Differences in the language processing of children
with (Developmental) Language Disorders
Rima Louiza Gebara S2740249
University of Groningen, Faculty of arts Thesis MA Neurolinguistics
Thesis supervisor: V.M. de Aguiar Date: 30-07-2020
Abstract Background
It is unknown to what extent the abilities of children with Developmental Language Disorders (DLD), children with Down Syndrome (DS), children with Hearing Impairment (HI), and children with (Developmental) Language Disorders ((D)LD; e.g. children with DLD and language disordered children associated with DS and HI) differ. Therefore, the aim of this study is to analyze differences in lexical processes between children with (D)LD, DLD, DS, HI in comparison to Typical Developing (TD) children to provide insights in the selective aspects of language processing while their overall language level is comparable.
Methods
The speech samples collected by Bol and Kuiken (1990) were used to empirically analyze the lexical processes between the groups of children with (D)LD, DLD, DS, HI, and TD. The groups of children with (D)LD, DLD, DS and HI were matched to TD-counterparts employing the mean length of utterances as a proxy for language level. To compare the groups all nouns and verbs were collected out of the speech samples. For every word, the psycholinguistic values were extracted out of different corpora. Inferences were made using non-parametric testing.
Results
The DS-group used words with significantly more phonemes than the TD-group and clinical groups. No significant differences were found between the DLD-group, DS-group, HI-group and the TD-group for the variable concreteness, frequency, Age of Acquisition, and phonological neighborhood. The (D)LD-group shows significantly more variation for the variable number of phonemes while they show a significantly less variation for the variable concreteness compared to the TD-group.
Discussion and Conclusion
The results indicate that there is a delay in the lexical semantics for the (D)LD-, DLD-, DS- and HI-group in comparison to the TD-group, but not an atypical pattern. Similarly, the results indicate that there is a delay in the lexical phonology and phonological buffer for the (D)LD- DLD- and HI-group. Children with DS might have greater difficulty with activating lexical phonologic representations than the TD-, HI- and DLD-group. However, it is also possible that they have a better developed phonological buffer. More research is needed to determine if children with DS have difficulties with activating their lexical phonology or with the phonological buffer.
Table of Contents
1. General Introduction ... 3
2. Theoretical background ... 4
2.1 Model of single word processing ... 4
2.2 Language impaired clinical groups... 4
2.2.1 Developmental Language Disorder (DLD)... 4
2.2.2. Language impairment across language domains in children with DLD ... 6
2.2.3 Hypotheses of lexical semantics ... 8
2.2.4 Language disorder associated with Down Syndrome (DS) ... 9
2.2.5 Language disorder associated with Hearing Impairment (HI) ... 11
2.3 Measurement of lexical abilities in children with language impairment ... 13
2.4 Nouns and verbs ... 15
2.5 Language processing and psycholinguistic variables ... 16
2.6 Objective and research question ... 19
2.7 Hypotheses ... 19
3. Method... 22
3.1 Participants within the database ... 22
3.2 Participants of this thesis ... 22
3.3 Data collection procedure ... 24
3.4 Data coding procedure ... 25
3.5 Statistical analyses ... 26
4. Results ... 27
4.1 Group means and group SDs... 27
4.2 (D)LD-group vs TD-group... 31
4.3 Comparing the DLD, DS and HI-groups with their matched TD-group ... 31
4.3.1 Qualitative comparison of the DLD, DS and HI-group ... 31
5. Discussion ... 37
5.1 Summary of findings... 37
5.2 Measures of lexical diversity ... 37
5.3 Psycholinguistic variables ... 38
5.3.1 (D)LD-group vs the matched TD-group ... 38
5.3.2 DLD-, DS- and HI-groups vs matched TD-group ... 39
5.4 Limitations, strengths, and future research ... 41
6. Conclusion ... 44
General Introduction
Language is one of the most complex systems of rules a person learns. Similar as every other facet of human development, language development is characterized by variation (Bates, Dale, & Thal, 1995). For most children language development is a fast, and seemingly effortless task. However, some children have a delayed language development, which can increase the risk of a negative effect on education, employment, and social and emotional problems (Laasonen et al., 2018). Many different causes are known for these language learning difficulties (Bishop & Leonard, 2014). In this thesis the spontaneous language of three language disordered groups, namely children with developmental language disorders, Down Syndrome and hearing impairment is investigated. It is suggested that there are differences and similarities in the lexical semantic and lexical phonology of the three language impaired groups (Rice, Warren, & Betz, 2005).
To evaluate and understand the linguistic abilities of children with language and communication disorders in comparison to typically developing children the descriptive method ‘speech sample analysis’ is frequently used in combination with the measures for expressive vocabulary size (Marques & Limongi, 2011; Templin, 1957). However, these measures are often limited, because they may mask important linguistic differences between children and it is not always clear what part of the linguistic knowledge of a child the measure actually reflects (Dethorne, Johnson, & Loeb, 2005; Malvern & Richards, 2002; Watkins, Kelly, Harbers, & Hollis, 1995).
To explore the differences and similarities in the linguistic abilities of language impaired children and typical developing children psycholinguistic variables can be used. However, few reports have examined children’s linguistic processing with psycholinguistic variables. Therefore, the objective of the present study is to examine whether the nature of the lexical semantic and lexical phonology, measured with psycholinguistic variables, differs between different clinical groups. Studying this will provide insight in the language representations and processing of different language impaired groups.
Next, the theoretical background provides information about the three clinical groups: children with developmental language disorders, children with down syndrome and children with hearing impairment. Then, the language measures for expressive vocabulary size are presented and differences between the acquisition of nouns and verbs will be discussed. Fourth, the psycholinguistic variables are explained, and the expectations of the performance of the clinical groups on the different variables as a result of the literature are discussed. Thereafter, the objective of the study, the research question and the hypotheses are presented.
2. Theoretical background 2.1 Model of single word processing
Language models provide frameworks to constructively interpret data (Gray & Kiran, 2013). An often-used model which is also employed in this thesis is the model of Ellis and Young (1988). This model is successful at explaining language deficits and incorporates auditory and visual input and output processes. Moreover, it illustrates links between auditory analysis, phonological lexicons, the semantic system, and the phoneme level of spoken, written or visual words (Bryan & North, 1994; Gray & Kiran, 2013). According to this model, to produce a spoken word and for example name a picture the item must first be presented to the semantic system (e.g. lexical-semantic) via auditory or written modalities or via the visual object recognition system. A search is then made to match the pattern of features to a store of known words. Once identified, the meaning of the word is accessed in the semantic system (e.g. lexical semantic). These representations only include its meaning, but not the word form. The semantic representations stored in the lexical semantics are called lemmas which are lexical items that include information about the syntactic and semantic features of the word, but not about its phonological structure (Biran & Friedmann, 2005). The lexical semantics do activate the phonological output lexicon (e.g. lexical phonology) in which the phonological structure of a word is stored. The phonological output lexicon consists of the phonological form of the word. Moreover, the word is retrieved from it as a sequence of sounds. As the right word form is
retrieved, the phonological (output) buffer acts as brief temporary storage that holds the letters as the words are formed. After the letters are properly sequenced, the word is then spoken (Biran & Friedmann, 2005; Bryan & North, 1994; Gray & Kiran, 2013).
2.2 Language impaired clinical groups
2.2.1 Developmental Language Disorder (DLD)
Language does not always develop as expected, leading to negative effects on social, academic and economic standing (Laasonen et al., 2018; Leonard, 2014). Children with a Developmental Language Disorder (DLD) have major problems in learning to talk and perform below age expectations on language measures, despite showing normal development in other developmental areas. Generally, children with DLD show a delayed or deviant pattern of language development (Leonard, 1998).
