesis

Since the 1970’s, the dominant view is that the underlying cause of the reading and spelling problems of people with dyslexia is a deficit in phonological processing and the representation of phonological information: the core phonological deficit hypothesis (Stanovich, 1988; Snowling, 1998; 2000; Vellutino et al., 2004; Bennett, Romano, Howard and Howard, 2008, Ramus and Ahissar, 2012; Pavlidou, 2010). Poorly specified phonological representations are hypothesized to cause poor phonological processing, in turn resulting in problems with letter-sound mappings and the decoding of printed words (Snowling, 1998). Adduced evidence for poor phonological processing in dyslexia is extensive. Important arguments in favour of the hypothesis are the findings that individuals with dyslexia to perform poorly on tasks measuring phonological awareness (explicitly attending to and manipulating speech sounds), verbal short-term memory and verbal working memory (for example non-word

(13)

repetition) and automatic retrieval of phonological forms (rapid automatized naming) (Ramus and Ahissar, 2012; Rispens and Been, 2007; Vellutino et al., 2004). Adults with dyslexia still show poor performance on all three tasks, relative to adults without dyslexia (Bruck, 1992; Ransby and Swanson, 2003; Shaywitz et al., 1999), despite receiving special education services such as in the case of the participants in the study of Shaywitz et al. (1999).

As awareness of distinct phonological categories is necessary for the establishment of sound-letter relations, poor phonological awareness as a cause for impaired literacy development intuitively makes sense. However, it is important to note that the relation between phonological awareness and reading ability is reciprocal: learning how to read and reading experience is also important for the development of phonological awareness (Pavlidou, 2010). Thus, poor phonological awareness could also be an effect of the reading disorder. This problem is touched upon in the longitudinal study of Shaywitz et al. (1999). They looked into the development of poor readers, average readers and superior readers from kindergarten through high school. They found that phonological awareness is the most robust predictor of reading ability, and enabled them to discriminate both between poor readers and average readers as well as between average readers and superior readers. They conclude that phonological awareness is an important, continuing contributor to reading and spelling ability in all readers. Importantly, intervention aimed at the improvement of phonological awareness seems to have positive effects on the reading and spelling skills of people with dyslexia (Howard, Howard, Japiske and Eden, 2006: Alexander and Slinger-Constant, 2004; Torgesen et al., 2001). The severe problems in non-word reading that have been revealed in people with dyslexia are also viewed as supporting evidence for underspecified phonological representations; as such a task measures phonological coding without depending on word knowledge. The deficit may be located in the phonological loop of working memory, which is important for retaining and rehearsing phonological information (Vellutino et al., 2004; Pavlidou, 2010). Van der Leij, van Bergen, van Zuijen, de Jong, Maurits and Maassen (2013) show that serial rapid naming and phonological awareness were predictors of later word reading fluency, although only moderately.

The exact nature of the underspecified (also called ‘poor’, ‘weak’ ‘fragile’, etc.) phonological representations that are claimed to cause dyslexia remains unclear (de Bree, 2007). Impaired speech perception is proposed to underlie the phonological deficit in dyslexia (Joanisse et al., 2000). This low-level skill can be tested with for example speech categorization/discrimination tasks. Some studies reveal differences between individuals with dyslexia and controls in tasks measuring speech perception (Godfrey et al., 1981; Werker and Tees, 1987). However, Joanisse et al. (2000) conclude that evidence for this hypothesis is not always convincing, since group differences tend to be small and not generalised to all areas of speech perception and all individuals with dyslexia. In their own study, in which they test children with and without dyslexia aged 7-9 years on speech contrast tasks, the authors found that only a

(14)

subgroup of the children with dyslexia showed poor speech perception. Meanwhile, the majority of the children with dyslexia appears to have normal speech perception capacities in combination with phonological impairments (for example in non-word reading), suggesting that phonological representations can be impaired with intact speech perception. Boets, Wouters, van Wieringen and Ghesquire (2007) report similar findings: only a small group of people with dyslexia, and even a group of people without dyslexia, showed impaired speech perception, indicating that impaired speech perception is not a marker of dyslexia. Still, in a longitudinal study, van der Leij et al. (2013) found evidence for an important role of speech perception in literacy development. They investigated 180 children with a familial risk (FR) of dyslexia (meaning that they had at least one parent with dyslexia) and a control group of 120 children (no-FR) from the age of 2 months up to 9 years. Thus, monitoring the children started before they were learning how to read. There was a significant correlation between early speech processing and later reading fluency. Early auditory processing abilities enabled the authors to discriminate not only between FR children and non-FR children, but also between FR children that later received the diagnosis dyslexia and children who did not. Thus, these results indicate that suboptimal speech perception is a factor in the impaired literacy development of individuals with dyslexia.

2.2.2 Criticism against the core phonological deficit hypothesis

Although evidence in favour of the core phonological deficit hypothesis seems plentiful and convincing, critical remarks have been made as well. For example, cases have been reported of individuals with dyslexia showing average proficiency on phonological tests, while still having problems with reading fluently. In the other way around, phonological problems are not always accompanied by reading problems (Pavlidou, 2010; Joanisse et al., 2000). Also, Nicolson, Fawcett, Brookes and Needle (2010) claim that therapy aimed at phonological processing yielded disappointing results (in contrast to Howard et al., 2006).

