Analysing Signs of Processing Limitations in Narratives of
9 to 12-year-old children with DLD: Macrostructure,
Microstructure and Dysfluencies
Master’s Thesis General Linguistics (Clinical Track), University of Amsterdam
Bente Jasmijn van der Sluis
Student Number: 12783277
Supervisor: Prof. Dr. Judith Rispens
Second Reader: Dr. Ileana Grama
Date: 20
thof June, 2020
External Project
Company Name: Royal Auris Group
First Supervisor: Dr. Britt Hakvoort
Second Supervisor: Dr. Iris Duinmeijer
1
Abstract
Background: Previous research has identified instances of imbalances between the investment in narrative microstructure (i.e. linguistic form) and macrostructure (i.e. narrative content) in stories told by children with developmental language disorder (DLD). This is thought to result from processing limitations. Previous researchers have also identified increased instances of dysfluencies in tasks with a high processing load, like generated storytelling.
Purpose: Both the finding of imbalances in macro- and microstructure, and increased
dysfluencies in tasks with relatively high processing loads, have only been sparingly investigated, with varying results. This may partly be due to the fact that there is a plethora of ways to measure narrative ability, thus different studies may arrive at different conclusions. The goal of this study is to consolidate, and re-investigate with updated methods, the sparse previous findings in
narrative examination which support the theory of a processing deficit at play in DLD. Logically, children with imbalances between micro- and macrostructure investment should also produce more dysfluencies, when both show evidence of processing deficits. Bidirectional imbalances (i.e. findings of both low linguistic form with high narrative content and high linguistic form with low narrative content) could also provide more power to the processing theory behind macro- and microstructure imbalances, due to the untypical nature of such a result for children with DLD. Method: Narratives were elicited from sixty-seven 9 to 12-year-old Dutch children with DLD, by making use of the story generation book Frog, Where Are You? (Mayer, 1969). Both macro- and microstructural performance was calculated by making use of all narrative variables which have proven themselves to be sensitive to the presence of disordered language. Dysfluencies were categorised as fillers, repairs and pauses.
Results: 49 out of 67 children showed bidirectional dissociations between their micro- and macrostructural investment. A correlation analysis showed the estimate of the correlation coefficient between microstructural and macrostructural measures to be non-significant. There was no increase of dysfluencies in the imbalanced groups, rather the balanced group showed a proportionate increase in certain dysfluency indices compared to the imbalanced groups; perhaps content and form are not the only narrative dimensions that suffer under cognitive capacity limitations.
Conclusions: Previous findings of (bidirectional) imbalances have been consolidated in this study, using all available and sensible variables. Further work has to be undertaken to better understand the function that dysfluencies may have in signalling processing restrictions.
2
Table of Contents
Chapter Page 1. Introduction……….………...4 2. Theoretical Background……….………6 2.1 Narrative Abilities in DLD……….………..6 2.1.1 Macrostructure………6 2.1.2 Microstructure……….……....82.1.3 Content versus Form……….10
2.2 Processing Hypotheses………...11
2.3 Current Study……….13
3. Research Questions and Hypotheses………...…………...……..16
3.1 Question and Hypothesis 1……….………16
3.2 Question and Hypothesis 2……….16
3.3 Question and Hypothesis 3……….17
4. Methods………18 4.1 Participants……….18 4.2 Materials…………...………..19 4.3 Procedures..………20 4.3.1 Task Administration………..20 4.3.2 Story Transcription………20 4.4 Data Analysis……….21 4.4.1 Macrostructure………..21 4.4.2 Microstructure………...22 4.4.3 Communicative Breakdowns………23 5. Results………..24
5.1 Analysis of Output Data…...………..24
5.2 Macrostructure………...24
5.3 Microstructure………25
5.4 Communicative Breakdowns……….26
5.5 Dissociation..………..26
3
6. Discussion………31
6.1 General Discussion……….…………31
6.2 Limitations and Future Directions..………....35
6.3 Clinical Implications………..35 7. Conclusion………37 References………38 Appendix A………..45 Appendix B………..46 Appendix C………..47
4
Chapter 1
1. Introduction
