Thesis in partial fulfillment of the master program
General Linguistics, University of Amsterdam
Nonword repetition and serial
rapid naming in children with
specific language impairment
Emma Klaassen (5819342)
8th of July 2014
Under the supervision of
Judith Rispens
Annette Scheper
1 Acknowledgements
In conclusion of the current study, I want to express my gratitude towards all the people who have helped during the writing of this thesis.
First of all, I want to thank my supervisors Judith Ripens and Annette Scheper for their guidance and the meetings that raised my motivation every time. Judith, thank you for helping me find an internship, for the useful advice on the content of this study and making me enthusiastic about statistics. Annette, thank you for giving me the opportunity of an internship at Kentalis, for sharing your expert clinical knowledge on language impairment and for the inspiring meetings. Moreover, thank you to all the linguists at the Spraak & Taal Ambulatorium (Speech & Language Centre) at Koninklijke Kentalis Eindhoven who have taught me about clinical practice and who supported me in the research process.
Secondly, I want to thank Willem for his teachings on statistics and Anne for her review of the prefinal version of this thesis. Furthermore, I am grateful to Mieke for her hilarious jokes, her laptop and her culinary support. Thank you, my dear Rob, for helping me in times of laptop emergencies, for the nightly dataset-repair session and for your patience and abundant love. Finally, I owe many thanks to my friends and family (in law) for the mental support during the process of this thesis.
2 Abstract
Background: Nonword repetition and serial rapid naming are two abilities that are impaired in children with SLI. Both involving different levels of processing, these abilities depend heavily on phonological processing. The lexical restructuring hypothesis predicts that the relative predictive strength of these associated skills changes as language develops. Aims: The aim of the current study is to investigate the skills associated with nonword repetition and serial rapid naming performance in a population of children with specific language impairment. Moreover, this research aims to study these skills from a developmental perspective, comparing the predictors for nonword repetition accuracy and serial naming speed in different age groups. Methods & Procedure: 143 Dutch children with SLI (age 6-11 years), treated at the Spraak & Taal Ambulatorium (Speech and Language Centre) at Koninklijke Kentalis, were assessed both linguistically as neuro-psychologically as a baseline measurement before treatment. Apart from a nonword repetition task and a serial naming task, measures for phonological short-term memory, receptive vocabulary and visual memory were included in the analysis. Results: phonological short-term memory and receptive vocabulary were associated with both nonword repetition and serial rapid naming abilities. It was found that the skills contributing to these abilities change as language develops: for serial rapid naming, phonological short-term memory capacity was an important factor in the older children as was vocabulary in the younger group. For nonword repetition performance, the opposite pattern was observed. Conclusion: Nonword repetition and serial rapid naming depend heavily on phonological processing and are associated with phonological short-term memory capacity and receptive vocabulary in a population of Dutch SLI-children. More research is needed to deepen our knowledge about phonological processing in these children and about this heterogeneous disorder in general.
Keywords: Specific language impairment, nonword repetition, serial rapid naming, phonological representations, lexical restructuring hypothesis
3
Prologue 4
1. INTRODUCTION 4
1.1 Specific language impairment 7
1.2 Nonword repetition 8
1.2.1: The nonword repetition (NWR) task 8
1.2.2: NWR in a theoretical framework 9
1.2.3: What the NWR task measures 11
1.2.4: NWR in children with SLI 13
1.3 Serial rapid naming 14
1.3.1: The serial rapid naming (SRN) task 14
1.3.2: SRN in a theoretical framework 15
1.3.3: What the SRN task measures 17
1.3.4: SRN in children with SLI 18
1.4 NWR & SRN: the common ground 19
1.4.1: Lexical restructuring 20
1.4.2: The influence of vocabulary on representations 21
1.4.3: Phonological representations in NWR 22 1.4.4: Phonological representations in SRN 23 1.5 Research rationale 24 2. METHODOLOGY 26 2.1 Participants 26 2.2 Materials 27
Nonword repetition task
Continue benoemen & woorden lezen WISC digit recall
Peabody Picture Vocabulary test (PPVT-III-NL) Block recall WMTB-C
2.3 Procedure 29
4
3. RESULTS 31
3.1 Descriptive statistics 31
3.2 Nonword repetition performance 32
3.3 Serial rapid naming performance 34
3.4 Skills associated with and contributing to NWR and SRN 35
4. DISCUSSION 40
4.1 Discussion of results 40
4.2 Limitations of the study 42
4.3 Clinical relevance 42
4.4 Suggestions for further research 43
5. CONCLUSION 44
REFERENCES 45
5 Prologue
As our scientific knowledge is expanding at a rate like never before, it is important not to lose sight of ‘the big picture’. Since researchers in the scientific world mainly work in highly specialized areas, at the risk of not linking their insights to related topics, it is necessary to connect their findings. By doing so, different insights are no longer isolated but can contribute to a network of knowledge. This kind of research may help to map concepts and theories, creating a better understanding of, in this case, specific language impairment, and a research field in general.
Furthermore, building a network of knowledge also comprises connecting scientific research with the clinical work field. As an cooperation between Amsterdam University and the Spraak & Taal Ambulatorium (Speech and Language Centre) at Konklijke Kentalis (Eindhoven, the Netherlands), this study aims to do just that. Evidence-based practice in which the gap between the scientific world and the clinical practice is bridged, is of great important. In this manner, there should be an ongoing influence and reinforcement of the two, aiming to optimize healthcare for children with specific language impairment.
I: INTRODUCTION
The acquisition of language is an important challenge that children face during development. It is quite remarkable that so many children manage to internalize one or more languages so effortlessly. There is, however, a subgroup of children who, despite unimpaired cognitive functions, experience difficulties in acquiring language. Investigating these children with Specific Language Impairment1 (SLI) proves to be interesting, as by doing so it is possible to gain insight into the normal development of the language system. The ability to repeat non-existing words, referred to as nonword repetition (NWR), and the ability to rapidly name highly familiar items, referred to as serial rapid naming (SRN), are known to be notoriously difficult for the majority of SLI-children (Gathercole & Baddeley 1990; Coady 2013). These abilities depend heavily on phonological processing (Metsala 1999; Coady 2013) and are associated with orthographic decoding skills (Bishop et al 2009). Moreover, both tasks are complex in the sense that they involve multiple levels of cognitive processing and there is no real consensus on what factors mediate nonword repetition accuracy and serial rapid naming speed (Coady & Evans 2008).