Ramus and Ahissar (2012) review studies about dyslexia, with special attention to areas in which individuals with dyslexia show normal performance. They argue that to pinpoint the cause of poor performance on certain tasks, it is important to find test conditions, deviating in one crucial parameter, where performance is normal. Although people with dyslexia have repeatedly been shown to exhibit problems in phonology, normal performance on phonological tasks has been found as well. Most noticeable is the lack of observed abnormalities in speech production (Ramus and Ahissar, 2012: Marshall, Harcourt-Brown, Ramus and van der Lely, 2009). Ramus and Ahissar (2012) argue that as speaking is an important function of phonology, a core phonological deficit should be noticeable in this area as well. Moreover, single picture naming, relying on lexical phonological retrieval, is not slower for people with dyslexia compared to controls (Ramus and Ahissar, 2012: McCrory, 2001). This contrasts with the finding of impaired rapid automatized naming in individuals with dyslexia, suggesting that the difficulties with

(15)

RAN result from the “sequential nature of the task” (Ramus and Ahissar, 2012, p. 107). Moreover, as is stated in the previous section, evidence for poor speech perception (like categorical perception) is not unambiguous: multiple studies do not find group differences, or only find impaired performance in a subgroup of children with dyslexia or in certain test circumstances, even when noise is added to speech perception experiments (Ramus and Ahissar, 2012: Ziegler et al., 2009, amongst others). Also, non-word repetition, while repeatedly shown to be impaired in dyslexia, is normal when the non-words are short (1-3 syllables), indicating accurate perception and production is possible when there is less pressure on verbal working memory (Ramus and Ahissar, 2012: Marshall and van der Lely, 2009).

Ramus and Ahissar (2012) stress the fact that a lot of tasks that are intended to measure phonology, are often complex and also require more general cognitive abilities than just phonological (de)coding. Ramus, Marshall, Rosen and van der Lely (2013) investigated phonological abilities in children with SLI and dyslexia, and did a factor analysis to investigate how these skills relate to overall linguistic performance. Their results lead to a distinction between phonological representations and phonological performance. Both these abilities explain unique variance in the linguistic ability of children with dyslexia and SLI, indicating that phonological skills are not a single construct. Research suggests that exactly phonological performance, which requires phonological representations and (an) additional skill(s) like meta-cognitive skills, short-term memory or serial retrieval, might be affected by dyslexia, while phonological representations themselves are relatively normal (Ramus et al., 2013: Wagner and Torgesen, 1987; Ramus and Szenkovits, 2008). In the same line, Ramus and Ahissar (2012) hypothesize that, instead of having incomplete phonological representations per se, the problems due to dyslexia may rather be suboptimal usage of these representations, as a result of poor detection of regularities in spoken and written language. Nicolson et al. (2010) suggest attention should shift from explaining the problems seen in dyslexia with a core phonological deficit, to investigating the learning mechanisms that underlie the acquisition of phonology.

This line of thought is supported by findings of more general (learning) deficits in dyslexia. Dyslexia seems to be linked to impairments in skills that cannot be directly related to a core phonological deficit, for example impairments in motor functions, executive functions, visuoperceptual functions and fundamental language problems not directly related to phonological processing (Hedenius et al., 2013, Vellutino et al., 2004; Pavlidou, 2010; Ramus and Ahissar, 2012; Bennett et al., 2008; Lum, Ullman and Conti-Ramsden, 2013, amongst others). Motor impairments in people with dyslexia have been reported most often (for the first time by Denckla, Rudel, Chapman and Kreger, 1985), although there is variation in findings across studies (Chaix et al., 2007). This growing pile of evidence of deficits related to dyslexia that are not phonology-related, has triggered proposals of a more general/fundamental underlying cognitive deficit. As poor performance has been found in a large variety of tasks, a lot of different theories have been proposed (Ramus and Ahissar, 2012). For example, hypotheses have been set up about

(16)

dysfunctions in rapid temporal processing, attention mechanisms, magnocellular function and general processing speed (Ramus and Ahissar, 2012: Tallal, Miller and Fitch, 1993; Stein and Walsh, 1997). A promising proposal is the procedural deficit hypothesis, stating that a deficit in procedural learning underlies dyslexia. This hypothesis is discussed in the next section.

2.3 The procedural deficit hypothesis

2.3.1 Description of the hypothesis

The procedural deficit hypothesis proposes that people suffering from dyslexia (but also people with specific language impairment (SLI)) have an impaired procedural memory system, causing (or contributing to) the problems with reading and spelling, and the more general cognitive and motor deficits they exhibit (Ullman, 2001, 2004; Ullman and Pierpont, 2005; Nicolson et al., 2010, Nicolson and Fawcett, 1990; 2007, amongst others). Importantly, this hypothesis implies that not a language-specific but general (in fact, not even unique to humans) learning mechanisms underlie language acquisition. Thus, the procedural deficit hypothesis in an attempt to explain all or most problems associated with dyslexia by means of one general underlying deficit, and to unite explanations of dyslexia and SLI. The fact that dyslexia and SLI show considerable overlap has led to the idea that both disorders might be attributed to the same underlying deficit, albeit in a different gravity (Kerkoff et al., 2013; Ullman and Pierpont, 2005). Both disorders have a strong genetic component, and there is a high comorbidity between SLI and reading problems: as many as 50% of the children diagnosed with SLI have been estimated to have a reading disorder (Hedeinius, 2013). Moreover, like individuals with dyslexia, children with SLI often exhibit deficits in more general cognitive and motor skills (Hedenius, 2013: Hill, 2001; Rechetnikov and Maitra, 2009).

Procedural learning (which is a form of implicit learning) is the ability to learn cognitive/motor habits and skills without explicit awareness or instruction, resulting in implicit, automatically used knowledge – for example learning how to ride a bike. Procedural memory mainly depends on the cortico-striato-cerebellar brain network (Pavlidou, 2010). It is important for the extraction of structural regularities and abstract rules from input, allowing us to adapt “our behaviour to recurring and sequential patterns in our environment” (Pavlidou, 2010, p. 52, referring to Goschke and Bolt, 2007). Procedural memory contrasts with declarative memory, which is involved in retaining events/experiences (episodic memory) and facts (semantic memory) (Ullman, 2004). The distinction between the two memory systems is confirmed by cases of amnesic patients that have severely impaired declarative memory while procedural skill learning is intact (Pavlidou, 2010: Cohen and Squire, 1980). In general, declarative knowledge is consciously retrievable, while procedural knowledge is implicit and used in an automatic way.