A typical 7- or 8-year-old child with DLD might talk like a 3-year-old child, using ungrammatical utterances (e.g., “me go there,” instead of “I went there”; Bishop, 2006). DLD is a heterogeneous category that varies in profile as well as severity of the disorder. However, in most cases children have problems with both understanding and producing spoken language. For example, the child could have difficulty with acting out a sentence using toys, such as “the boy is chased by the dog”. The child may
show confusion as to who is doing what to whom. DLD is difficult to diagnose because it often occurs in children who are otherwise developing normally, with no hearing problems or physical handicaps that explain the difficulties (Bishop, 2006). The most prominent problems among children with DLD are in the areas of morpho-syntax, phonology, and lexicon (Bishop, 1997). Approximately 7% of the population has DLD. Boys are affected three times more than girls (Tomblin et al., 1997).
Children with language problems can be diagnosed with DLD from the age of 5. At that age it becomes reasonable to classify children because the language problems are unlikely to resolve without help (Bishop, 2017). For many years there has been an ongoing debate about how to diagnose and how to define language impairments. Previous studies and some current studies still use the term Specific Language Impairment (SLI). However, this term is rejected recently (Bishop, 2017), because the term implies that only language is affected. Differently, DLD may co-occur with conditions, such as cognitive disorders, behavioral disorders and motor disorders. In this thesis the term DLD will be used.
The consensus paper of Bishop (2017) also stated adjusted criteria for diagnosing DLD. The first point for diagnosing is that test scores from standardized tests provide useful information, but it should not be used as the single criterion to identify DLD. In addition to the tests there should be evidence of a functional impairment (e.g. education or social interaction). However, these functional impairments depend on subjective judgements, which causes scope for biased and uneven outcomes (Bishop, 2017).
Formerly exclusion criteria were used for diagnosing DLD. However, this way of diagnosing has been noticeably criticized (Reilley et al., 2014). Nowadays the focus is on differentiating conditions, which indicates that when there is a condition ‘X’, the term ‘language disorder X’ would be managed as a substitute for DLD. The differentiating conditions are biomedical conditions (e.g. a genetic syndrome), sensorineural hearing loss, neurological disease, conditions of Autism Spectrum Disorder (ASD) or intellectual disability. It is noticeable that the term differentiating conditions does not include milder neurodevelopmental disorders such as ADHD. This is because, in contrast to the term SLI, the term DLD indicates that DLD can co-occur with neurodevelopmental disorders (Bishop, 2017). Moreover, the diagnosis of SLI required a non-verbal IQ within normal limits. However, IQ does not determine the responsiveness to therapy. Therefore, non-verbal-IQ should not be a diagnostic criterion (Bishop, 2017).
Although DLD is diagnosed most often during childhood, the associated difficulties do persist later on. To cope with the risks caused by DLD, it is essential to understand the interactions between protective and harmful factors that affect the developmental manifestation of DLD better (Laasonen et al., 2018). However, there is not yet consensus in the literature about the origin of the impairment
(Bishop, Whitehouse, Watt, & Line, 2008). There is growing evidence that genetic factors play an important role (Li & Bartlett, 2012). At the neural level, the perisylvian brain areas that contribute to language processing are often affected within people with DLD (Badcock, Bishop, Hardiman, Barry, & Watkins, 2012). However, the precise mechanisms that cause the neural abnormalities in DLD are not known. Presently, the range of cognitive or behavioral difficulties associated with DLD is not even entirely understood. For example, cognitive difficulties have been suggested to include nonverbal as well as verbal domains, and the linguistic markers of DLD seem to vary from one language to another language (Leonard, 2014). The language of children with (Developmental) Language Disorders (D)LD can be heterogeneous, with different degrees of impairment in different language functions and domains.
2.2.2. Language impairment across language domains in children with DLD
The domain of morphology, which is the study of morphemes, languages smallest units of meaning and the domain syntax, which is the part of grammar that prescribes how words can be combined into phrases and sentences, are two well-known areas of weakness for children with DLD (Leman et al., 2012; Rice et al., 2005). Children with DLD differ in their morphosyntax from both typically developing age-matched children and typically developing younger language-matched children. The most common error pattern for children with DLD is to omit these morphemes more frequently and for a longer period than typically developing children (Rice, 2003; Rice, 2004; Rice & Wexler, 1996).
Late onset of word acquisition is frequently the first sign of DLD (McGregor, Newman, Reilly, & Capone, 2002). Several studies have examined initial mapping of new words. The studies found that in the experimental word-learning situations, children with DLD are weaker at establishing initial maps of new words compared to their age-matched peers. Rice, Buhr, and Nemeth (1990) compared children with DLD and typical developing (TD) children matched on Mean Length of Utterance (MLU) and found that despite the equal MLU, children with DLD scored lower on the fast mapping of words, which is an important skill for new word learning, than the MLU-matched TD children. This difficulty is thought to be in the origin of the reduced vocabulary levels displayed by children with DLD (Rice, Buhr, & Oetting, 1992; Rice, Oetting, Marquis, Bode, & Pae, 1994). Given the difficulty with novel word learning, most children with DLD perform below their age-matched peers on expressive and receptive vocabulary measures but perform similarly to younger children (Rice, 2003; Rice, 2004; Rice et al., 2005).
Concerning the lexical semantics, it is suggested that children with DLD have deficits in the lexical semantic domain and are more likely to accomplish below age peers (Leonard, 1998; Sabisch, Hahne, Glass, von Suchodoletz, & Friederici, 2006). According to McGregor et al. (2002) children with
DLD are less likely to learn lexical labels and semantic features of new words than age-matched peers. Moreover, it is suggested that naming errors are the result of less elaborate semantic
representations. In addition, Nash and Donaldson (2005) found that children with DLD are poorer at learning the meanings of low-frequency words (e.g. gauntlet, polka), because of the weakness in word learning of children with DLD. They also describe that these children have poorer lexical semantic representations for words stored in their long-term memory. Therefore, retrieving these words could be more difficult (Sheng & McGregor, 2010). However, Sheng and McGregor (2010) found that there is big variability in lexical semantic skills among children with DLD. The children with DLD with deficits in the lexical semantics were most likely to have word-finding difficulties too. Moreover, the lexical semantics of a child with DLD can be adequate to the vocabulary of the child but not to the child’s age (Pizzioli & Schelstraete, 2011).
In contrast, Drljan and Vuković (2019) found that children with DLD provided fewer semantic responses and more errors compared to both typically developing vocabulary- and age-matched children. These data showed that children with DLD have difficulties with lexical semantics and it exceeds their overall vocabulary delays. Moreover, the results of some studies show that difficulties in the lexical semantic processing continues through the school-age period within children with DLD (Mainela-Arnold, Evans, & Coady, 2010).
Children with DLD make more naming errors during object naming, action naming and story retelling than their typically developing peers. This demonstrates that children with DLD have difficulties with the retrieval of words from their long-term memory (McGregor, 1997). Even when children with DLD name correctly, they need more time to retrieve the names than their typically developing peers do. These difficulties may emerge due to impaired lexical phonology and lexical semantics (Lahey & Edwards, 1996; Leonard, Nippold, Kail, & Hale, 1983).