1 SLI is diagnosed when a child fails to develop language at the typical rate despite normal general intellectual
6 The general questions addressed in this study are, firstly, which skills contribute to (1) nonword repetition and (2) serial rapid naming performance for different age groups in a population of children with SLI and, secondly, how do these two abilities relate to each other in this population of language impaired children? The aim of this thesis is to investigate the abilities underlying performance on both NWR and SRN tasks. In pursuing this goal, performances on a nonword repetition task and a serial rapid naming task are analyzed in a group of Dutch SLI-children. The relation between these two measures has so far not been studied profoundly and could prove to be interesting from both a theoretical and a clinical point of view. The motivation for this study is therefore twofold.
Scientific curiosity is the first driving factor behind the current study. Despite our growing understanding of the language system, the interconnectedness of children’s various language abilities is a field of study which has many secrets to uncover. At brain level, our language system is highly complex and connective, involving different modules and an inconceivable number of neurons. The question of how different language skills are related to each other, especially in the developing, highly connective and adaptive system of children, is thus an interesting one. Moreover, research on the correlations between linguistic abilities in children with SLI could deepen our understanding of the nature of this language disorder itself.
Secondly, this study is of clinical significance in diagnosing and treating children with SLI. Although standardized linguistic tests aim to assess various language skills, locating the locus of breakdown in the language system proves a real challenge due to the complexity of the majority of tests. That is to say, only few linguistic assessment tools measure only one language skill. Partly due to the complex and multilayered nature of language itself, most of the language tools test a range of skills. In addition, clinicians such as language therapists and clinical linguists know from experience that the scores of some linguistic measures seem to be highly correlated. Therefore, it can be concluded that it is not always known which language skills are actually measured by a certain assessment tool. Knowledge of the correlation between different linguistic tests could therefore be meaningful as clinicians will be better able to identify core problems based on the language profiles of specific language impaired children.
In this study nonword repetition and serial rapid naming are discussed separately at first. For both abilities, a model is presented and factors mediating performance are explicated. Moreover, impairment in these abilities found in children with SLI is discussed. Thereafter, the common ground between NWR and SRN is elaborated on by discussing phonological representations, the lexical restructuring hypothesis as formulated by Metsala (1999) and its influence on both NWR
7 and SRN. As the theoretical framework is set out, the research rationale and methods are explained. Subsequently, the results of the study are presented and discussed. Finally, the findings on the skills contributing to NWR and SRN performance and the relation between the two abilities are discussed in the light of previous research and the clinical relevance of this study is emphasized. This thesis will start, however, with a short, general overview of the research area of specific language impairment.
1.1 Specific language impairment (SLI)
Specific language impairment is a disorder that is based on criteria of exclusion. Only when other factors that could influence language development are ruled out, the label of ‘specific language impairment’ is given. Thus, children diagnosed with SLI have, by definition, no severe deficits in hearing, cognition or neurological problems and display normal non-verbal IQ (Joanisse & Seidenberg 1998). Specifically, language is impaired in these children. SLI surfaces in different forms across languages. Although the problems that children with SLI are facing are dependent on language typology, many symptoms seem to be shared cross-linguistically. Deficits in morphosyntax, phonology and word finding are prevailing in this population.
Although this disorder has been studied extensively, there are still many unresolved issues; even about the nature of this impairment no real consensus has been established. There are many different theories on this fundamental issue, each emphasizing different levels of processing and breakdown. Viewing these accounts from a meta-perspective, two main positions can be distinguished: the domain-specific and the domain-general account.
The domain-specific (DS) account assumes that there is a deficit in the grammatical knowledge of the child. The other viewpoint, the domain-general (DG) account, argues that children with SLI suffer from an information-processing deficit that interferes with different levels of language (Joanisse & Scheidenberg 1998). These two positions make different predictions about SLI. Whereas the DS account predicts that the core deficits of SLI are restricted to grammar, the DG account predicts that SLI should co-occur with other (non-linguistic) deficits such as perceptual or motor impairments and that there is a relation between the general deficits and SLI. The predictions made by the two accounts are, however, not always disjoint. It is, therefore, not easy to distinguish DS from DG empirically.
There are some difficulties when investigating SLI. First of all, determining when language development is actually impaired proves a task of its own. Secondly, because of the exclusion
8 criteria of SLI, the children diagnosed with this disorder have very different behavioural profiles (Joanisse & Seidenberg 1998). This heterogeneity makes it difficult to find conclusive correlations in research. Thirdly, comorbidity with other disorders such as developmental dyslexia and ADHD is high (Williams & Lind 2013). Because of this, it is hard to determine causal relationships between the cause and the consequences of language impairment and to learn more about the nature of SLI. Fourthly, it is difficult to distinguish between a deficit in knowledge and a deficit in processing based solely on experimental data. Experimental tasks can never directly tap representations but only the outcome of some sort of processing. Furthermore, processing and knowledge are not independent, so impaired processing on a language level cannot be taken as evidence for either position (van der Lely 2005). To sum up, SLI proves to be a very diverse disorder with many symptoms, possible causes and a heterogeneous patient group.
1.2 Nonword repetition
Every word we have ever learned was once unfamiliar to us. The first step in word acquisition is the perception and awareness of the phonological form of a word. The capacity to repeat an unfamiliar phonological sequence is one of the most basic, but nonetheless one of the most important, language abilities in word learning (Archibald & Gathercole 2007).
1.2.1: The nonword repetition (NWR) task
The capacity to repeat nonwords, which is thought to mimic the first step in actual word learning, can be assessed in a test known as the nonword repetition task. Although seemingly simple at first glance, the NWR task requires multiple levels of processing and various cognitive skills, among them acoustic analysis, phonological decoding, phonological assembly and articulation. (Coady & Evans 2008). Because of the involvement of several other cognitive processes, a deficit in any of these levels results in disrupted performance on a NWR task. Below average test scores, therefore, do not provide information about a locus of breakdown in the language system. Interpreting the results of a NWR test can thus be problematic (Coady & Evans 2008). As Coady & Evans (2008) state ‘any attempts to ascribe accuracy differences to one or two underlying skills to the exclusion of other supporting skills are doomed to fail.’ (Coady & Evans 2008). Although the complexity of the NWR task can be considered a drawback, the task makes a good diagnostic tool for specific language impairment, precisely because it taps so many underlying skills. It provides a reasonably specific and sensitive clinical marker that can be administered quickly
9 (Conti-Ramsden et al 2001). Thus, as Coady & Evans (2008) state, ‘While using the [NWR] as an identifying tool says nothing about the nature of underlying deficits, it does provide a way to group children so that these deficits can be explored.’