(17)

According to some researchers, the declarative and procedural memory system are also important for language. Declarative memory underlies the representation of idiosyncratic and word-specific information (the mental lexicon), while procedural memory supports the acquisition and use of “rule-governed aspects of grammar, across syntax, morphology and phonology” (Hedenius et al., 2013, p. 3925). Ullman (2004) describes procedural memory as a computational system, supporting the acquisition and use of (linguistic) regularities, for example “combining items into complex structures that have sequential and hierarchical relations” (p. 246). In this view, procedural memory is important for language acquisition, as it helps to discover the underlying system in the huge amount of immensely varied input. Statistical distributions of sounds, syllables, morphemes and words can be learned to discover the underlying structure of the language (Saffran, 2003; Arciuli and Simpson, 2012). An example of statistical learning by young infants acquiring their first language is word segmentation. Children use statistical information about the distribution of speech sounds to discover word boundaries in the continuous stream of speech (Saffran, 2003).

Nicolson and Fawcett (1990) and Nicolson et al. (2010) propose the ‘Automatization deficit framework’, entailing the idea that individuals with dyslexia have difficulties with automatizing skills. Automaticity, “the end product of procedural learning” (Nicolson and Fawcett, 2007), results in effortless use of skills. A deficit in automatizing skills would result in a slow, effortful execution of skills like reading (Nicolson and Fawcett, 2010). This is in line with Shaywitz et al. (1999), who report that adults with dyslexia read accurately but slow (non-automatic). Laasonen et al. (2014) and Hedenius (2013) suggest two routes how an impairment in procedural memory could negatively influence literacy development: causing the phonological difficulties that in turn cause the reading problems, or directly disadvantaging the acquisition of orthographical knowledge (see Figure 1).

Although explicit instruction is important in learning how to read, procedural learning plays an important role as well (Gombert, 2003; Pavlidou, 2010; Ise, Arnodldi, Bartling and Schulte-Körne, 2012). Arciuli and Simpson (2012) for example showed that procedural learning is related to reading ability in adults and typically developing children. Procedural learning might play an important role in the automatization of phoneme-grapheme mappings and in the acquisition of implicit phonological rules and categories: the detection of distributional patterns is important for phonetic categorisation (Pavlidou, 2010; Jiménez-Fernández, Vaquero, Jiménez and Defior, 2011; Kerkhoff et al., 2013; Wanrooij, 2015). Moreover, procedural learning is important for the acquisition of distributional patterns in orthography and probabilistic graphotactic constraints that are not explicitly taught. For example, /ɛt/ is more often written as ‘ette’ (as opposed to ‘ète’, ‘aite’ or ‘ête’) after the letter ‘v’ than after the letter ‘f’ in French (Pacton, Fayol and Perruchet, 2005). These types of constraints are not taught explicitly, but are important for spelling and automatic reading. Moreover, while many specific spelling conventions cannot be captured in explicit rules, children do have implicit knowledge about the distribution of letters and letter

(18)

patterns which they use during spelling (Ise et al., 2012: e.g. Pollo et al., 2009). Pacton Perruchet, Fayol and Cleeremans (2001), show that 80% of the French 1st_{grade children participating in their study decide} that befful looks more like a real word than bekkul (double ‘f’ is allowed in French spelling, while double ‘k’ is not). These children have not received formal instruction about reading yet. In the same line, Ise et al. (2012) show that poor spellers have difficulties acquiring distributional properties of letter strings and frequent letter chunks, using a written artificial grammar learning task. In the following sections, a summary of previous research on procedural memory in dyslexia is provided.

Figure 1 – Two routes of how impaired procedural memory could lead to reading/spelling problems.

2.3.2 Previous research on procedural memory functioning in dyslexia

2.3.2.1 Artificial Grammar Learning tasks

Procedural learning is often tested with Artificial Grammar Learning (AGL) tasks (for the first time by Reber, 1967; furthermore e.g. Saffran, Aslin and Newport, 1996; Gómez, 2002). Such tasks usually consist of a training/familiarization phase during which participants are exposed to a continuous stream of letters/speech sounds or sequences of visual shapes. The stimuli are constructed based on particular ‘grammatical’ rules. For example, particular sounds/letters or syllables are allowed to occur consecutively, while certain combinations never occur. During the training phase, participants are not explicitly informed about an underlying pattern in the input. After the training phase, it is tested if the participants are able to discriminate between grammatical and ungrammatical strings of sounds/syllables or shapes. If participants are able to do this, this implicates procedural learning of the grammar to which they were exposed in the training phase. Young infants are already able to discriminate between strings of sounds that are grammatical or ungrammatical according to the ‘new grammar’ in an AGL task (Saffran, 2003). A

Suboptimal procedural memory Disrupted acquisition of phonological categories/rules Problems (automatization) sound-letter mappings Difficulties reading/spelling Suboptimal procedural memory

Poor detection of cues for spelling/automatized

reading

(19)

characteristic of these type of AGL tasks is that participants usually have no explicit knowledge of the rules of the new grammar, as procedural learning results in implicit knowledge (Pavlidou, 2010).