Dollaghan (1987) showed that children with DLD were able to map the referent and context of a new word after minimal exposure but were not able to map enough of its phonological form to support the production, unlike typically developing peers. Bishop (1997) demonstrated that children with DLD could recognize new words but could not produce new words. This suggests that the lexical phonology of children with DLD is underspecified within their lexicon. In addition, school-age children with DLD need to hear significantly more of the spoken word than their typically developing peers to recognize newly learned words, but not to recognize familiar words (Dollaghan, 1998). Moreover, there is good evidence that DLD is associated with difficulties in processing speech (Joanisse & Seidenberg, 1998). Bishop and Snowling (2004) suggest that impaired lexical phonology skills are a key deficit in DLD. Moreover, some researchers suggested that these difficulties affect the development
of lexical phonology and that degraded lexical phonology is the cause of deviant acquisition of morphosyntax (Benton, 1964; Joanisse & Seidenberg, 1998).
In addition, previous research demonstrated that children with DLD have impairments in repeating lengthy nonwords and general nonword repetition which indicates problems with the phonological buffer (Conti-Ramsden, 2003; Dollaghan & Campbell, 1998; Roy & Chiat, 2004).
2.2.3 Hypotheses of lexical semantics
There are two different hypotheses about the lexical semantic representations of children with DLD. The storage hypothesis suggests that children with D(LD) are slower at vocabulary learning. Therefore, they are less familiar with the words in their lexicon in comparison to TD children. This results in less elaborate representations and fewer lexical connections (Brackenbury & Pye, 200; Kail, Hale, Leonard, & Nippold, 1984). Support for the storage hypothesis is provided by Kail et al. (1984). They examined children who needed to remember as many words as possible in repeated free recall (e.g. children recalled the list three times in a row), and in cued recall (the investigator provided the category names as retrieval cues). The results showed that the children with language impairment recalled fewer words than their age matched peers in both cued and free recall. Moreover, the pattern of repeated free recall implies that children with a language impairment were less likely to store the word that was presented. In addition, they were less consistent in their retrieval of words than their age matched peers.
The retrieval hypothesis assumes that children with D(LD) have difficulties with the process of accessing lexical-semantic information because of inefficient word retrieval mechanisms. The difficulties are not located within the lexical entries or their connections (Fried-Oken, 1987; Newman & German, 2002). Support for this hypothesis is provided by Newman and German (2002) who examined the naming accuracy of 320 TD children and children with word-finding difficulties under a variety of lexical factors. These lexical factors are word frequency, age-of-acquisition (AoA), neighborhood density, neighborhood frequency, and stress pattern. The results show that all factors influence lexical access in children. Moreover, significant results show that AoA effects decreas with the maturation of TD children. In contrast, the effects continue to influence the lexical access of children with word finding difficulties. This could indicate that impairments in accessing words might have prevented them from developing a strong access path to the words. These findings are suggested to support a view of lexical access in which access paths to words are strengthened with successful use. Both hypotheses are not mutually exclusive, namely a child with D(LD) could have poorly elaborated representations and difficulties with accessing them (Newman & German, 2002).
2.2.4 Language disorder associated with Down Syndrome (DS)
Down syndrome is one of the most deeply studied genetic syndromes. This is due to the fact that Down syndrome is the most frequent known genetic cause for intellectual disability with a prevalence of one child in 700 births (Abbeduto, Warren, & Conners, 2007; Centers for Disease Control and Prevention, 2006). It is caused by a trisomy of chromosome 21. This genetic deviation results in a distinct facial appearance (e.g. flat nasal bridge, small mouth, small ears), heart and respiratory problems, and cognitive impairment. Down syndrome can be diagnosed with a non-invasive prenatal screening (Lou, Lanther, Hagenstjerne, Petersen, & Vogel, 2020).
This cognitive impairment varies from a mild impairment with an IQ between 50 and 70, to a moderate impairment with an IQ between 35-50, or to a severe impairment with an IQ of 20-35 (Bull, 2011; Chapman & Hesketh, 2000). However, the severity of the impairment differs per cognitive domain. Children with Down syndrome often perform more effectively in social situations than would be predicted on the basis of the cognitive impairments. Overall, young children with DS can be affectioned, empathic, and engaging (Moore, Oates, Obson, & Goodwin, 2002; Wishart & Pitcairn, 2000). These social abilities could improve with early intervention techniques. However, the level of social functioning is highly variable (Bull, 2011).
Language, however, is one of the greatest impaired domains in Down syndrome and it even may be the biggest obstacle to live independently and to have a meaningful inclusion in society (Chapman & Hesketh, 2000; Chapman, 2003). Children with DS are characterized by an absence of developmental homogeneity between their cognitive and linguistic competence. Namely, the linguistic competence of children with DS is weaker than what could be expected based on their overall cognitive level (Chapman & Hesketh, 2000).
However, children with DS are also at risk of deficits in language development because of reasons beyond the corresponding cognitive skills. For example, deficits in the motor coordination that are associated with DS could affect the motor movements that are required for the speech production system (e.g. phonation, articulation, and respiration). Moreover, children with DS have an increased prevalence of middle ear infection which could cause a hearing loss (Miller, 1992). In addition, their spontaneous speech is often incomprehensible (Rice et al., 2005).
The difficulties within the morphosyntactic domain emerge in an early phase of linguistic development. Namely children with DS produce simpler and more telegraphic sentences, and more frequently omit function words than their typically developing age-matched children (Vicari, Caselli, & Tonucci, 2000). When the age increases, the split between morphosyntactic and lexical abilities decreases, and a generalized picture of linguistic difficulty emerges (Fowler 1990; Miller 1992;
Fabbretti et al., 1997). The comprehension domain remains better than the production domain (Rondal, 1995; Fabbretti et al., 1997).
Previous studies have investigated the components of linguistic abilities of children with DS and have found that phonology and morphosyntax are areas of difficulty, but all language areas are affected, especially in expressive language. There are studies that report that the lexical
development of children with DS is moderately preserved compared to their other linguistic abilities (Fabbretti, Pizzuto, Vicari, &Volterra, 1997; Fowler, 1990). However, a number of inconsistencies can be found in the literature on this topic both in early stages and in older children (Galeote, Sebastián, Checa, Rey, & Soto, 2011). Namely, the study of Caselli et al. (1998) shows that productive
vocabularies develop at practically the same mental age in children with DS and in TD-children. However, Roberts, Price, Barnes et al. (2007) found that children with DS had lower productive vocabulary than TD-children.
The lexical semantics is also a domain of deficits in children with DS, despite the individual variability. Namely, the onset of learning the first word is often delayed. Moreover, early expressive vocabulary growth is often behind (Berglund, Eriksson & Johansson 2001; Mervis & Robinson, 2000). In addition, Andreou and Katsarou (2016) found that the lexical semantics is an area of difficulties in children with DS in both expressive and receptive language.
The expressive vocabulary levels of preschoolers, children in elementary school, and adolescents with DS are delayed compared to their nonverbal cognitive levels (Chapman, Schwartz, & Bird, 1991; Miller 1988). In addition, more recent findings show that children with DS have difficulties in producing many words in various categories and especially in verbs. This is measured using the fast mapping technique. This technique is described as a cognitive strategy and it allows children to produce as many words as possible from a grammatical category (Andreou & Katsarou, 2016; Nash & Snowling 2008).