Previously, it was thought that the NWR task was very promising as a pure assessment tool since it involved processing of novel material and was viewed as a content-free language measure (Coady & Evans 2008), hereby providing the opportunity to measure STM capacity without the inference of language. Moreover, since no context knowledge is required for the test, it minimizes cultural and dialectal biases (Coady & Evans 2008). As an presumed indicator of short-term memory (STM) ability, the NWR task was assumed not to draw on prior linguistic knowledge stored in the mental lexicon (Bishop et al 1996). In other words, no long-term lexical knowledge was thought to interfere with this test. This assumption, however, was proven wrong when it was found that lexical and sublexical properties mediated repetition accuracy scores (Edwards et al 2004). Bishop et al (1996) even claimed that ‘the apparent purity of nonword repetition is deceptive’. NWR performance is affected by the extent to which nonwords resemble actual words; nonwords that are close to real words in the sense of phonology or morphology are easier to repeat than those that are not (Bishop et al 1996). Due to these findings, Coady & Evans (2008) argued that the NWR test can be used to examine the structural properties of the mental lexicon, both in typically developing and language impaired individuals. Despite ample research on and with the nonword repetition task, there is still no real consensus on what the task measures (Coady & Evans 2008). In general, the task has been used extensively as a measure of phonological memory, lexical access, speech production and phonological processing.
1.2.2 NWR in a theoretical framework
When studying language phenomena, one should always try to embed linguistic findings into a broader cognitive framework. In young children, the brain is not yet highly specialized and displays a high degree of plasticity and connectivity. Since language and cognition are so tightly linked in the developing brain, the application of broad models of cognition is useful in studying language acquisition. The view of language as an integrative part of cognition has gained popularity since Baddeley proposed his model of working memory in the early 1990’-s (Baddeley 1992) (see Figure 1). His well-known and established model of working memory (Baddeley 1992) is an example of how language abilities can be implemented into a cognitive model. This multiple-component model comprises specialized temporary memory systems, in which memory traces are actively maintained. According to Baddeley’s model, working memory includes a
10 component for the processing of auditory information, the phonological loop, and one for processing visual information: the visuospatial sketchpad2. The phonological loop, more generally referred to as phonological short-term memory (P-STM), in turn, comprises two subsystems: (1) the passive phonological store where auditory information is retained and (2) the active articulatory control system, in which phonological information is rehearsed subvocally and kept active as to avoid time-related decay. Lastly, Baddeley’s model includes a central module, the central executive, which is involved in control and regulation of the working memory (Baddeley 1992) The working memory system does not operate in isolation. Rather, this multiple-component system is connected to the mental lexicon in long-term memory (LTM) where word forms and associated meanings are stored. This connection to LTM is not displayed in Figure 1.
Figure 1: Baddeley’s model for Working Memory
In order to repeat a nonword, several levels of information processing must be passed through. First of all, the speech stream must be accurately perceived. Once this is has been accomplished, the nonword is broken down into smaller segments. Consequently, the nonword must be kept active in P-STM long enough to formulate and execute an articulatory plan (Coady & Evans
2 In 2000, Baddeley added another component to his model: The episodic buffer (Baddeley 2000). This additional
11 2008). The phonological loop is the most important subcomponent for NWR. Since this rehearsal process is crucial for NWR performance, the phonological loop is essential for this task. NWR accuracy is, therefore, seen as reflective of phonological loop functioning.
1.2.3 What the NWR task measures
The two main factors mediating NWR accuracy will be discussed: P-ST and receptive vocabulary. Furthermore, the influence of phonotactic probability (PP) on NWR scores will be explained.
1. P-STM
The initial idea proposed by Gathercole & Baddeley (1990) almost 25 years ago that NWR performance was purely an index of P-STM, has long been discarded. The finding that deficits in NWR scores persisted even when differences on an independent measure of STM were taken into account (Achibald & Gathercole 2007) gives support for the claim that NWR performance depends on more than P-STM alone. However, P-STM is still considered to be an important factor mediating NWR accuracy, as it can be concluded that repetition accuracy is correlated with phonological storage capacity. Performance on a NWR task decreases as the P-STM becomes more inefficient and the functioning of the phonological loop declines (Archibald & Gathercole 2007). This phonological storage capacity is not only determined by the endurance of the phonological string but also by the processing of speech. Gathercole (2006) argues that children with SLI may suffer from a double deficit, combining an impairment of P-STM storage capacity with a problem in processing novel auditory stimuli.
2. Receptive vocabulary
It has long been clear that NWR accuracy and measures of receptive vocabulary are correlated. This is plausible since NWR mimics actual word learning (Coady & Evans 2008). The relationship between these two linguistic measures is bidirectional but not constant throughout development. The associations between NWR scores and vocabulary size change with age and lexical development, with NWR dominating influence before the age of 5 but a lesser influence after this age (Coady & Evans 2008). Moreover, in younger children, NWR and vocabulary knowledge are mediated by a third variable, phonological representations, whereas for older children, it is only vocabulary knowledge that supports NWR scores (Coady & Evans 2008).
On the other hand, the relationship between NWR abilities and vocabulary size in the SLI population, seems to be unidirectional (Coady & Evans 2008). Munson et al (2005) found that on
12 a NWR task SLI-children performed similarly to a group of typically developing (TD)-children matched on receptive vocabulary scores. When we look at the study of Botting and Conti-Ramsden (2001), however, the opposite pattern is not observed: children with similar non-word repetition abilities don’t necessarily have similar standard vocabulary scores (Botting & Conti-Ramsden 2001). Thus, it seems that for SLI-children, vocabulary size predicts NWR scores but that repetition ability does not predict vocabulary size (Coady & Evans 2008).
3. Sublexical effects: Phonotactic probability
The frequency of occurrence of individual phonemes and phoneme combinations within a language, is referred to as phonotactic probability (PP) (Storkel et al. 2006). PP expresses the probability that a sequence of sounds will occur in a lexical item (Edwards et al 2004). Children are sensitive to these phonotactic properties and this sublexical effect influences NWR accuracy (Edwards et al 2004). Edwards et al (2004) found that in TD-children and –adults, nonwords constructed of phonemes occurring and co-occuring frequently in a language (high PP) are repeated more accurately than nonwords with low frequent phonemes (low PP). Moreover, children with smaller vocabularies display a larger influence of frequency on accuracy than children with larger vocabularies (Rispens, Baker & Duinmeijer in press).