AGL tasks can involve several types of grammars, for example a grammar containing non-adjacent dependencies. Non-non-adjacent dependencies are frequent in language, for example in patterns as ‘isX-ing’. There is a morphosyntactic relation between is and –ing, but the dependent elements are interrupted by variable linguistic material (for example in he is happily singing). Young children (aged 1;7) are able to learn these non-adjacent dependencies in their native language when the distance between the elements is no longer than three syllables (Kerkhoff et al., 2013). Gómez (2002) conducted an artificial grammar with three-element strings like aXb and cXd (e.g. pel wadim rud or vot wadim jic), in which a+b and c+d were non-adjacent dependencies, with a variety of items (X) that could interrupt them. Importantly, the same X elements (like wadim) occurred both in aXb and cXd. Adults are able to learn the dependency between a+b / c+d in this task: they were able to discriminate between grammatical strings like pel X rud and ungrammatical strings like vot X rud. However, learning only took place when the set of middle elements was highly variable (24 different items): a varied set of X items makes the non-adjacent dependencies stand out more. In a head-turn preference version of the experiment, infants aged 1;5 were also able to discriminate between grammatical and ungrammatical strings (Gómez and Maye, 2005). Thus, this type of learning is already available to young infants. However, adolescents with SLI are shown to have difficulties with non-adjacent dependency learning. Hsu, Tomblin and Christiansen (2008) show that adolescents with SLI failed to learn non-adjacent dependencies. Instead, the results indicate that the participants with SLI attempted to “learn the materials by rote memorization”, or in other words, focused on ‘adjacent dependencies’ (Hsu, Tomblin and Christiansen, 2008, p. 6). This finding indicates that procedural learning might be affected in SLI.

The AGL paradigm has been used with individuals with dyslexia as well. For example, Pavlidou, Williams and Kelly (2009; 2010) compared performance of children (9-12 years old) with and without dyslexia on a visual, non-linguistic AGL task. This task consisted of a training phase in which the participants were presented with sequences of symbols subject to specific rules, followed by a grammaticality judgment test. It turned out that the children with dyslexia performed more poorly on a AGL task: they failed to show procedural learning of the grammar rules. In another study, Laasonen et al. (2014) compared adults with dyslexia and ADHD to a control group (Project DyAdd), and found that the adults with dyslexia did not perform above chance in grammaticality judgment in a visual AGL task (with symbol sequences). Moreover, significant correlations were found between explicit knowledge of the AGL grammar and phonological processing and reading in the dyslexia group. Moreover, Ise et al. (2012) investigated visual artificial grammar learning (with symbol sequences) in children with good and poor spelling skills, and found that poor spellers showed less efficient procedural learning. Non-adjacent dependency learning was investigated in infants with a familial risk of dyslexia by Kerkhoff, de Bree, de

(20)

Klerk and Wijnen (2013) with an auditory AGL task. They found that, in contrast to the control group, the at-risk children could not discriminate between grammatical and ungrammatical strings, indicating an impairment in procedural learning. Finally, Grunow, Spaulding, Gómez and Plante (2006) looked at a heterogeneous group of adults with language-based learning disabilities (they had received therapy for language impairment, dyslexia and/or learning disabilities) on an auditory non-adjacent dependency learning task, and found that the language-disabled group was not able to learn the non-adjacent dependencies. The control group benefited from high variability of X (in strings like aXb), while the non-impaired group did not.

2.3.2.2 Serial Reaction Time tasks

Procedural memory is also often tested with the Serial Reaction Time (SRT) task, tapping the procedrual learning of non-linguistic motor sequences (Nissen and Bullemer, 1987). In this task, participants watch four dots/boxes/figures on a computer screen, and have to press corresponding buttons when one of the dots lights up as quickly and accurately as possible (see Figure 2). Unbeknownst to the participant, there is a pre-defined sequence in the elements. Procedural learning of this sequence results in the decrease of reaction time in predictable trials, while reaction times get slower in when a random sequence is implemented. Most participants are not conscious of the fact that they learned a fixed sequence (Hedenius et al., 2013). Like with AGL tasks, participants with SLI are shown to have difficulties with SRT tasks as well (e.g. Lum, Conti-Ramsden, Page and Ullman, 2012).

Figure 2 – Serial Reaction Time task.

Several researchers report impaired performance of individuals with dyslexia on SRT tasks. For example, Vicari et al. (2003) and Vicari et al. (2005) showed that children with dyslexia are impaired on two types of procedural learning tasks: an SRT task and the Mirror Drawing test. In the Mirror Drawing test, the children had to draw a star from an example picture, while looking at the picture and their hand though a mirror. They did this in four different sessions. Procedural learning was measured as an increase in speed and accuracy of drawing. The children with dyslexia performed more poorly on both tasks

(21)

compared to the control group. In another study, Stoodley, Harrison and Stein (2006) showed that adults with dyslexia performed more poorly on a SRT task than adults without dyslexia. Similarly, Menghini et al. (2006) did an fMRI study with adult participants. Individuals with dyslexia showed absence of implicit learning in a SRT task. Thus, adults with dyslexia still seem to have difficulties with procedural learning. Interestingly, Jiménez-Fernández et al. (2011) compared implicit to explicit sequence learning in an SRT task. In the explicit variant, the participants were told to pay attention to a sequence in the stimuli. They found that children with dyslexia were only impaired on the implicit version of the task, supporting the hypothesis that implicit learning is affected in this group. Moreover, they showed normal performance on a visual task measuring non-sequential implicit learning, suggesting that the procedural memory deficit in dyslexia is not extended to all forms of procedural learning.