As previously stated, lexical phonology is an impaired domain in children with DS. In typical development, lexical phonology is thought to be a result of extending lexical knowledge (Metsala, 1999). Therefore, it is suggested that general language delay in Down syndrome leads to difficulties in lexical phonology. The studies of Cossu, Rossini, and Marshall (1993) and Evans (1994) showed that it is challenging to quantify lexical phonology in children with DS using tasks that are designed for typically developing children. These tasks gave more beneficial results when used to investigate the phonological processing skills of young adults with Down syndrome (Fowler, Doherty, & Boynton, 1995). Therefore, assessment variables in lexical phonology tasks should be considered carefully within this population. In this way, the tasks specifically measure lexical phonology skills, without being needlessly complicated by other general cognitive skills (Kennedy & Flynn, 2003).
A deficit in phonological buffer is a well-documented feature of DS as well (Fowler, Doherty, & Boynton, 1995; Jarrold & Baddeley, 1997; Jarrold, Baddeley, & Hewes, 2000). Namely tasks such as nonword repetition and digit span are difficult, including the production of intelligible speech sounds and problems in oral language and phonological short-term memory (Abbeduto et al., 2007; Miolo, Chapman, & Sindberg, 2005; Seung & Chapman, 2000). For example, the digit spans of children with DS are generally lower than the digit spans of other children with the same receptive vocabulary level (Jarrold & Baddeley, 1997; Seung & Chapman, 2000). This deficit has been a selective impairment of the phonological buffer and is often attributed to low phonological storage capacity (Jarrold et al., 2000).
2.2.5 Language disorder associated with Hearing Impairment (HI)
The hearing acuity of children is graded as normal hearing when it is within 20 dB. The severity of a hearing loss is classified as mild if it is within 20–40 dB, moderate if it is within 41–55 dB, severe if it is between 70–90 dB, and profound if it is 90 dB or more. Although there is no acknowledged demarcation, the individuals with a severe or profound hearing loss are generally referred to as deaf. Individuals with a mild or moderate hearing loss are generally referred to as hard of hearing. One of the most common hearing deficits in children of more developed societies is sensorineural hearing loss. For example, in the USA, sensorineural hearing loss occurs around three times more often than Down syndrome. Sensorineural hearing loss is a versatile condition that causes medical, social, and culture difficulties. Early diagnosis of hearing loss is necessary to achieve a satisfactory linguistic and cognitive development (Smith, Bale, & White, 2005). To diagnose children with hearing losses nowadays neonatal screenings for hearing losses and audiologic evaluation are used (Farinetti, Raji, Wu, Wanna, & Vincent, 2018). Audiologists, speech therapists, doctors and other professionals mainly use the term hearing impaired (HI) to describe the individuals with any degree of sensorineural hearing loss, and this will be done in this thesis too (Smith et al., 2005).
The common consequences of HI for children includes a significant delay in language acquisition and therefore in their academic achievements. HI affects all aspects of oral language acquisition, because children with HI are not able to access and extract information from the oral language around them (Eisenberg, 2007; Moeller, Tomblin, Yoshinaga-Itano, Connor, & Jerger, 2007; Pimperton & Kennedy, 2012). The language delay can appear in all levels of severity of HI, thus from a mild to a profound HI (Smith et al., 2005). A child that has HI since birth is particularly vulnerable to have a disordered and delayed language development, because the auditory deprivation is during the sensitive period for language acquisition (Kuhl, 2004). Namely, if a child does not or cannot receive appropriate language input during this early sensitive period it has an impact on the neural pathways
in the brain that support language acquisition. If the same absence of language input appears later in life it does not have an impact on the neural pathways (Kral, Hartmann, Tillein, Heid, & Klinke, 2001; Shepherd & Hardie, 2001).
Many professionals in both special education and health care have supported early identification of HI and consequent intervention to improve the language development and academic outcomes of deaf and HI individuals (Smith et al., 2005). As mentioned above a hearing loss always affects the speech and language development (Miller, 1987). Most HI children have difficulties in different linguistic domains, namely vocabulary, phonology, and morphosyntax performance are below their age level. However, considerable linguistic heterogeneity is reported (Delage & Tuller, 2007). HI thus can limit the development of concepts and words, language comprehension and language communication.
Regarding the lexical semantics of children with HI, the study of Jerger et al. (2006) found that the children with HI develop normal lexical semantic representations, but the dynamics of their semantic processing seem to be changed by the presence of HI in childhood. They seem to have an abnormally prolonged semantic stage of processing, which could reflect limitations in rapid object naming and slowed picture naming (Kail, 1996). Moreover, children with HI have a more limited and slowed vocabulary development than their peers. Because of the small vocabularies it could be expected that children with HI have an underdeveloped semantic structure, but there is a large heterogeneity among HI children (Luckner & Cooke, 2010; Johnson, 2009; Pittman, & Schuett, 2013; Walker & McGregor, 2013).
The influence of having a HI since birth is assumed to play an important role in building knowledge about lexical phonology (Levelt, Roelofs, & Meyer, 1999; Tye-Murray, 1992). The HI can degrade and filter the auditory input which results in less specified auditory phonological knowledge. Therefore, visual speech could become a crucial source for lexical phonology knowledge. Moreover, the lexical phonology may be more closely tied to articulatory codes than auditory codes. In addition, it is stated that children with moderate HI have impoverished lexical phonology, with abnormally poor phoneme discrimination and phonological awareness (Briscoe, Bishop, & Norbury, 2001). However, cognitive cues could play an important role in the linguistic processing to compensate for their poor phonological skills (Kallioinen et al., 2016).
Regarding the phonological buffer it is administered that children with HI have difficulties with the repetition of nonwords. In comparison with the children with DLD the impairments were of similar magnitude. However, the children with DLD had poorer repetitions and therefore, could be
differentiated from children with HI on phonologically complex nonwords (Briscoe, Bishop, & Norbury, 2001).
2.3 Measurement of lexical abilities in children with language impairment
Language assessments usually involve a combination of standardized tests, parental questionnaires, and elicitation of spontaneous speech. From spontaneous speech, the most used measures are the Type Token Ratio (TTR; the ratio of different words to total words used), the D, which represents the vocabulary diversity of a speech sample, and the Mean Length of Utterances (MLU; Klee, Stokes, Wong, Fletcher, & Gavin, 2004; Marques & Limongi, 2011; Templin, 1957).
The MLU is a global used measure of productive language ability and is frequently used within clinical and research settings. MLU was originally popularized by Brown (1973). It has been recommended as a useful measure for both diagnosing language impairments and for monitoring treatment progress (Miller, 1981; Paul, 2000). MLU is a measure that is often used for matching language disorder children with younger typical developing children on the same language level. Another common way for comparing language disorder groups is to match on chronical age. As mentioned above MLU is a measure often used for both diagnosing language impairments and for monitoring treatment progress. By matching children on MLU, it is possible to identify if there are selective aspects of language processing that differ between groups, even if their overall productive abilities are comparable (Rice et al., 1992). However, when the groups are matched on MLU it is suggested that the chronical age of a child does not influence the results. Nevertheless, when a difference is found between children with language problems and younger typical developing children with an approximately equal language level the difference could be due to an age difference or due to language problems (de Jong, 1994). It is hard to distinguish between both possible causes (Bishop, 1992). Early research from Morehead and Ingram (1974) compared children with DLD to TD children with equal MLUs and found that they did not differ in the syntactic language development.