Studies have been performed to investigate whether children with SLI are subject to the same sublexical effect of PP. Research supports the idea that the facilitatory sublexical PP effects in children with SLI are slower and less accurate but the magnitude and facilitation due to PP is similar to that of TD-children (Edwards et al 2004). Rispens et al (in press) studied groups of children with and without SLI and reading impairment (RI) and found an effect of PP and this effect was especially pronounced in children with RI. This effect of low PP was significantly larger for this group of children compared to the SLI-only and the TD-group, regardless of similar vocabulary scores. They concluded that even though there is an indication that NWR is associated with vocabulary, itself does not predict the sensitivity for PP. Moreover, the group of children with RI-only was more negatively affected by low frequency phoneme sequences in the NWR task despite a mean overall performance on the task that was comparable to the TD-group. Archibald & Gathercole (2009) argue that the NWR task taps P-STM, referring to a deficit in the functioning of the phonological loop as the temporary storage in the phonological buffer. The difficulties that SLI-children face with language learning can, according to these researchers, be associated with P-STM impairment. Metsala (1999), however, states that it is not phonological memory but rather phonological sensitivity that mediates NWR. Phonological sensitivity refers to
13 the process of breaking down phonological strings into smaller units. In other words, the ability to segment a phonological string into phonemes. This In her study, she found that both factors contributed uniquely to NWR performance in a group of 4-5 year old TD-children. For the 3-4 year olds, however, only phonological sensitivity accounted for unique variance on the NWR task. Lastly, Bowey (2001) argues that the distinction between memory and sensitivity cannot be made since they are both manifestations of an underlying phonological processing ability (Coady & Evans 2008).
Although there is agreement on the statement that language abilities and phonological memory are related, there is, however, no consensus on how these factors are related and what the direction of influence is. While Gathercole and Baddeley (1990) argue that phonological memory determines language skills, other researchers state that phonological memory is a consequence of language abilities (Coady & Evans 2008). The question whether phonological working memory is really distinct from phonological knowledge or that both encompass phonological awareness, although interesting, is beyond the scope of this study.
1.2.4 NWR in children with SLI
It is well-known that children with language impairment face great difficulties when asked to repeat a unfamiliar phonological string. Even younger TD-children who are matched on language abilities outperform children with SLI on a NWR task. Thus, scores on the NWR tend to be disproportionately impaired in SLI children. When independent measures of STM are taken into account, the deficits in nonword repetition seem to persist (Archibald & Gathercole 2007). Despite the heterogeneity of the SLI-group regarding area of difficulty and severity of the impairment, NWR scores fall below the average in the vast majority of this group. Furthermore, even in individuals whose obvious language impairments have resolved and who score within the normal range on expressive language, NWR scores tend to remain below average (Bishop et al 1996), revealing the persisting underlying cognitive deficit in phonological processing.
Naturally, these findings give rise to the question why SLI-children experience difficulties with repeating nonwords. Since the NWR is a complex task, breakdown can occur at many different levels. First of all, perceptual deficits might influence repetition accuracy but quite some studies have discarded this possibility (Gathercole & Baddeley 1990; Coady & Evans 2008). It seems plausible that children with SLI do not experience problems in perceiving minimal phonemic pairs when embedded in short (non)words. In longer (non)words, however, phonemic contrasts
14 are less accurately discriminated. Secondly, phonological encoding seems to be less efficient in SLI-children, leading to less robust phonological traces in memory and impeded repetition accuracy. At the next level of processing, Archibald and Gathercole (2007) propose that poor performance on the NWR task is the result of impairment of the STM. Despite the heterogeneity of the SLI population, the co-occurrence of P-STM deficits is a fairly consistent finding across many studies (Achibald & Gathercole 2006; Bishop et al 1996; Conti-Ramsden et al 2001). Moreover, Archibald and Gathercole (2007) argue that deficits in nonword repetition are more marked than those in standard measures of STM. Additionally, mild difficulties in motor planning and articulation have been reported and could contribute to deficits found in SLI-children on nonword repetition. In conclusion, as Coady & Evans (2008, p.21) stated: ‘the difficulty experienced by children with SLI in repeating non-words appears to stem from difficulty with all components of the task, including speech perception, phonological encoding, phonological memory, phonological assembly, and articulation’.
1.3 Serial rapid naming
Serial rapid naming (SRN) involves the naming of a continuous serie of highly familiar items as rapidly as possible (Bowey et al 2005). SRN has shown to be most strongly associated with reading abilities (Bowey et al 2005). First investigated by Dencka & Rudel (1974), the interest in serial rapid naming has greatly increased the last decades.
1.3.1 The serial rapid naming (SRN) task
Rapid naming abilities can be measured in a rapid naming task usually consisting of several subtasks. These tasks, involving the naming of color names, digits, letters, pictures and simple words (monosyllabic and high frequent), have been interpreted as the speed with which lexical items can be accessed and articulated. Since the task entails only highly familiar items, it is assumed that these naming responses are overlearned or automatized (Bowey et al 2005). Naming speed is conceptualized as the integration of attentional, perceptual, phonological, as well as semantic, motoric and memory subproceses. In order to achieve speed in naming, all these processes require precise timing within each component and across all components (Wolf et al 2000). Like the NWR task, SRN is thus also a complex task. In naming tasks, these processes are cascaded and there is no consensus on what is actually measured by rapid naming tasks (Bowey et al 2005). Interpretation of results in SRN is therefore not simple. Naming speed has generally
15 been assumed to be reflective of a range of abilities including phonological processing, lexical access and global (non-linguistic) processing speed and is indicative of the efficiency of these processes. Naming speed has also been found to correlate with phonological awareness (Kirby et al 2006). However, it should be emphasized that phonological processes represent only a subset of the multiple processes involved in naming (Wolf et al 2000).
As was done in the current study, a distinction is frequently made between alphanumeric naming speed, indicated by the rapid naming of letters and digits, and symbolic, semantic or non-alphanumeric naming speed, usually indicated by the naming of colors and pictures (Denckla & Rudel 1974). Lastly, it should be noted that the rapid reading of words gives an indication of reading speed and literacy fluency.
1.3.2 SRN in a theoretical framework
Serial rapid naming is a complex task involving visual information processing and rapid lexical retrieval. Wolf et al (2000) discuss a model for visual naming in which several factors are incorporated. First of all, visual processing at multiple levels takes place, analyzing high and low spatial frequencies. This allows for identification processes and integration of information about the stimulus with knowledge from long-term memory in the form of mental representations. The specificity and quality of these representations influence the speed of processing and lexical access. In the model, affective factors – additional components that may affect integration of information – are represented. When all information is integrated, an articulatory plan is formulated and motor commands translate this plan into an item name. For all these subcomponents, processing speed is of influence. Slow processing in one component, can lead to disrupted timing in the next. This whole process occurs within 500 ms in TD-adults (Wolf et al 2000) (See Figure 2).
There are two main accounts for serial naming speed: the orthographic processing efficiency account and the global processing speed account. Whereas the first argues that impaired naming speed may be an indicator of difficulties in orthographic rather than phonological processing, the second account proposed that SRN measures purely reflect global processing speed (Bowey et al 2005). So far, there is still no consensus on one theory of serial naming speed that can account for all the findings that have been found.
16 Figure 2: A model for visual rapid naming (Wolf et al 2000)
17 1.3.3 What the SRN task measures
Regardless of the high complexity of SRN, it is commonly assumed that there are two main factors that the SRN task measures: lexical retrieval and general processing speed. Both factors will be discussed in this section.