Storytelling is an intrinsic and integral part of being human. For thousands of years,
narratives have been passed on to family, friends and younger generations by use of written- and oral language (Fulford, 1999). No wonder, therefore, that storytelling has become a staple in our present-day academic, professional and social activities. However, successful and gripping storytelling is not a given, innate skill for human beings; our holistic narrative abilities are moulded by uniting multiple components, including cognitive, linguistic, and social-emotional abilities (Torng & Sah, 2020). As Trabasso and Rodkin (1994) explain, narratives are made up of a causal chain of events, that almost always leads to an ultimate goal that is attempted to be realised by the narrative. Therefore, cognitive processing and other executive functions like inhibition, updating, working memory, planning, attention and organisation play a large role in successful storytelling (Montgomery, Polunenko &
Marinelli, 2009). Besides, no gripping, relatable narrative can be told if a person’s theory of the world is not well-developed; unspoken, implicit references cannot be made if world knowledge is lacking, and the conceptualisation of the story will be poor (Duinmeijer, De Jong & Scheper, 2012). Also, good linguistic abilities are imperative if one wants to recount a decent narrative, as telling a story requires advanced phonetic, phonological, pragmatic, semantic, and morpho-syntactical capabilities (Manhardt & Rescorla, 2002).
Thus, it may come as no surprise that children with difficulties in one or more of these domains also have difficulties with developing adequate narrative abilities. For instance, children with Developmental Language Disorder (DLD) display difficulties with language even under conditions of normal intelligence and neurological functions. Their difficulties include impaired lexical, syntactical, phonological, and pragmatic skills (Leonard, 2014); non-linguistic cognitive and social cognitive problems including world knowledge, Theory of Mind and social understanding (Bishop, 1992; Craig, 1995; Blankenstijn & Scheper, 2006); and reduced executive functioning (e.g. Weyandt & Willis, 1994; Im‐Bolter Johnson & Pascual‐Leone, 2006; Bishop & Norbury, 2003). Resultantly, they have been found to be poorer in narrative production compared to typically-developing children (e.g. Colozzo, Gillam, Wood, Schnell, & Johnston, 2011; Tsimpli, Peristeri, & Andreou, 2016).
Furthermore, studies like Colozzo et al. (2011), Gillam and Johnston (1992), and Lam, Scheper and Rispens (2015) have found that skills of children with DLD – and not of
5
typically-developing children -- in macrostructure (i.e. narrative content) and in microstructure (i.e. linguistic form) seemed dissociated. For instance, adequate use of conjunctions at microstructural level does not correlate with the ability to construct causal relations at macrostructural level. Previously, researchers like Colozzo et al. (2011) and Fichman, Altman, Voloskovich, Armon-Lotem, and Walters (2017) have already
hypothesised that children with DLD are poor at juggling different parts of narrative tasks due to the burden that such a task’s complexity puts on their processing systems. That is, excelling microstructural performance might cause macrostructural performance to drop due to processing constrainments, or the other way around. However, only three relatively dated studies have explicitly and chiefly investigated dissociations between macro- and
microstructural ability in children with DLD, and only one has investigated pupils in the same age range as the current study. Plus, all previous studies included a fairly small number of children, which makes generalisation to a population problematic. Besides, it is only hypothesised that processing deficits cause this imbalance in certain children. Support for this hypothesis might be found in any bidirectionality of potential imbalances, and also instances of communicative breakdowns or dysfluencies. As previous researchers explain (MacLachlan and Chapman, 1988; Wagner et al., 2000), dysfluencies increase when children with DLD are subjected to narrative tasks with a greater processing load. Therefore, the hypothesis might be better supported if children with imbalances between micro- and macrostructure investment also produced more dysfluencies.