Despite its frequent use it remains relatively unclear what MLU reflects, in terms of a child’s linguistic knowledge. Namely, it is relatively unexplored to what extent MLU is influenced by other language domains, such as semantics or phonology (Dethorne et al., 2005). Moreover, Eisenberg, Fersko, and Lundgren (2001) have reported that the MLU can identify some, although not all, preschool children with language impairment. They suggest that a low MLU can be interpreted as supporting a diagnosis of language impairment, however, a high MLU cannot determine if a child has no language impairment (Eisenberg, et al., 2001).
The measure number of different words (NDW) in a speech sample is a very straightforward measure of lexical diversity (Miller, 1991). However, this measure is hard to compare across children
because it is influenced by sample size. Namely, the NDW will increase as the size of the sample increases until the speaker reaches his or her total active vocabulary. However, the rate at which the NDW enlarges will slow down, but the absolute value of the reported NDW continues to grow. This makes it hard to compare children in a developmental context. To compare children with the measure NDW all sample sizes should match the smallest sample size, measured in lexical tokens. Thus, cutting samples based on the number of utterances is not sufficient (Owen & Leonard, 2002). If the NDW is calculated for a sample with a standardized number of utterances or by time, this influences the number of lexical tokens and therefore the MLU (Klee, 1992). Namely, children with a lower MLU will produce fewer number of words in, for example, a 200-utterance sample. Therefore, they will have a lower total number of words, than children with a high MLU and vice versa. Additionally, the total number of words in a speech sample (TNW) is related to talkativeness when the speech sample is limited by time (Owen & Leonard, 2002).
TTR is a ratio of the NDW used in a (semi-) spontaneous speech sample to the NTW in the sample (Templin, 1957). This measure had been used in the past years to estimate children’s lexical proficiency in as well the clinical as the research setting. The TTR has been recommended as a measure to quantify children’s vocabulary diversity and identifying a language disorder and/or describing the strengths and weaknesses of a child with language learning difficulties (Miller, 1981; Retherford, 1993). A large TTR means that words are less frequently repeated in the speech sample (Owen & Leonard, 2002). TTR has also been used in research of DLD (Burroughs, 1991). Burroughs (1991) reported that children with low lexical diversity were judged as less talkative and less mature than children with high lexical diversity. However, the general limitation of TTR as an index of language development or disorder has occurred. The TTR is highly influenced by sample size (Malvern & Richards, 2002; Templin 1957). The more utterances in a sample, the more words tend to repeat which causes a decrease of the TTR (Hess, Haug, & Landry, 1989; Richards, 1987; Watkins et al., 1995). In addition, the TTR is a measure of vocabulary size, and thus cannot say much about at which level of processing the impairment evokes and how severe the impairment is at individual levels of language processing (Owen & Leonard, 2002).
To solve the problem with the TTR that is highly influenced by sample size, Malvern and Richards (2002) developed the measure D for lexical diversity. The calculation of the D is based on the chance of introducing new vocabulary into increasingly longer speech samples. However, Owen and Leonard (2002) found that even though D appears to be less influenced by sample size than older measures such as TTR, it may not be totally independent of sample size. Moreover, when measuring the D and NDW score from spontaneous speech samples they suggest that children with DLD only
have subtle lexical deficits, because the children have control over the topic of conversation and which lexical items they use.
2.4 Nouns and verbs
Another way to investigate these developmental problems is to use the measures of noun and verb production (Longobardi et al., 2015). The categories noun and verb are basic parts of speech and exist universally across languages (Lyons & John, 1968; Robins, 1952; Robins, 1979). Notably, the two categories do not follow the same course in language development (Longobardi et al., 2015). Namely, the ‘noun bias’ is a universal disposition to acquire nouns before verbs in children. This causes the acquisition of verbs to stay behind that of nouns (Gentner & Kuczaj, 1982). In addition, nouns and verbs carry different types of information. For example, temporal information (e.g. tense) is often carried on verbs. However, plurality is often marked on nouns. Moreover, the order of nouns and verbs in a sentence is syntactically fixed. It is suggested that the slower development of verbs in comparison to nouns, is caused by the greater conceptual and grammatical complexity of verbs compared to nouns (Gentner, 1982; McDonough, Song, Hirsh‐Pasek, Golinkoff, & Lannon, 2011). Therefore, it is suggested that only children at a certain age with sufficient cognitive ability can comprehend the concepts that are represented by verbs (Benedict, 1979; Gentner, 1982). However, it is suggested that verbs cause less semantic activation and less semantic feedback in lexical decision than nouns which could make nouns more complex to learn (Cordier, Croizet, & Rigalleau, 2013).
Moreover, Hadley, Rispoli, and Hsu (2016) found that spontaneous production of different verbs and parent-reported verb lexicon size were better predictors than nouns for developmental language problems. The best lexical indicator of grammatical complexity after six months appeared to be children’s spontaneous speech production of lexical verbs at 24 months. Moreover, a larger diversity of verbs will increase variation in the sentences produced (Hadley et al., 2016).
Verbs and nouns also differ in the level of concreteness. Nouns tend to be rated as more concrete than verbs. It is also stated that concrete words are processed faster than abstract words (Kounios & Holcomb, 1994; West & Holcomb, 2000). Therefore, it could be that nouns are easier to process (Colombo & Burani, 2002). Moreover, nouns tend to be acquired earlier than verbs, which could mean that the age of acquisition (AoA) of a word is a significant predictor of the differences in noun verb acquisition. However, Colombo and Burani (2002) found the AoA cannot uniquely account for the processing difference between nouns and verbs. Marinellie and Chan (2006) suggested that for both nouns and verbs, the gap between the knowledge of high- and low-frequency words progressively decreases with age (Marinellie & Chan, 2006). However, since word frequency and AoA are highly correlated, the role of word frequency is not totally clear (Bastiaanse, Wieling, & Wolthuis,
2016). In contrast, Gillette et al. (1999) suggested that the learnability of a word is not completely based on lexical class (e.g. verb or noun), therefore different variables may distinguish TD-children from children with language problems when extracted from verbs and nouns.
2.5 Language processing and psycholinguistic variables
An approach that tries to explain the way in which children process speech and language at a cognitive or psychological level is the psycholinguistic approach. This approach aims to formulate hypotheses about the psychological processes that may be impaired within the speech and language development (Baker, Croot, McLeod, & Paul, 2001). This is mostly done by developing theoretical models. Models can capture the important components of a system and make the relationships among those components clear. Most of the models are based on healthy people or on behavioral dissociations that are documented in populations with neurological impairments. The dictation of these models can be displayed in box-and-arrow diagrams (Rofes, 2015). It is stated that it is possible to use the box-and-arrow models to investigate the speech processing skills of children (Menn, 1978; Smith 1973). Each box represents the level of representation or processing, and the relationship between the boxes are called the arrows (Levett, 2006).
The model that is referred to in the book Whitworth, Webster and Howard (2014) and in this thesis is an adjusted version of the model of Patterson and Shewell’s (1987). This model is an adaptation of the earlier Logogen models. Early Logogen models are models that account for both the types of errors and the factors influencing reading performance (e.g. imageability of a word) in individuals with deep dyslexia. These Logogen models were a typical ‘box and arrow’ processing diagram. The language-processing model for single words of Whitworth et al. (2014) based on the model of Patterson and Shewell’s (1987) provides a functional working model of language processing. Namely, it administers a description level that can be used to lead an assessment process that can identify undamaged and disrupted processes. Every component of the model is necessary to account for the processing of single words. If a box or arrow were removed it would consequently result in a system that is unsuccessful in at least one language-processing task (Whitworth et al., 2014).