Correlations between procedural learning on a SRT task and literacy skills have also been found. Howard et al. (2006) compared young adults with and without dyslexia on an alternating SRT task and a spatial context learning task. In the alternating SRT task, sequential dependencies are interrupted by random trials (for example ArBrCrDr), allowing for continuous measure of pattern learning throughout the experiment. In the spatial context learning task, the participants had to find a visual stimulus (a T) on a screen with 12 distractor stimuli, and decide its orientation by pressing a button (left or right pointing). The location of the stimulus was predicted by the spatial configuration of the distractor items. The authors report impaired sequence learning but superior spatial context learning in the dyslexia group. Moreover, a significant positive correlation was found between sequence learning and reading ability. In another study, Hedenius et al. (2013) found intact procedural learning in children with dyslexia, but a specific impairment in the consolidation phase (no implicit knowledge of the sequence after a night’s sleep), using an alternating SRT task. A significant positive correlation between sequence learning and reading ability was found as well. Finally, Bennett et al. (2008) investigated young adults with dyslexia on a sequential learning task. They found no group differences, but a positive correlation between learning scores and reading ability: good readers learned the sequence better than poor readers. This significant correlation was found for both test groups separately. Importantly, based on a meta-analysis of 14 studies, Lum, Ullman and Conti-Ramsden (2013) concluded that individuals with dyslexia have impaired procedural learning abilities in the SRT task relative to the control participants, although the effect size decreases for older participants.

There are studies, however, that have not found evidence for impaired procedural learning in participants with dyslexia. For example, Kelly, Griffith and Firth (2002) found intact procedural learning in a SRT task for adults with dyslexia. Moreover, Rüsseler, Gerth and Münte (2006) investigated adults with dyslexia, comparing performance on a SRT task and a visual AGL task, and found them to be unimpaired in both tasks. However, in both studies (Rüsseler et al., 2006 and Kelly et al., 2002), overall longer response times for the dyslexia group are reported, meaning the participants with dyslexia were slower on the task

(22)

compared to the control group. This indicates that although procedural learning is not absent in the dyslexia group, it is slower than in the control group. In another study, Laasonen et al. (2014), who did find poor performance in the AGL task, found normal performance in the SRT task in adults with dyslexia, relative to a control group. The fact that the participants in these experiments were all adults could be an explanation, as Lum et al. (2013) showed that deviation of procedural learning ability was less strong for adults with dyslexia compared to children with dyslexia, although adults with dyslexia have been shown to have poor performance too. Inconsistent results could be due to the type of sequence that is used and the time interval between elements (Hedenius et al., 2013), as well as the number of exposures to the pattern (Lum et al., 2013). The heterogeneity of individuals with dyslexia may also cause inconsistent results, as the degree of dyslexia differs strongly between participants and possibly between investigated groups.

In sum, individuals with SLI and dyslexia seem to show impairments in procedural learning in SRT tasks and AGL tasks, supporting the procedural deficit hypothesis, although some counter-evidence has been found as well. Several studies also found correlations between procedural learning abilities and reading/spelling abilities. Whether adults with dyslexia are able to learn non-adjacent dependencies in a AGL task remains an open question.

(23)

3. The current research

3.1 Rationale and research questions

The core phonological deficit hypothesis is not a satisfactory explanation of developmental dyslexia for multiple reasons. First, it does not explain the more general deficits that have been observed repeatedly in individuals with dyslexia (Hedenius et al., 2013, Vellutino et al., 2004). Moreover, the phonological deficit hypothesis would predict more serious phonological deficits than actually observed in dyslexia: problems in speech for example seem to be non-existent (Ramus and Ahissar, 2012). A lot of tasks on which individuals with dyslexia perform poorly do not purely measure phonological processing, but require other cognitive skills as well (Ramus et al., 2013). Rather than a core deficit in phonology, a more general learning deficit that manifests in multiple (cognitive) domains is suggested in the current study. The procedural deficit hypothesis is able to explain the difficulties with literacy and the more general cognitive impairments observed in individuals with dyslexia, without denying the specific phonological performance difficulties observed in the developmental disorder. Procedural memory is involved in noticing regularities in (language) input. Learning how to read and spell requires procedural learning (Gombert, 2003): for example for the acquisition of orthographical regularities and (probabilistic) graphotactic constraints that are not explicitly taught, but that are important for automatic reading and spelling of words (Ise, Arnoldi, Bartling and Schulte-Körne, 2012). Moreover, the acquisition of phonological categories and rules is related to procedural memory, as well as the acquisition of motor skills (Grunow et al., 2006; Kerkhoff et al, 2013). In sum, a suboptimal procedural memory system in dyslexia is possibly an explanation of difficulties that individuals with dyslexia experience in literacy, phonology, and more general (cognitive) domains, while not predicting such severe phonological problems as the phonological deficit hypothesis. In other words, its predictions fit the actual observed characteristics of dyslexia better.

Several researchers have shown that children and adults with dyslexia perform more poorly on implicit sequence learning in a SRT task (Vicari et al., 2003, 2005; Stoodley et al., 2006; Menghini et al., 2006; Jiménez-Fernández et al., 2011, Howard et al., 2006; Hedenius et al., 2013; Du and Kelly; 2012; Bennett et al., 2008). Moreover, difficulties with visual AGL tasks are found as well (Pavlidou et al., 2009, 2010; Laasonen et al., 2014; Ise et al., 2012). Non-adjacent dependency learning could be a crucial skill for the acquisition of patterns and categories in language (Kerkhoff et al., 2013). So far, this skill has only been investigated in individuals with SLI (Hsu et al., 2008), children with a familial risk of dyslexia (Kerkhoff et al., 2013) and a heterogeneous group of language-impaired adults (Grunow et al., 2006). All of these groups show impairments in non-adjacent dependency learning. Obviously it is not certain who of the at-risk children in the study of Kerkhoff et al. (2013) will actually develop reading problems, and the group of language-impaired individuals in the study of Grunow et al. (2006) is too varied to claim anything

(24)

about dyslexia in particular. Thus, it remains unclear whether adults with dyslexia would have problems with non-adjacent dependency learning.