1. Lexical retrieval
Lexical retrieval is an important factor in SRN. Naming speed is often constrained by the speed with which a certain lexical item can be retrieved from the mental lexicon. Kirby et al (2006) argues that if SRN reflects lexical access efficiency, ‘it may thus be an index of how much and how quickly information may be entered in working memory’ (Kirby et al 2006, p.462). Involving a number of separate processing components (Messer & Dockrell 2006), lexical access in itself is a complex process. Determining the locus of impairment and considering the way in which deficits might impact the process of retrieval, is essential in understanding lexical retrieval. There is, however, no detailed developmental model of lexical access. Adult models have been constructed but these models are not applicable to children (Messer & Dockrell 2006) since their cognitive and language system are less specialized and still more connective and flexible.
For adults, naming is assumed to include three stages: object identification, name activation and response generation (Messer & Dockrell 2006). It is clear that to produce a word, both semantic information, at the level of the lemma, and phonological information, at the level of the lexeme, have to be activated. At the lexeme level, phonological processing includes the activation and selection of the phonological form of a word. Models of lexical retrieval suggest that activation spreads from the semantic to the phonological level and emphasize the importance of fast and accurate lexical selection at both levels. Moreover, the role of competitors influencing selection has been found at the lemma and the lexeme level (Messer & Dockrell 2006).
2. General processing speed
General processing speed is a very broad term referring to the efficiency of information processing in the cognitive system. This processing speed is of importance in all aspects of language and language acquisition. In a task of naming speed, its role is even more emphasized.
18 Processing speed can be administered using reaction times (RT’s) for a wide range of tasks including several modalities.
Some correlation studies have been executed in the last ten years on the relation between naming speed and other linguistic and cognitive abilities. One of them is a study by Bowey et al (2005) who investigated the relation between naming speed (alphanumeric and nonsymbolic), word reading, phonological processing, global processing speed, articulation rate and block recall. A wide range of tasks were administered and it was found that each measure of global processing speed, phonological processing ability and word reading, but no measure of serial naming speed or articulation rate, correlated with block design (Bowey et al 2005). This 2005 study gave support that the association between alphanumeric naming speed and word-reading skill is predominantly dominated by phonological processing ability. According to the authors, neither the global processing speed nor the orthographic processing efficiency account were able adequately to explain this association (Bowey et al 2005).
1.3.3 SRN in children with SLI
As we have seen, serial rapid naming is mediated by lexical retrieval and general processing speed. Both of these factors are problematic for most of the SLI-children (Messer & Dockrell 2006; Miller et al 2001). Many studies have found that children facing language difficulties in terms of word-finding difficulties (WFD), SLI and dyslexia, are slower in naming than control groups. Children suffering from severe reading abilities also show deficits in SRN (Bishop et al 2009).
First of all, lexical access is impaired in many children with SLI: WFD are among the most common symptoms of SLI (Messer & Dockrell 2006). The current understanding is that these difficulties are the result of impairment in the processes underlying lexical retrieval. It was found that the rapid naming of letters and numbers was also deficient in a group of children with WFD. This group was, however, only more slow with semantically complex stimuli (Messer & Dockrell 2006.
Secondly, there are indications that general processing speed is also impaired in children with language deficits. Whereas in children with WFD, there is support for the claim that slowing is solely language-based, for SLI-children slowing seems to extend to general information processing (Miller et al 2001). In children with dyslexia, there is no consensus on whether generalized slowing, independent of phonological ability, contributes to their language difficulties (Messer & Dockrell 2006). The relation between impairment in processing speed and language impairment is a difficult one. Furthermore, Lahey et al (2001) argued that for children with SLI,
19 there is no direct linear relation between general speed of processing and severity of the language impairment (Lahey et al 2001). It was also found that some children with SLI do not appear to show deficits in general information processing (Miller et al 2001). Thus, although the influence of generalized slowing in more domains (Coady 2013) is still a matter of debate (Miller et al 2001; Lahey et al 2001), the contributing factor of inefficient phonological processing and lexical retrieval is evident (Coady 2013).
Despite the presence of language deficits, Bishop et al (2009) found a subgroup of SLI-children who performed within the normal range on a serial rapid naming task. Importantly, this group was not simply the less severely affected ones. In this study, children with and without language impairment and with and without reading impairment were investigated. It was concluded that there is a subgroup of SLI children who learn to decode words and nonwords accurately. In other words, it was observed that when children were not impaired on SRN, they also performed within the normal range on a NWR task. These children were characterized by semantic and syntactic problems: weak vocabulary and poor sentence comprehension. Moreover, deficits in phonological processing were observed but these were not severe.
1.4 NWR & SRN: A common ground
Although nonword repetition and serial rapid naming are two distinct abilities mediated by different factors, they share some underlying skills. In this chapter, the common ground between NWR and SRN is formulated. This common ground is twofold: first of all, phonological representations represent a shared factor. Secondly, according to Bishop et al (2009) both abilities involve phonological decoding. In this section, phonological representations and lexical restructuring is elaborated on.
Phonological representations – the abstract, mental representations of a phonological string – develop as children acquire language. These representations are of great importance in both NWR and SRN. Both abilities can be defined in terms of representations. Whereas NWR entails the creation, storage and articulation of novel phonological representations, SRN comprises the automatized retrieval of existing lexical representations3 from LTM.
3 The difference between phonological representations and lexical representations is that the latter also incorporates
20 1.4.1: Lexical restructuring – from holistic to segmental
In the early stages of vocabulary acquisition, new words are represented in a holistic manner, possibly in terms of acoustic or articulatory patterns (Munson et al 2005). Metsala (1999) proposed that there is a shift from holistic representations towards more segmented phonemic representations as age increases and language development proceeds. According to Gathercole (2006), ‘the child is forced to improve the economy of its organization by shifting towards employing more analytic representations of sublexical structure, relating either to syllables or phonemes’ (Gathercole 2006, p.529). This assumption, referred to as the lexical restructuring hypothesis (Metsala 1999), states that as lexical restructuring takes place, a decrease in sensitivity to larger phonetic units (syllables) should give room for increased sensitivity to smaller phonetic units such as phonemes (Coady & Evans 2008). Furthermore, the more segmented the representations are, the more flexible they become, thereby facilitating phonological generalizations (Metsala 1999). In the process of lexical restructuring, representations become more robust and fine-grained. As this restructuring process takes place, production of phonemes are characterized by variability. Once representations become more independent of context, children gain facility with incorporating them into novel patterns and creating transient phonological representations (Coady & Evans 2008). Children with SLI are assumed to have more holistic and fragile representations. This could be due to their small vocabularies. Moreover, it could be the case that these children did not have enough opportunity to develop precise representations since they need more exposures in order to extract the constituents of a phonological string. This could be the result of difficulties with the formation of phonological representations due to problems with the storage in working memory.