One of the hypotheses that focus on testing the intactness of specific components of the language model is the critical variable approach. This approach implies assessing different levels of intactness and disruption of processing. Moreover, it describes the effect of different variables (e.g. frequency, word length) on performance (Shallice, 1988). Therefore, it investigates variables that could affect the chance that a task will be performed correctly by a client (Nickels & Howard, 1995a, p. 1281; Shallice, 1988). There are several factors that could be manipulated to provide information during the assessment. The variables included in this thesis give rise to error patterns from which assumptions can be drawn.
The following psycholinguistic variables will be included to examine differences in the linguistic representations of the three language disordered groups and typical developing children within their spontaneous speech. Those variables are: frequency, which implies how often a word appears in spoken language (Freq; Goh, Suárez, Yap, & Tan, 2009; Stokes, 2010); concreteness, which is the degree of abstraction associated with the concept a word describes (Conc; Gerhand & Barry, 2000; Howell & Becker, 2001; Richardson, 1975; ); Age of Acquisition, which is the average age at which a word is learned (AoA; Barry, Morrison, & Ellis, 1997; Gerhand & Barry, 1998; Xue, Liu, Marmolejo-Ramos, & Pei, 2017;); length in phonemes, which indicates the number of phonemes a word contains (NPh; Nickels & Howard, 2004; Johnston, Johnson, & Gray, 1987); and, phonological neighborhood, which implies the number of words existing that are phonologically similar to a word (phNE; Goh et al., 2009; Jones & Brandt, 2019).
Lexical semantics
The variable concreteness plays an important role in the acquisition of words and is processed in the lexical semantic system (Gerhand & Barry, 2000). More concrete verbs are learned before less concrete nouns, which results in the few early verbs in children’s vocabulary (Gentner & Kuczaj, 1982). Moreover, the presentation of a concrete word activates a wider lexical semantic network than the presentation of an abstract word (Grondin et al., 2019). In addition, concrete words (e.g. chair, bed) are generally processed more quickly and completely than abstract words (e.g. love, idea; Barber, Otten, Kousta, & Vigliocco, 2013; Whitworth et al., 2014).
Lexical phonology
Frequency is a variable that is suggested to play an important role in word learning in children (Gerhand & Barry, 2000; Stokes, 2010). It is suggested that more frequent words are accessed and produced more easily than less frequent words. If it is hard for someone to produce less frequent words, it could indicate an impairment in the lexical phonology. The variable AoA is said to correlate with frequency and the AoA effect reflects the demands of word processing in the phonological lexicon as well (Wijnendaele & De Deyne, 2000). Gierut and Morrisette (2013) suggested that phonology is the relevant domain responsible for AoA effects in children. In adults Brysbaert, Wijnendaele and De Deyne (2000) found that the AoA effect reflects the demands of word processing in both the lexical phonology but also in the lexical semantic system. AoA is associated with the speed with which the representation of words can be activated. Namely, words that are acquired first are easier to access than words that are learned later (Izura et al., 2011; Lachman, Shaffer, & Hennrikus, 1974; Monaghan & Ellis, 2010; Stadthagen-Gonzalez, Bowers, & Damian, 2004; Whitworth et al., 2014). The effect of
the AoA can be found in various experimental tasks, namely in picture naming (Meschyan & Hernandez, 2002), word naming, and lexical decision (Morrison & Ellis, 1995).
Moreover, effects of the variable phonological Neighborhood reflect the demands of word processing at the level of lexical phonology (Dell, Chang, & Griffin, 1999; Schwartz, Dell, Martin, Gahl, & Sobel, 2006). Words that come from phonologically dense neighborhoods have many phonological neighbors (e.g. words that sound the same as many other words in the target language) and are thought to be learned earlier in the development (Storkel, 2004). Words of dense neighborhoods need less exposures to be learned than words with less phonological neighbors (Storkel, 2004). Words with a high phonological neighborhood density consist of frequently occurring sounds. These sounds are held better in the memory during short term processing, which supports the development of very detailed long-term word memory tracks (Sosa & Stoel-Gammon, 2012; Storkel, 2004; Walley, Metsala, & Garlock, 2003). Previous studies have shown that the variable phonological neighborhood is correlated with word frequency. Namely, higher frequency words tend to have more neighbors and lower frequency words tend to have fewer neighbors (Landauer & Streeter, 1973). Moreover, word frequency shows how often a person could encounter a word, which could have impact on when and if a word is learned (Rice et al., 1994).
Regarding the number of phonemes, the reverse length effect can reflect processing demands in the phonological lexicon. This effect implies that longer words are easier to produce than short words. However, there are very little individuals who are better at producing long words than short words (Best, 1995; Lambon Ralph & Howard, 2000; Whitworth et al., 2014).
Phonological buffer
The effect of the variable number of phonemes is thought to be indicative for difficulties in the phonological buffer. In general, better performance on words with fewer phonemes suggests a problem in the phonological buffer (Whitworth et al., 2014). Research shows that the number of phonemes is correlated with the variable phonological neighborhood. Namely, longer words tend to have fewer neighbors and shorter words tend to have more neighbors (Pisoni, Nusbaum, Luce, & Slowiaczek, 1985). As previously stated, children with (Developmental) Language Disorders ((D)LD; e.g. children with DLD and Language Disordered children associated with DS and HI) all have impairments in the phonological buffer. Children with DLD and HI have mostly difficulties with repeating (lengthy) nonwords. However, when comparing children with DLD and HI on the repetition of nonwords children with DLD score lower (Abbeduto et al., 2007; Chapman, & Sindberg, 2005; Conti-Ramsden, 2003; Dollaghan & Campbell, 1998; Roy & Chiat, 2004; Seung & Chapman, 2000). Moreover, children with DS have problems with nonword repetition and with digit span tasks, including the
production of intelligible speech sounds and problems in oral language and also phonological short-term memory (Abbeduto et al., 2007; Miolo, Chapman, & Sindberg, 2005; Seung & Chapman, 2000).
2.6 Objective and research question
As previously stated, there are differences and similarities in the lexical semantic, lexical phonology and phonological buffer processing abilities of the three language impaired groups (LD-groups): children with DLD, children with DS and children with HI. To explore these differences and similarities in the representations and processing of the language of language impaired children and typical developing children psycholinguistic variables can be used. However, few reports have compared those groups and examine their linguistic processing with psycholinguistic variables. Therefore, the objective of the present study is to examine whether psycholinguistic variables extracted from verbs and nouns in spontaneous speech can be used to characterize language processing differences between children with DLD, Down Syndrome, and Hearing Impairment. These three clinical groups are compared to MLU-matched typically developing children. Studying this might provide insight in the selective aspects of language processing that could differ between the groups, even though their overall productive abilities are comparable (Rice et al., 1992).
The main research question of this thesis is: ‘Do children with language disorders associated with DS and HI, and children with DLD differ in language processing from language-matched TD children?’. The second research question is: ‘Are those differences in language processing compared to TD children distinct between groups?’.
2.7 Hypotheses
Regarding the measures of lexical diversity, it is hypothesized that the children with (D)LD and the TD-groups will score similarly, because they do not significantly differ on the measure MLU, and the measures MLU, TTR, NDW and TNW influence each other. Given the difficulties of (D)LD-children with linguistic processing described in the theoretical background is hypothesized that the (D)LD-group will score weaker on all psycholinguistic measures and thus will differ in their linguistic processing from typically developing children. This implies that the children with (D)LD will use more high-frequent words and words with a high AoA rating compared to TD children. Moreover, children with (D)LD will use more concrete words and words of a higher phonological neighborhood compared to TD children. In addition, children with DLD will use words with less phonemes than TD children.