To investigate procedural learning ability and its relation to literacy skills in dyslexia, the current study investigates linguistic and non-linguistic procedural learning in Dutch adults with and without developmental dyslexia. To this end, an auditory non-adjacent dependencies AGL task and a visual SRT task will be administered, as well as several tasks measuring literacy skills. Usually, an AGL task consists of a familiarization phase during which the participant receives linguistic input following a certain pattern, and a test phase where the participant is asked to choose between grammatical and ungrammatical strings. A new addition for the current study is an online processing component during the familiarization phase of the task: participants are asked to press a button when they hear a certain ‘word’, the occurrence of which becomes predictable when the pattern is learned. Reaction times will be measured during the familiarization phase. A random block is added to the experiment, enabling us to discover potential learning effects. Thus, instead of only comparing grammaticality judgments between groups, a more detailed comparison of learning throughout the experiment can be made between adults with and without dyslexia. This addition should make the task more sensitive to group differences. Because of the use of two types of procedural learning tasks (auditory non-adjacent dependency learning and visual sequence learning), we will gain more insight into the exact type of procedural learning that may be impaired in this group, and how these abilities are related to reading and spelling ability. As control tasks, visual short-term memory (STM) and visual selective attention will be tested as well. A visual STM task is chosen, because verbal STM is shown to be impaired in individuals with dyslexia (Pavlidou, 2010). It is suggested that working memory/STM as well as attention are related to procedural learning (Janacsek and Nemeth, 2013; Toro, Sinnett and Soto-Faraco, 2005). Especially since the SRT task is a visual task, it is interesting to investigate whether visual STM and visual selective attention are related to performance on this task. In sum, the research questions of this study are as follows:

o Are literacy skills poor in Dutch adults with dyslexia, compared to Dutch adults without dyslexia?

o Is linguistic/non-linguistic procedural learning impaired in Dutch adults with dyslexia?

o Is procedural learning of visual-motoric sequences in an SRT task impaired in Dutch adults diagnosed with dyslexia, compared to Dutch adults without dyslexia?

o Is procedural learning of non-adjacent dependencies in an auditory AGL task impaired in Dutch adults diagnosed with dyslexia, compared to Dutch adults without dyslexia? o How is procedural learning ability related to literacy skills?

(25)

o Is this relation different for adults with dyslexia and adults without dyslexia?

o Are visual selective attention and visual short-term memory related to procedural learning? o Could an impairment in visual selective attention and/or visual short-term memory

explain potential difficulties with the two procedural learning tasks in the participants with dyslexia?

3.2 Hypotheses

The main hypothesis is that procedural memory is impaired in adults with dyslexia and can (partly) explain their difficulties with literacy abilities. Results from previous studies indicate that procedural learning in SRT tasks is impaired in children and adults with dyslexia, thus poor performance is expected on this task for the dyslexia group. Auditory AGL tasks with a non-adjacent dependency pattern have not been tested on adults with dyslexia before. However, adolescents with SLI, children at risk for dyslexia and a heterogeneous group of language-impaired adults have been shown to perform more poorly than controls on non-adjacent dependency learning in a AGL task. As there is an overlap of impairments seen in dyslexia and SLI (Ullman and Pierpont, 2005), this type of procedural learning is expected to be impaired in adults with dyslexia as well. On the other hand, it could also be the case that, while children at risk for dyslexia might be impaired on this task (Kerkhoff et al., 2013), adults with dyslexia have ‘caught up’ with adults without the reading disorder. In that case they might be able to acquire non-adjacent dependencies, possibly by compensating for their impaired procedural memory with their declarative memory system (Nicolson et al., 2010). In other words, difficulty with the acquisition of non-adjacent dependencies might be specific to individuals with SLI and children with/at risk of dyslexia. In any case, the control group is expected to be able to learn auditory non-adjacent dependencies (Gómez, 2002). As the sequence in the AGL task is more complex than in the SRT task (adjacent dependencies versus non-adjacent dependencies), the AGL task is expected to provide more difficulties for the adults with dyslexia. On the other hand, the SRT task relies more on learning motor sequences than the AGL task. Since dyslexia is associated with motor impairments, this could mean the SRT task is more difficult for individuals with dyslexia. However, if dyslexia is indeed associated with a general underlying procedural memory deficit, impairment in both tasks is expected, assumed that both tasks tap procedural learning.

Concerning the literacy tasks, the dyslexia group is expected to perform more poorly compared to the control group, as previous studies have shown that literacy difficulties in dyslexia persist into adulthood (Pavlidou, 2010). As procedural memory is important for the acquisition of literacy skills, performance on both procedural learning tasks is expected to be related to reading/spelling ability: individuals who perform better at the procedural learning tasks, are expected to perform better on the literacy tasks as well. This relationship is expected to come forward for both groups. The two procedural

(26)

learning tasks might tap (slightly) different skills, and thus different correlations between the tasks and literacy-related abilities might come forward. Both type of tasks (SRT and AGL) have been associated with literacy, but no relationship between specifically non-adjacent dependency learning in an AGL task on the one hand and reading/spelling ability on the other hand has been established before. As non-adjacent dependency learning might be a crucial skill for the acquisition of patterns and categories in a language, this relationship is expected (Kerkhoff et al., 2013). Unfortunately, on the basis of the current study it is not possible to gain more insight in the contribution of the two different routes of how an impaired procedural memory can disadvantage literacy acquisition (Figure 1). Both routes are expected to play a role.