With the shift towards more segmented representations, phonological awareness (PA) grows. PA is defined as the ability to access sublexical phonemic units. As Metsala (1999) proposes, PA is constrained by these underlying representations. That is to say, sublexical units cannot be accessed when representations are holistic. The emergence of segmented representations is thought to be a prerequisite for the development of phonological awareness. In other words, the same mechanisms that drive segmental restructuring also plays a role in the emergence of PA (Metsala 1999). Metalinguistic knowledge facilitates the repetition of nonwords and word learning (Gathercole 2006).
A short remark on phonological representations has to be made at this point: phonological representations are difficult to measure and cannot be directly explored but have to be observed in surface behaviour. In studying linguistic output, one can only derive information
21 about the phonological representations.. It is even a matter of debate whether they can be inferred from repetitions of nonwords at all. It may be impossible to distinguish between the representations themselves and the processes operating on them (Coady & Evans 2008).
1.4.2: The influence of vocabulary on phonological representations
As the lexical restructuring hypothesis proposes, the refinement of lexical representations is related with vocabulary size. Children with larger vocabularies have more segmented, and therefore more flexible, phonological representations. As vocabulary knowledge expands, there is a need to discriminate between an increasing number of phonologically similar entries. In other words, phonological knowledge of representations incrementally emerges as lexical development proceeds. Receptive vocabulary can, therefore, be interpreted as an indirect measure of phonological representations.
Edwards et al (2004) propose that if children acquire phonological representations based on generalizations over the lexicon, the quality of the representations is directly linked with vocabulary size. Representations of familiar sublexical patterns can be accessed more easily and they can be more flexible rearranged into novel patterns. Smaller vocabularies provide less support for abstracting knowledge about phonemes, independent of specific contexts. Children with SLI tend to have smaller vocabularies since they require more exposures before generalizations can be made. Representations in these children are therefore thought to be more fragile.
Although there is consensus on that vocabulary growth and lexical restructuring are associated, the issues of how these properties are related and the direction of influence on each other is unresolved. Two main hypotheses can be distinguished, each making different claims about the interplay between vocabulary and representations (Gathercole 2006).
First of all, the phonological storage hypothesis, as supported by Gathercole (2006), states that the quality and endurance of phonological representations determines vocabulary acquisition. Variation in the first variable, can explain variable rates of vocabulary growth observed across individuals. According to Gathercole (2006), any factor that impacts the quality of phonological storage, will influence the forming of representations.
Secondly, the lexical restructuring hypothesis, as endorsed by Metsala (1999), claims that restructuring of representations is driven by vocabulary growth. Representations have to become more segmented as pressure to discriminate between similar sounding words increases. Words
22 with many similar phonological neighbourhoods (neighbourhood density: ND) require more specified phonological representations. More specifically, when two word forms are very similar and can only be distinguished by one phoneme, representations have to be specific enough to mark this difference. The increase in ND during lexical development is the main mechanism driving segmental restructuring of phonological representations.
Whether lexical restructuring drives vocabulary growth or vice versa, is difficult to assess. It seems plausible that there is not a unidirectional but a bidirectional influence between these two properties. Furthermore, it is questionable if the representations can be separated from the lexical effects since phonological development occurs in the context of lexical development and vice versa (Coady & Evans 2008).
1.4.3: Phonological representations in NWR
Summarizing, phonological STM, receptive vocabulary and PP all mediate nonword repetition performance. Rispens and Baker (2012) studied the relative contributions of these factors to NWR performance in children with SLI and/or reading impairment (RI) and two groups of TD-children.. It was found that for the 5-6 year old TD-children, lexical restructuring was of great importance. The authors concluded that P-STM plays a significant role in NWR performance for the group of 5-year-old children, but it appears that the role of phonological representations is a better predictor of NWR at age 5 years than at age 8 years. Thus, Rispens & Baker (2012) propose that the predictive strength for NWR performance of the quality of phonological representations changes during development, decreasing as language development proceeds/less predictive for NWR performance over time (Rispens & Baker 2012).
Furthermore, Metsala (1999) argues that nonwords that are low in PP are more sensitive to segmental representations than nonwords with high PP. If poor performance on the NWR stems from the underspecified phonological representations, demands on segmental recombination will be greatest for low PP nonwords. Difficulties observed in repetition in children with small vocabularies should thus be greatest for especially those nonwords (Metsala 1999; Edwards et al 2004). Moreover, Edwards et al (2004) studied the relationship among PP, age, articulatory ability and vocabulary size on a NWR task. They found that it was the measure of expressive vocabulary size rather than age that was the best predictor of overall repetition accuracy and PP effects. The larger the vocabulary, the smaller the effect of PP on repetition accuracy (Edwards et al 2004). This finding gives support for the hypothesis that children with larger vocabularies have more context-independent representations. Moreover, it gives support
23 for the idea that phonological templates support the creation of novel phonological representations. As Edwards et al (2004) propose: ‘Children with larger vocabularies may have both more practiced phonetic representations of the particular sequences from having encountered them previously and more robustly abstracted representations of individual phonemes’ (Edwards et al 2004, p.25).
Summarizing, there are two variables that mediate the link between vocabulary and nonword repetition. First of all, P-STM, the ability to keep incoming phonological information active in memory. Secondly, phonological representations largely determine performance on nonword repetition tasks. P-STM and the phonological representations are not entirely independent from each other. It is only possible to keep representations of novel items active in the phonological memory when they are robust. More specified representations will result in more flexibility in encoding phoneme sequences and thus better representations of the nonword. There is a bidirectional influence of lexical development and phonological representations: vocabulary growth leads to more segmented representations and vice versa (Metsala 1999).
1.4.4: Phonological representations in SRN
The specificity of phonological representations are also important for SRN abilities. Kirby et al (2006) found that there is a correlation between phonological awareness and SRN speed. Phonological awareness, in turn, is associated with the quality of phonological representations. is constrained by phonological representations. Another study, executed in 2009, found that in children, performance on rapid naming tasks correlates with early reading skills, when phonological awareness (PA) and verbal IQ are controlled (Lervag & Hulme 2009). Together with measures of PA, non-alphabetic SRN abilities – measured before reading instruction has begun - has been found to provide a strong predictor of later literacy skills in children (Lervag & Hulme 2009). According to Lervag & Hulme (2009), global processing speed cannot account for these findings. Generally, it can be concluded that naming speed correlates positively with age and with familiarity of the item. According to Coady (2013), these effects are presumably mediated by the quality of the underlying lexical representations. These representations become more robust and specified with increasing age and familiarity.