It is imperative to know whether children with DLD actually experience processing limitations, in order to identify the difficulties and bottlenecks of children with DLD on narrative tasks. This may lead us to better understand the underlying deficit that generates their linguistic problems, and to allow us to recognise what strategy can be adopted to help shape the children into the best storyteller they can be, as this will ensure greater success in their professional, academic or social lives (e.g. Miniscalco, Hagberg, Kadesjö, Westerlund & Gillberg, 2007; Tannock & Schachar, 1993). As such, the current study will again investigate the macro- and microstructural abilities of children with DLD, and analyse the instances of dysfluencies among different subgroups, while employing modern procedures and considering the newest metrical and theoretical insights. Firstly, an overview of previous research on this topic and the current study’s objectives will be given in Chapter 2 and 3. In Chapter 4 and 5, the methodology and subsequent results will be explained. Chapter 6 and 7 make up the discussion and conclusion.
6
Chapter 2
2. Theoretical Background
2.1 Narrative Abilities in Developmental Language Disorder
As explained before, successful storytelling depends on many factors (Hudson & Shapiro, 1991; Kaderavek & Sulzby, 2000; Colozzo et al., 2011): world knowledge, knowledge that is specific to the genre of the task like conventional introductions (e.g. ‘Once upon a time…’), structural knowledge (e.g. causal events that create a plot), and linguistic skill (e.g. complex syntax and devices used for referencing, and to create causal and temporal connectives). Importantly, these skills need to be managed and employed in real-time during live narration, thus upping the narrative task’s processing requirements (Liles, 1993; Owens, 2007).
Children with Developmental Language Disorder (DLD), a developmental language impairment diagnosed under conditions of normal nonverbal measures of intelligence, are known to perform more poorly on narrative tasks than age-peers, sometimes both in their stories’ macrostructural (i.e. narrative content) and microstructural (i.e. linguistic form) performance (e.g.Fey, Catts, Proctor-Williams, Tomblin & Zhang, 2004; Scott & Windsor, 2000). These findings will be explored below.
2.1.1 Macrostructure
The term macrostructure is used to indicate the storyteller’s production of global narrative content; the higher-order hierarchical organization like story episodes (Justice et al., 2006). Certain studies that have investigated macrostructural ability in children with DLD have found a reduced performance on macrostructural indices compared to typically-developing (TD) children (e.g. Merritt & Liles, 1987; Liles, 1987; Olley, 1989; Duinmeijer et al., 2012; Manhardt & Rescorla, 2002). For example, children with DLD have been found to produce less story grammar elements and propositions (Bishop & Donlan, 2005; Merritt & Liles, 1987; Reilly et al., 2004), and fewer ‘cognitive state terms’ (Bishop & Donlan, 2005;
Manhardt & Rescorla, 2002). That is, compared with TD children, children with DLD did not to the same extent mention whether the character seemed confused, happy, etc., -- which is part of macrostructural story evaluation.
In most previous studies, macrostructure is typically solely expressed in terms of episodic structure, as measured by Stein and Glenn’s (1979) story grammar (SG) conceptual
7
framework; Stein and Glenn (1979) state how the best story must include a setting, an initiation of a problem or plot, an internal response to the initiation, an attempt to resolve the problem, an outcome, and a final reaction to this outcome. At the base of the SG framework lies the notion that narratives are a series of events that are mostly causally (i.e.expressing a cause) or temporally linked (i.e. with regard to time). This notion is supported by a corpus-based, pioneering study conducted by Tomai, Thapa, Gordon, and Kang (2011: 77), where 283 TD children’s individual narrations of identical stories were analysed in order to investigate ‘the role that causality plays in determining whether subjects will mention a particular story event’; the authors found that any mentioned narrative event was nearly always causally linked to a previous event.