Two control tasks are included in the experiment: a task measuring visual selective attention and a task measuring visual short-term memory. These tasks are expected to yield similar performance across groups, as a specific procedural learning deficit is hypothesized in dyslexia. If the control tasks would also yield poor performance in the dyslexic group, this would indicate a more general cognitive impairment which could explain problems with procedural learning tasks. Previous research states that visual STM and visual selective attention might be related to procedural learning ability. If this relationship is found, it is expected to be more profound for the SRT task, due to its visual nature.

In sum, the dyslexia group is expected to perform more poorly on the two procedural learning tasks and the literacy tasks, while performance on the control tasks is expected to be similar across groups. Moreover, procedural learning ability is expected to be related to literacy skills in both groups, while visual short-term memory and visual selective attention are related to procedural learning skills. See Table 1 for a summary of the hypotheses.

Table 1 – Summary of the hypotheses.

Dyslexia Control AGL

SRT Literacy tasks Control tasks

Procedural learning skills → Literacy skills Attention/STM skills → Implicit learning skills

(27)

4. Method

4.1 Participants

In total, 30 adults participated in the study. The experimental group consisted of 15 adults diagnosed with dyslexia (7 female and 8 male participants). Their ages varied between 19 and 36 years old (M = 26.6 years, SD = 5.17). The age at which they were diagnosed with dyslexia varied between 6 and 15 years old (M = 11.9, SD = 3.3). The diagnoses were made during their primary or secondary schooling, by teachers or external evaluators in multiple test sessions. Reading and spelling ability was declared seriously below age-appropriate level, and in discrepancy with general cognitive functioning. Most candidates received therapy aimed at improving their literacy skills. No other language disorders (for example SLI) were reported by the participants in the experimental group. The control group consisted of 15 adults without dyslexia or other language-related disorders (8 female and 7 male participants). Their ages ranged between 23 and 32 years old (M = 26.3, SD = 2.72). There was no significant difference concerning age between the two groups: t(28) = 1.72, p = .86. See Table 2 for an overview of the participants in both groups.

All participating adults are native monolingual speakers of Dutch. Further, the subjects did not have any hearing difficulty or serious visual problems (wearing glasses was not considered as such). Three participants in the test group were diagnosed with ADHD, while two participants in the control group reported this diagnosis. One participant in the control group reported being diagnosed with PDD-NOS (an autism spectrum disorder, which is usually considered a relatively mild form of autism). To avoid previous knowledge of the tasks, (former) Linguistics and Psychology students were excluded from participation. Familiarity with tasks was mentioned by some participants, but this was only the case for the participants with dyslexia who had prior experience with the reading tasks. Familiarity with the AGL and SRT tasks was not reported by any of the subjects. This is crucial since these tasks are directly related to implicitly learned patterns.

Importantly, the control group was matched to the experimental group in terms of level of education, since education strongly influences literacy. In the Netherlands there are three main levels of education after high school: mbo (middelbaar beroepsonderwijs, intermediate professional education),

hbo (hoger beroepsonderwijs, higher professional education) and wo (wetenschappelijk onderwijs, scientific education: university level). The test group and control group both consisted of two people with

mbo as highest education level, seven people with hbo as highest education level and six people with wo

(28)

Table 2 – Participants. LoE = Level of Education.

TEST GROUP CONTROL GROUP

Age Gender ADHD? LoE LoE Age Gender ADHD?

1 28 M yes mbo mbo 26 M yes

2 24 F no wo wo 23 F no 3 21 F no hbo hbo 25 M no 4 35 M no hbo hbo 25 F no 5 27 M no wo wo 28 F no 6 22 M no wo wo 25 M no 7 36 F no hbo hbo 25 F no 8 35 F no wo wo 26 F yes 9 19 M no wo wo 27 F no 10 23 M no hbo hbo 32 M no 11 26 F yes wo wo 32 M no

12 28 M yes hbo hbo 27 M no

13 35 M no hbo hbo 24 F no

14 26 F no hbo hbo 23 M no

15 24 F no mbo mbo 28 F no

4.2 Material

All participants were tested on two types of procedural learning: procedural learning of non-adjacent dependencies in an auditory AGL task, and learning of an implicit sequence of stimuli in a visual SRT task. Moreover, literacy skills were tested: spelling of Dutch words and pseudo-words, fast reading of Dutch words and pseudo-words and rapid automatized naming of colours, letters and objects. As control tasks, visual attention skills and visual short-term memory were assessed. See Table 3 for an overview of the tasks. All tasks will be discussed in depth below.

Table 3 – Overview of the tasks.

PROCEDURAL LEARNING LITERACY CONTROL TASKS

AGL task Auditory non-adjacent

dependencies EMT Fast reading of Dutch words

Corsi block tapping

Visual STM

SRT task Visual sequences Klepel Fast reading of non-words d2 Selective visual attention Spelling Spelling Dutch words and

non-words RAN Rapid naming of colours,

(29)

4.2.1 Procedural learning tests

4.2.1.1 Auditory AGL task

Procedural learning of non-adjacent dependencies was tested with a newly developed auditory AGL task (similar to Gómez, 2002, adapted to Dutch phonotactic rules), taking approximately 20 minutes to administer. The artificial language consists of strings with the structure of aXb, where a and b are non-adjacent dependencies, and X is a variable element: tep X lut (version 1) and sot X jik (version 2). As Gómez (2002) found that adults without language disorders require a highly variable X (24 different elements) to learn the dependency relation between element a and b, the same variability is adapted in the current experiment. Importantly, there is a consistent relationship between two elements that are non-adjacent (tep + lut and sot + jik), while there is no relationship between adjacent elements (X +

tep/lut/sot/jik). An online processing component was added to the task: reaction times were tracked during the training phase of the task, allowing for a more detailed insight in the learning process.