Secondly, Coady (2013) studied word-frequency and PP effects in rapid naming tasks in children with and without language impairment. There was a main effect of group and of word frequency found. Moreover, results suggested that children with SLI were more vulnerable to
24 increased competition associated with words with high PP. These children were slower in naming of pictures whose labels were low-frequent words with high PP (Coady 2013).
Summarizing, phonological representations and their specificity play an important role in both NWR and SRN. The specificity of the phonological representations increases as language develops. More specifically, this process is assumed to be driven by vocabulary growth. Due to this restructuring process and the effect it has on NWR and SRN, it is interesting to study these two abilities from a developmental perspective.
1.5 Research rationale
The aim of this study is to analyze the underlying skills required for rapid naming and nonword repetition in SLI-children. The current study does not make any claims about NWR as a possible clinical marker since performance is analyzed relative to the other children from the Speech and Language Centre at Kentalis. The lexical restructuring hypothesis (LRH) (Metsala 1999) argues that the specificity of phonological representations changes as language develops. Moreover, Rispens & Baker (2012) found that the relative strength of factors contributing to NWR accuracy changes with age. These findings make it interesting to study different NWR and SRN within a population of SLI-children from a developmental perspective to investigate whether the relative contribution of skills changes as language development continues. The research questions are formulated as follows:
Which skills contribute to (1) NWR and (2) SRN performance for different age groups in a population of children with specific language impairment (SLI)? How do NWR and SRN relate to each other in this population?
First of all, it is hypothesized that receptive vocabulary and phonological STM are associated with both NWR and SRN performance assuming that in the youngest children, lexical restructuring is of great influence on NWR and SRN scores. It is predicted that performance on both tasks is most strongly associated with receptive vocabulary in the youngest children and with P-STM in the oldest children. As age increases and language abilities develop, the dependence on phonological representations is expected to decline (Bishop et al 1996) and since receptive vocabulary is an indirect measure of these representations, the predictive strength of vocabulary is expected to decrease as age increases. Moreover, as visual processing is still developing in the youngest age group, it is predicted that for these children Block Recall is associated with the rapid
25 naming of colours and pictures. Furthermore, in the oldest children, it is predicted that phonological representations are sufficiently specified and that NWR and SRN performance is only constrained by P-STM capacity.
Secondly, it is hypothesized that nonword repetition and serial rapid naming share some underlying skills since they both rely on phonological decoding and the quality of phonological representations. It is, thus, predicted that there is a correlation between NWR scores and different subtests4 of the SRN task in our population of SLC-SLI children.
4 The subtests of the rapid naming task are grouped into three composite measures: (1) semantic naming speed
including the naming of pictures and colors; (2) alphanumeric naming speed including the naming of letters and digits; and (3) literacy naming speed including two subtests for rapidly reading words.
26 II: METHODOLOGY
2.1 Participants
The children in the SLI group were recruited via the Speech and Language Centre of the Koninklijke Kentalis in Eindhoven in The Netherlands. This centre is specialized in communication disorders as the result of language impairment, hearing impairment or behavioural disorders such as ADHD and autism. Children with speech and language disorders who do not benefit from regular language therapy are tested and treated intensively for a period of three months. The department offers short-term, specialized treatment. Moreover, this treatment has a multidisciplinary approach in which linguists, language therapists, neuropsychologists and communicative councilors collaborate as a team to diagnose and decide on treat strategies for the children with language impairment. The data used in this study is collected at two moments in time: first, at the start of enrollment as part of baseline diagnostics and secondly, 6 months after finishing the specialized treatment.
Most children who receive treatment at the Speech and Language Centre of Kentalis are diagnosed with SLI. This, by definition, means that their nonverbal IQ scores fall within the normal range, that no hearing impairment has been ascertained and that their language abilities are below the age norms. Children who have an additional behavioural disorders or hearing impairment are also given treatment.
To qualify for treatment at the Speech and Language Centre of Kentalis, children visiting a speech and language therapist should have made no substantial progress within the last six months. Secondly, it should be evident that the child’s progress at school is stagnating. As a result to these admission requirements, it is plausible that our group of SLI-children is not representative of SLI children in general. To mark this potential difference, we will refer to our group as Speech and Language Center SLI children (SLC-SLI group).
For the current study, a group of children between the age of 6 and 11 with specific language impairment was selected. None of the children had a diagnosis or a strong indication of a comorbid behavioural disorder. These children were deliberately excluded since it is very difficult in these children to differentiate between the effect of their disorder and their language impairment. Moreover, all children had Dutch as their only native language. Even though normal ranging nonverbal IQ scores are part of the definition of SLI, for the sake of homogeneity, we only included children with IQ scores above 85 (on scale of 100).
27 After applying these inclusion criteria, those SLC-SLI children were selected who were examined at all the relevant tests. In this way, a group of 143 children with SLI (age range: 72 – 143 months) was selected for inclusion in the current study. The group consisted of 38 6-year olds, 37 7-year olds, 37 8-year olds and 16 9-year olds, 11 10-year olds and 4 11-year olds. With 56 girls and 87 boys in our group, gender distributions were unequal. See table 1 for an overview of the group characteristics.
Table 1: Characteristics of the SLC-SLI group. Mean and standard deviation are reported in months.
Age N Mean age (SD) Boys Girls
6 38 76.89 (3.73) 23 15 7 37 87.7 (3.27) 21 16 8 37 100.20 (2.95) 22 15 9 16 112.94(3.92) 11 5 10 11 124 (3.62) 7 4 11 4 141 (3.36) 3 1 Total 143 87 56 2.2 Materials
The children who are treated at the Speech and Language Centre of Kentalis are administered a test battery including linguistic and neuropsychological tests. Moreover, audiological tests are administered and auditory processing is evaluated. Assessing cognitive functioning is essential since adequate cognition forms a precondition for language acquisition. A subset of the tests administered at the SLC is used in the current study.
Nonword Repetition: The ability to repeat nonwords was tested using a Dutch nonword repetition (NWR) task developed by Rispens (Rispens & Baker 2012). In this test, a nonword is presented auditory only once and this nonword has to be repeated immediately by the participant. The test consists of 40 stimuli ranging in length (2-5 syllables) and PP (high/low). Complexity is controlled for since all syllables have a CV-structure, except for the final syllable that has a CVC-structure. For more details on construction of the stimuli, see Rispens & Baker (2012) or Rispens et al (in press). Performance was scored with either right (1) or wrong (0). Percentage Phonemes Correct (PPC) was not calculated. Raw scores are reported in percentages of correctly repeated nonwords. No standardized scores are available since the test is (not yet) norm-referenced.