Regarding the lexical semantics, a developmental delay in children with DLD is well documented (Dockrell, Messer, George, & Ralli, 2003sheng; Sheng & McGregor, 2010). Children with
DS show difficulties in this domain as well. However, there is a big variability among children with DS (Berglund, Eriksson & Johansson 2001; Mervis & Robinson 2000). The study of Jerger et al. (2006) suggested that children with HI develop normal lexical semantic representations, but at the dynamics of their semantic processing seem to be changed by the presence of HI in childhood. For example, they seem to have an abnormally prolonged semantic stage of processing, which could reflect limitations in rapid object naming and slowed picture naming (Kail, 1996). However, it is not examined if that will have an impact on their spontaneous speech too (Kail, 1996). The variable concreteness is the only variable that can indicate problems in just the semantic lexicon (Gerhand & Barry, 2000; Storkel, 2009). Therefore, this thesis expects that especially children with DLD will use more concrete words than their matched TD-group. The children with DS and HI will score equal in comparison to their matched TD-group.
Regarding the lexical phonology, this thesis expects that children with the most severe lexical phonology problems, thus DS and DLD, will show the biggest effect on the variables that reflect processing in the phonological lexicon: frequency, AoA and phonological neighborhood. Therefore, the DS group and DLD group will use higher frequent words, words with lower AoA rating and higher phonological neighborhood density compared to the TD-group and the children with HI. This is because, as mentioned above, it is determined that impaired lexical phonology is a key deficit of children with DLD (Bishop & Snowling, 2004). Moreover, it is suggested that general language delay in Down syndrome leads to difficulties in the domain of the lexical phonology (Metsala, 1999). Children with HI can be limited in their lexical phonology development too. However, there is a large heterogeneity among HI children. Additionally, cognitive cues could play an important role in the linguistic processing to compensate for their poor phonological skills (Kallioinen et al., 2016).
The variable number of phonemes can reflect troubles with the processing demands in the lexical phonology, but also in the phonological buffer. If there is an impairment in the lexical phonology the reverse length effect will appear. However, there are very little individuals who are better at producing long words than short words and thus the reverse length effect is not likely to show (Best, 1995; Lambon et al., Whitworth, et al., 2014).
The phonological buffer of children with HI is administered to be a domain of difficulties. However, when comparing children with DLD and children with HI, children with DLD are expected to have bigger problems with the phonologic buffer (Briscoe et al., 2001). Moreover, children with DLD and also DS are known to have a deficit in phonological buffer (Conti-Ramsden, 2003; Dollaghan &
Campbell, 1998; Fowler et al., 1995; Jarrold & Baddeley, 1997; Jarrold et al., 2000; Roy & Chiat, 2004). Therefore, it is hypothesized that both the DLD-group and DS-group will show the typical length effect on the variable number of phonemes and thus use words with fewer phonemes than the TD-group and children with HI.
3. Method 3.1 Participants within the database
In the present study the spontaneous speech samples of three language disordered groups of children (LD-groups) and of typically developing children (TD-group) recruited by Bol and Kuiken (1990) were used. The spontaneous speech samples were obtained by Bol and Kuiken (1990) at the primary schools of the participating children and stored in the Child Language Data Exchange System project (CHILDES; MacWhinney, 2001).
To select the participants pairwise matching was used. The participants were chosen out of four groups, namely three (D)LD-groups (e.g. DLD, DS, HI) and a TD-group. The first (D)LD-group was the group of children with DLD who attended a school in Amsterdam for special education specialized in DLD. The IQ of the children with DLD was tested and fell within normal ranges. Therefore, the children with DLD had no sufficient psychological or intellectual restraints that could explain their problems with language acquisition.
The second (D)LD-group was the group of children with DS. All children were diagnosed as suffering from trisomy 21. Their mental age was 3.6 years or older. The level of intelligence was measured with a Dutch adaption of the Merrill-Palmer Preschool Performance Test (Stutsman, 1926). Their intelligence score differed from 20-56 with an average IQ of 40.7.
The third (D)LD-group was the group of children with HI. All children had sensorineural hearing loss or a mixed hearing loss. Moreover, all children were diagnosed before the age of 1.6 years. The hearing losses varied from 40- 85 dB pure tone average on the better ear, and the average hearing loss of the children with HI was 65.8 dB. The IQ score of the children with HI was within the average range.
According to the speech therapists of the children within these three (D)LD-groups the language production level of the selected children was not higher than the language production level of typically developing four-year-old. Another requirement was that their chronological age should be at least 3.6 years.
The TD children were recruited from playgrounds and day-care centers in Amsterdam. The group consists of 31 Dutch children (16 male and 15 female). Sixteen children have been recorded twice with at least 6 months’ time in between the recording moments. Therefore, the total number of recordings becomes 47.
3.2 Participants of this thesis
For the first research question a group of 40 (D)LD-children were selected out of the DLD-, DS- and HI-group and matched on MLU and gender to 40 TD-children. As can be seen in Table 1 the age of the TD-group had a mean of 2.28 with an SD of .64. The (D)LD-children had a mean age of 8.34 with
a SD of 4.77. The TD-group consisted of 19 females and 21 males. The (D)LD-group consisted in total of 17 females and 23 males. From the DLD-group 13 children (5 female, 8 male) were included, from the DS-group 14 children (6 female, 8 male) were included and from the HI-group 13 children (6 female, 7 male) were included. The mean MLU of both groups was 3.26. With the independent T-test the mean MLUs of both groups were compared. There was no significant difference found with p-score of 1.
For the second research question each (D)LD-group was matched to their own TD-group, which makes three different groups of participants. This allowed including all children with (D)LD available, and then creating a matched group for each (D)LD group. Some TD children were used as controls for more than one clinical group. As can be seen in Table 2 the first group consisted of 20 children with DLD (5 female and 15 male) with a mean age of 5.6 and a SD of .59. The TD-group consisted of 20 children of whom 7 female and 13 male. The mean age of the TD-group was 2 with a SD of .59. The MLU of the TD group was 3.665 with an SD of .94 and the MLU of the DLD group was 3.68 with an SD of .96. The MLUs did not significantly differ, namely with an independent T-test a p-score of 0.96 was found.
As can be seen in Table 3 the second group consisted of 20 children with DS of whom 10 male and 10 female, with a mean age of 13.5 and a SD of 4.14. The matched TD-group included 20 children (9 females 11 males) and the mean age was 1.97 with a SD of (.54). The mean MLU of the groups were equal (2.855). With the independent t-test the mean MLUs were compared to test if the groups significantly differed, the mean MLUs did not differ (p=1).
The third (D)LD-group consisted of 20 children with HI (9 female, 11 male) with a mean age of 5.67 and a SD of 1.41, as can be seen in Table 4. The TD-group consisted of 20 matched children of whom 10 female and 10 male. The mean MLUs were equal (3.215). This was also confirmed by the independent T-test that found a p-value of 1.