However, recent findings in the field of narrational macrostructure suggest that the SG framework might not cover all of a narrative’s actual global content; Fichman et al. (2017) analysed retold narratives produced by 35 TD bilinguals and 14 bilinguals with DLD in both their first language, Russian, and their second language, Hebrew. Importantly, the authors measured both the inclusion of SG elements (e.g. the setting, initiating event and ending), and causal relations between SG elements (i.e. enabling, physical, motivational, and psychological relations) (see Table 1). They found that causal relation-performance differed more between the two groups than performance on SG elements. For example, the narratives of the children with DLD contained significantly fewer enabling and physical relations than the typically-developing group, and causal relations produced by the DLD group were more often unsuitable for the grander plot. The authors concluded that SG elements along with causal relations could significantly distinguish the narratives of the two groups; this confirms that one might need to examine a wider array of macrostructural features in narratives, as more diverse measures might give a better chance of spotting disordered language
performance than only measuring a child’s story grammar adherence.
Causal Relation Definition Example
Enabling Connects an attempt to a goal or to
another attempt in the following episode The dog tried to catch the bees (attempt). The boy saw it and decided to run away (goal). Physical Connects attempts and outcomes within
episodes The dog tried to swim away (attempt). But he got stuck on the boy’s head (outcome).
Motivational Connects goals and attempts within
episodes He said shhh (attempt). Because he wanted to hear the frog (goal). Psychological Establishes relations between internal
responses (IR) and attempts or outcomes. The boy was happy (IR). Because they found the frog (outcome). Table 1: Causal Relations
8
Fichman et al.’s (2017) method of analysing the narratives by use of causal relations is based on a causal network model designed by Trabasso and Sperry (1985). Causal relations interact with SG elements in the causal network model in order to shape the narrative (see Figure 1).As prefaced before, causal relation systems seem to be more sensitive to language impairment than only SG elements when assessing narrative performance (Trabasso & Sperry, 1985; Trabasso & van den Broek, 1985). This has also been reported by Kupersmitt and Armon-Lotem (2019), who investigated the narration of 150 children (5;0-7;0), of whom 45 were Hebrew monolinguals (19 with DLD), 57 were English–Hebrew bilinguals (20 with DLD) and 48 were Russian–Hebrew bilinguals (21 with DLD). They found that all children with DLD performed worse on the expression of causal relations than on SG elements.
A last index that can be used to assess macrostructural ability is evaluation. As
Bamberg and Damrad-Frye (1991) explain, evaluation by the speaker can be used as a way to elaborate on narrative content and orally interpret causal links. The speaker may also use evaluation to give more insight into their perspective of an event, and in order to engage listeners more. Examples are the use of frames of mind, evaluative words, hedges, character speech, and attention getters, like sound effects. Evaluative devices used by children with DLD have only been examined a handful of times. However, Manhardt and Rescorla (2002), and Reilly et al. (2004), using the Frog, where are you? story (Mayer, 1969), described how the children with DLD in their studies used less evaluative devices than the TD children. Yet, Norbury and Bishop (2003), also using Frog, where are you?, stated there were no
Figure 1: Causal Network Model, where the blocks are SG elements and the arrows are the causal relations
9
differences between groups in terms of evaluation. Though, it has to be noted that different age-ranges were included between the studies (Manhardt and Rescorla 8;0-9;0 / Reilly et al. 3;0–12;0 / Norbury and Bishop 6;0–10;0), which is problematic for comparison as narrative abilities evolve as children get older. Also, as Liles (1985) explains, how many evaluative comments children produce depends greatly on the amount of interaction, shared knowledge and joint attention between the examiner and the examinee; it is not obvious from the
description of the studies, how familiar and friendly the researchers were with the children. It is clear that evaluation used in narration by children with DLD is a topic that still requires further research.