28 Continue Benoemen & Woorden Lezen (CB&WL): Serial rapid naming and reading abilities are tested using the CB&WL (translated as Continues Naming & Reading Words). The ‘Continues Naming’ part of the test includes four subtests in which colours, (monosyllabic) digits, pictures of objects and letters have to be named as quickly as possible. The test is inspired by the Rapid Automatized Naming (RAN) test of Denckla & Rudel (1974). Additionally, the CB&WL comprises two subtests in which words, both monosyllabic and multisyllabic have to be read as fast as possible. This rapid naming task can be used to screen for reading difficulties and automation deficits. Raw scores are reported in number of seconds that the participant takes to complete the task. A high raw score is thus associated with slow naming speed. The scores are standardized on a Wechsler scale ranging from 1 to 19 (mean: 10, sd: 3). The CB&WL is norm-referenced for children between the ages of 6 and 16 years (Bos et al 2007).
Peabody Picture Vocabulary Test (PPVT): Receptive vocabulary was tested using the third edition and Dutch version of the Peabody Picture Vocabulary Task (PPVT III-NL (2005) Schlichting). The PPVT measures the receptive vocabulary of a child by testing the understanding of the meaning of spoken words. Four illustrations are shown to the participant and the target word is read aloud. The participant has to choose which illustration fits the presented word best. Difficulty increases as the test proceeds. The test is interrupted when the participant incorrectly identifies six consecutive items. Raw scores are determined by the number of correct answers. Standard scores are reported in WBQ (mean: 100, sd: 15). This test is norm-referenced for an extremely large age range (2;2 to 90 years of age) and is used widely in clinical practice. (Dunn, Dunn & Schlichting 2005).
WISC-digit recall: For giving an indication of short term memory capacity by measuring digit span, a subtest of the Wechsler Intelligence Scale for Children (Dutch version Wechsler & Kort 2005) was used. In this test, sequences of numbers are read aloud by the experimenter and the participant has to repeat this sequence either forward (the original order) or backwards. As the test proceeds, the number of to-be-repeated digits increases, hereby rising STM demands. The number of digits that can be repeated accurately (digit span) constitutes the raw score. Scores are standardized on a Wechsler scale (mean: 10, sd: 3). The WISC is norm-referenced for children between 6 and 16 years of age.
Block recall WMTB-C: Visual memory capacity was measured using one of the eight subtests of the Working Memory Test Battery for Children. In the block recall test, children view nine wooden blocks located randomly on a board. The experimenter taps a sequence of blocks and
29 the child’s task is to exactly repeat this sequence. Sequence length increases as the test continues. Performance is reported in raw scores (number of correct answers) and standard scores (mean: 100, sd: 15) (Pickering & Gathercole 2001).
2.3 Procedure
The participants were administered on all the tests on two different moments: first at the start of enrollment as a baseline measurement and secondly 6 months after the treatment was finished as a posttest. In this study, only the data of the pretest is included in the analysis. All tests were administered within one morning at the Speech and Language Centre of Kentalis. The order of the two testing varied: some children started first with the linguistic tests and then with the neuropsychological tests. Other children were tested in the opposite order. All children were tested in quiet rooms while the whole session was videotaped. Parents were seated in the next room and could watch their child on a video screen. For this study, only the pretest data was included.
2.4 Data analysis
The program IBM SPSS Statistics® (version 20) was used to analyze the data. Participants were divided into three age groups. See Table 2 for the details of the different groups.
Table 2: Details of the three age groups
Age group Age N
1 6 and 7 year olds 75
2 8 and 9 year olds 53
3 10 and 11 year olds 15
The six subtests of the CB&WL were clustered into three composite measures of naming speed. The scores from the naming of digits and letters were combined and averaged into an alphanumeric SRN score. Moreover, the scores of the naming of colors and pictures were joined into a semantic SRN score. Lastly, the scores of the two word reading subtests were averaged into a literacy SRN score. This clustering of scores is often done since the subtasks rely on different underlying processes (Bowey et al 2005).
30 Since not all data was normally distributed and there was no homogeneity of variance for all tests, only non-parametric tests were used throughout the analysis. First of all, a descriptive statistics for all measures are reported for the different age groups.
Secondly, nonword repetition performance and serial rapid naming performance was analyzed in depth. NWR results were analyzed with the Wilcoxon signed-rank test to investigate whether there was an effect of age, length and PP for both the whole group and an effect of length and PP for the different age groups. Moreover, the performance of the SRN test was examined to test whether there was an effect of age.
Thirdly, correlations (Spearman’s rho) between the various tests over the whole group and within the different age groups were calculated. Furthermore, for the different age groups multi-regression analyses were performed. For the multi-regression analyses, the second and third age group were conjoined to enlarge group size.
31 III: RESULTS
3.1: Descriptive statistics
A summary of descriptive statistics for all tests for each age group is presented in Table 3.
Table 3: Descriptive statistics of all tests scores for each age group. Performance is reported in standard
scores, expect for the NWR score which is reported in percentage of correctly named nonwords.
Task1 Age Group 1
Mean (SD) Age Group 2 Mean (SD) Age Group 3 Mean (SD) NWR 0.28 (0.11) 0.38 (0.13) 0.45 (0.12) PPVT 95.58 (10.08) 94.91 (10.98) 94.36 (7.72) Digit recall 7.44 (2.44) 7.46 (2.97) 5.83 (2.55) Block recall 97.90 (17.77) 98.90 (11.98) 96.87 (15.47) SRN-sem 7.58 (3.06) 7.95 (3.50) 7.77 (3.79) SRN-alnu 7.95 (3.06) 8.12 (3.40) 7.88 (3.29) SRN-literacy 7.56 (2.65) 8.35 (2.94) 7.96 (3.47) N 75 53 15
1 the PPVT and Block recall are norm-referenced for TD-children with a mean standard score of 100 (SD: 15). The digit recall and the SRN tasks are standardized on a Wechsler-scale ranging from 1 to 19 (Mean: 10, SD:3).
From Table 3, it becomes evident that on average the SCL-SLI children perform well within the normal range on the PPVT and Block recall. This means that on average receptive vocabulary and visual memory are age-appropriate in the current population. Moreover, overall scores on Digit recall and the SRN tests are below average but less than 1 SD below the mean. Increase in all standard scores expect for the PPVT is observed in the second age group relative in the youngest children. The results of the third age group are remarkably: all standard scores are lower than the second age groups. Performance on Digit recall is especially impaired, almost 1.5 SD below the mean for TD-children, indicating deficits in phonological STM. This could potentially be an explanation for the persisting language deficits observed in these children. However, the third age group is small (N=15) so these results should be interpreted with care. Nevertheless, it seems plausible that this group comprises children suffering from severe and persisting language impairment. As predicted, performance on the NWR improves with increasing age. Whether this is the result of general development or that this improvement is accounted for by enhanced underlying skills is not clear. Overall, standard deviations in de second age group are larger than in the first group. The large variance in test scores reflects the heterogeneity of this group.