Participants were told to help Appie de aap (‘Appie the monkey’) collect bananas. They could do this by listening to the ‘monkey language’ and pressing the space bar at the right moment, when either the word lut (version 1) or jik (version 2) is heard. They were instructed as follows: ‘Press the button as fast as possible when lut/jik comes’. This instruction was repeated at the start of every block. If the button is pressed correctly, the monkey has a banana in his hand and looks happy. However, when the button is pressed at the wrong moment, or is not pressed when lut/jik is said, the monkey does not receive the banana and looks disappointed. After every block, the participants got feedback about their performance (how many bananas were collected during the block). See Figure 3 for images of the positive and negative feedback during the experiment.

(30)

The task starts with a training phase, consisting of 7 blocks (see Table 4 and 5 for the different blocks in version 1 and 2). The first block is a practice block, during which the 24 different two-syllable X elements (banip, densim, dieta, domo, dufo, fidang, gopem, hiftam, kasi, kengel, kubog, loga, movig,

naspu, nilbo, noeba, pitok, poemet, rogges, seetat, sulep, vami, wadim and wiffel) and the components of the dependency pairs (tep, lut, sot and jik) are presented to the participant randomly. The participants are asked to press the space bar when they hear lut (version 1) or jik (version 2). All X elements are presented once in this practice block, while lut and jik are presented 6 times each, as the participant is practicing to react to those stimuli.

Table 4. Version 1 training phase AGL.

Block 1 Block 2 – Block 5 Block 6 Block 7

Type Practice Training Random Training

Stimuli

All 24 X elements, tep, lut, sot and jik

randomly.

24 times tep X lut, 24 times sot X jik, X = 24 different elements. 24 times lut in random position, 24 times jik in random position.

24 times tep X lut, 24 times sot X jik. X = 24 different

elements. Instruction ‘Press button when you

hear lut’

‘Press button when you hear lut’

Press button when you hear lut’

‘Press button when you hear lut’

Table 5. Version 2 training phase AGL.

Block 1 Block 2 – Block 5 Block 6 Block 7

Type Practice Training Random Training

Stimuli

All 24 X elements, tep, lut, sot and jik

randomly.

24 times tep X lut, 24 times sot X jik, X = 24 different elements. 24 times lut in random position, 24 times jik in random position.

24 times tep X lut, 24 times sot X jik.

X = 24 different elements. Instruction ‘Press button when you

hear jik’

‘Press button when you hear jik’

Four training blocks follow after the practice block. In each of these training blocks, the participants hear 48 strings (in both versions 24 times tep X lut and 24 times sot X jik), with 24 different Xs. Thus, all X elements occur once in the tep X lut construction, and once in the sot X jik construction. Again, the participants are asked to listen to the auditory input and respond to lut (version 1) or jik (version 2). Importantly, the auditory input is the same in both versions, but the participants are asked to respond to different ‘words’. During these training blocks, there is a pattern that could potentially picked up by the subjects. Consistently, every time tep is said, one of the 24 different X elements follows, whereafter

lut follows next. In the same manner, every time sot is said, one of the 24 different X elements follows, whereafter jik follows next. When a participant was to (implicitly) learn this non-adjacent dependency

(31)

relationship, she/he would be able to anticipate the moment she/he has to press the space bar: when

tep/sot is heard, lut/jik will come after X, no matter which element X is. Such a learning effect is expected to manifest as decreasing reaction times during the training blocks. The time in between the word within is string is shorter than the time between two strings, so that the strings are clearly distinguishable.

After the four training blocks, a random block is implemented. During this block, participants again hear 48 strings consisting of three elements, and are still asked to respond to lut (version 1) or jik

(version 2), again 24 times. In contrast to the preceding training blocks however, the order of the elements is now randomized. In version 1 (where the participant respond to lut the input was as follows: 24 times X X lut, 12 times X X jik and 12 times tep X X. In version 2, the input was as follows: 24 times X X jik, 12 times sot X X and 12 times X X lut. Thus, tep and lut, sot and jik are never combined, and neither are tep

and jik or sot and lut. As the occurrence of lut and jik is no longer predictable in this block, reaction times are expected to increase.

After this random block, another training block (equal to the first four training blocks) is implemented. Again, 48 strings (24 times tep X lut, 24 times sot X jik) are presented to the participants, and they are asked to press a button when they hear lut (version 1) or jik (version 2). Reaction times are expected to decrease during this block, since the items to which they have to respond are once again predictable. When participants can take advantage of the returned predictability of lut/jik, they are expected to respond faster in the final training block compared to the random block.

A grammatical judgment task follows after the training phase. The participants are asked to choose, between two sentences, which sentence sounds the most like the new ‘monkey language’ that they have heard for the past 20 minutes. Every trial consists of a grammatical and a ungrammatical sentence (see Table 6 for the sentence types). In the first 4 trials, grammatical strings with known X elements (densim, kasi, kengel and dita) are contrasted to ungrammatical strings with the same X elements. Grammatical strings are structured either tep X lut or sot X jik. Ungrammatical strings have the structure tep X jik or sot X lut. Thus, the grammatical strings in these trials are strings they had heard before during the training blocks. In the latter 4 trials, grammatical strings with new X elements (pergon,

vebong, bispa and dapni) are contrasted to ungrammatical strings with the same X elements. Again, grammatical strings are structured either tep X lut or sot X jik, while ungrammatical strings have the structure tep X jik or sot X lut. The sentences with the new X elements are chosen to test for generalization of the rule: do the participants still choose the string that follows the rule tep X lut / sot X jik when they have not heard those specific sentences before?

Procedural artificial grammar learning and visual sequence learning in adults with developmental dyslexia: Testing the procedural deficit hypothesis.