Table 1
Characteristics of the (D)LD-group and matched TD-group
Characteristics TD (D)LD Age (Mean, SD) 2.28 (.64) 8.34 (4.77) Female 19 17 Male 21 23 Total 40 40 MLU (Mean, SD) 3.2575 (1.01) 3.2575 (1.09)
Table 2
Characteristics of the DLD-group and matched TD-group
Characteristics TD DLD Age (Mean, SD) 2.55 (.5904) 5.63 (.59) Female 7 5 Male 13 15 Total 20 20 MLU (Mean.SD) 3.665 (.94) 3.68 (.96) Table 3
Characteristics of the DS-group and matched TD-group
Characteristics TD DS Age (Mean, SD) 1.97 (.54) 13.5 (4.14) Female 9 10 Male 11 10 Total 20 20 MLU (Mean, SD) 2.855 (1.04) 2.855 (0.966) Table 4
Characteristics of the HI-group and matched TD-group
Characteristics TD HI Age (Mean, SD) 2.29 (.62) 5.67 (1.41) Female 10 9 Male 10 11 Total 20 20 MLU (Mean, SD) 3.215 (.915) 3.215 (.886)
3.3 Data collection procedure
The spontaneous speech samples of the three (D)LD-groups were recorded at their schools. The children were playing with their speech therapists in a free-play situation. One of the two investigators of the study of Bol and Kuiken (1990), was also present in the room where the child was playing with the speech therapist. Occasionally the investigator participated in the conversation. Of all recordings of each child the utterances were transcribed by the investigator of the paper of Bol and Kuiken (1990), that was present at the recording of the spontaneous speech.
The TD children were audio recorded at home for one hour in a free play situation at their homes. During the recording two observers and at least one of the parents was present. Their
recordings were transcribed by one of the investigators. Afterwards, 100 successive utterances from the 11th minute of the audio recording onwards were transcribed.
3.4 Data coding procedure
The data sets are part of the CHILDES database and consequently available for academic purposes (Bol & Kuiken, 1990; MacWhinney, 2001). The spontaneous speech samples were transcribed according to the conventions of Codes for the Human Analysis of Transcripts (CHAT), the coding system of the Child Language Data Exchange System project (CHILDES; MacWhinney, 2001). The CHAT transcription format was designed for analysis by the program Child Language Analysis (CLAN; MacWhinney, 2017). The mean length of utterance (MLU) was already calculated by Bol and Kuiken (1990) and Type Token Ratio (TTR) was calculated after extracting the nouns and verbs. All numbers of different nouns and verbs and the total nouns and verbs per child were counted. Thereafter the number of different words was divided by the number of total words.
In CLAN the nouns and verbs were extracted from the transcripts. It was not possible to extract separately the nouns and verbs, because the samples were not tagged morphologically. Therefore, it was not possible to search for specific grammatical classes. This means that the proper nouns, adjectives, adverbs, determiners, prepositions, pronouns, conjunctions, and repetitions of words in the transcript are not removed. Therefore, all repetitions and words that did not belong to the grammatical classes noun or verb, were removed by hand. Subsequently, the list of nouns and verbs was checked for synonyms which occurred in CHILDES and replaced them if necessary, with the most adult form of the word. This decision was based on which words could provide most valid psycholinguistic data in the program ‘SUBLEX-NL’ (e.g. mam/ma/mama; Keuleers et al., 2010). Furthermore, for the verbs all conjugates were removed whereas the infinite form of the uttered words remained.
Subsequently, values of psycholinguistic variables were extracted from different corpora for each word. The norms of SUBTLEX-NL, which is a database of Dutch word frequency based on 44 million words from film and television subtitles (Keuleers, Brysbaert & New, 2010), was used for obtaining the lexical statistics of frequency (Freq; Baayen, Piepenbrock, & Gulikers, 1995). For the variables phonological neighborhood and phonological length, the database CLEARPOND was used, which is the Cross-Linguistic Easy-Access Resource for Phonological and Orthographic Neighborhood Densities (Marian, Bartolotti, Chabal, & Shook 2012). For the variables: Age of Acquisition (AoA; Kuperman, Stadthagen-Gonzalez, & Brysbaert, 2012), and Concreteness (Conc; Brysbaert, Warriner, & Kuperman, 2014) the norms of Brysbaert, Stevens, De Deyne, Voorspoels, and Storms (2014) were used. They collected concreteness ratings for 30,070 words and AoA ratings for 31,178 words and stored them in spreadsheets. Following, the lexical statistics of frequency, AoA and concreteness were
copied for each word with the formula VLOOKUP in Microsoft Excel. Moreover, Microsoft Excel was needed to merge and save all acquired data. Additionally, individual means and standard deviations (SD) were calculated for all word properties. The statistical analyses were calculated with SPSS Statistics 25.
3.5 Statistical analyses
The data was first analyzed in a descriptive way. The mean and SD of the age, gender and MLU were described and put in Tables 1-4. In addition, the group means and SDs of all variables of all four groups were calculated with Microsoft Excel and placed in Table 6-9. To examine if the data was normally distributed the Shapiro-Wilk test was administered. The data was not normally distributed for all variables as can be seen in Table 5. Therefore, nonparametric statistics were used. First, the Mann Whitney U test was administered to determine differences between the lexical statistics of the(D)LD-group and their matched TD-group. Subsequently, the Mann Whitney U test was administered to examine if there was a difference between the lexical statistics of the DLD-group and their matched TD-group, the DS-group and their matched TD-group and the HI-group and their matched TD-group. Thereafter, the outcomes of the comparisons of each (D)LD-group to their matched TD-group was compared qualitatively. Moreover, the data was represented in boxplots. A significance level of p < 0.05 was used and tested two-tailed to examine if the mean was significantly greater or smaller.
4. Results
This section presents the results of the analysis described in the previous section by providing and representing the descriptive statistics and the results of the statistical analyses in Tables 6-9 and text. The results will be discussed per research question, in the same order as described in the introduction. In Figure 1 – 14 the distribution of the data is represented in boxplots.
To examine if the data was normally distributed the Shapiro-Wilk test was administered. Out of the 112 variables 24 variables were not normally distributed. The non-normal distributed variables are displayed in Table 5. Therefore, non-parametric testing is used for all variables to ensure that the results can be compared.
4.1 Group means and group SDs
In Tables 6-9 the group means and group Standard Deviations (SDs) are shown for all variables per group. Table 6 shows the group means and group SDs of the (D)LD-group and the matched TD-group. Table 7 shows the group means and group SDs for the DLD-group and their matched TD-TD-group. Table 8 shows the group means and SDs of the DS-group and their matched TD-group and Table 9 shows the group means and SDs of the HI-group and their matched TD-group.
Table 5
Non-normally distributed data per variable per group
Variable Group Test Statistics
TTR DS W(20)=.902, p=.045 TD-DS W(20)=.894, p=.032 NTW DLD W(20)=.891, p=.028 MeanFrequency TD-DLD W(920)=.899, p=.04 HI W(20)=.864, p=.009 MeanAoA TD-LD W(41)=.915, p=.005 TD-HI W(23)=.875, p=.008 MeanNumber of phonemes DS W(920)=.898, p=.039 SD Frequency LD W(39)=.889, p=.001 TD-LD W(41)=.923, p=.008 DS W(20)=.898, p=.038 TD-HI W(23)=.893, p=.018 SDAoA TD-LD W(41)=.436, p=.000 TD-DS W(20)=.490, p=.000 SDConcreteness TD-LD W(41)=.386, p=.000 TD-DS W(20)=.417, p=.000 SDNumber of phonemes LD W(39)=.900, p=.002 TD-LD W(41)=.58, p=.000 DLD W(20)= .903, p=.047 TD-DLD W(920)= .904, p=.048 TD-DS W(20)=.633, p=.000 SDPhonological Neighborhood LD W(39)=.927, p=.014 TD-LD W(41)=.887, p=.001 DS W(20)=.901, p=.044