2.1.2 Microstructure
The term microstructure is used to indicate the storyteller’s production of word-and-sentence level language, i.e. linguistic form. This includes cohesion, lexicon and syntax. Almost all studies on microstructure in the narratives of children with DLD have found a reduced performance on it. For instance, researchers like Fey et al. (2004), Gillam and Johnston (1992), Norbury and Bishop (2003), and Reilly et al. (2004) have found stories told by children with DLD to be less grammatically accurate. Other researchers have found their stories to have a lower Mean Length of T-Unit in Words (MLT-W) -- where a T-unit is a sentence with all its dependent clauses included (Bishop & Donlan, 2005) -- and also to be less syntactically complex (Bishop & Donlan, 2005; Fey et al., 2004; Gillam & Johnston, 1992; Norbury & Bishop, 2003; Reilly et al., 2004).
In 2006, Justice et al. developed and published a clinical tool designed for investigating school-aged children’s performance in narration, and specifically in
microstructural narration: The Index of Narrative Microstructure (INMIS). The variables used in this tool include productive (e.g. linguistic output volume) and complexity variables (e.g. grammatical complexity), which are microstructural variables that had prior to the Justice et al. study been identified in the literature as sensitive measures to language impairment (e.g. Baltaxe & D’Angiola, 1992; McFadden & Gillam, 1996; Gillam & Johnston, 1992; Liles, 1985, 1987). Productivity is defined as the total sum of words, the T-unit’s mean length in words (MLT), the lexical diversity (i.e. the total number of different words), and the narrative’s number of T-units. Researchers like Klee (1992) and Strong et al. (1998) have shown that lexical diversity measures can reliably identify differences in child vocabulary use. As Justice et al. (2006) explain, story length is on the other hand a sensitive measure of story output, as language output progresses with language ability and during
10
ageing (e.g. Atkins & Cartwright, 1982; Strong et al., 1998). Structural complexity can be measured by how many T-units contained at least two finite verbs, and how many
coordinating and subordinating conjunctions were used.
An additional microstructural measure is cohesion, which can be measured by the inclusion of five cohesive ties in narratives (Halliday & Hasan, 1976); lexis, conjunction, ellipsis, substitution and reference (see Table 2). Cohesive narratives are formed by overt connections between the T-units in a story (Halliday & Hasan, 1976; Liles, 1985). Several researchers, like Baltaxe and D’Angiola (1992) and Liles (1985), have shown that children with DLD have a reduced performance on the use of cohesive ties compared to TD children.
2.1.3 Content versus Form
Interestingly, some studies find decreased macrostructural ability in children with DLD (e.g. Colozzo et al., 2011; Duinmeijer et al., 2012; Liles, 1987; Manhardt & Rescorla, 2002; Merritt & Liles, 1987; Olley, 1989; Torng & Sah, 2020; Lam, Scheper & Rispens, 2015), while other researchers find that macrostructural narration is not significantly affected in children with DLD (e.g. Boudreau & Hedberg, 1999; Kaderavek & Sulzby, 2000; Norbury & Bishop, 2003; Tsimpli et al., 2016). Furthermore, some studies that used Stein and Glenn’s (1979) SG metric to measure macrostructure production (e.g. Merritt & Liles, 1987; Olley, 1989; Manhardt & Rescorla, 2002), reported fewer SG components for children with DLD than for TD children. On the other hand, different researchers (e.g. Boudreau & Hedberg, 1999; Kaderavek & Sulzby, 2000; Norbury & Bishop, 2003) report no group differences in SG usage between both groups.
An explanation for the observed differences in outcome may lie in methodological incongruities between studies, which in turn may make these studies invalid for comparison with one another. For instance, Kaderavek and Sulzby (2000) used only three story
components in their assessment of children’s macrostructural ability, while Merritt and Liles (1987) used six story components in their story grammar metric. As Torng and Sah (2020) suggest, it could be the case that the metric used by Kaderavek and Sulzby (2000) is
Cohesive Tie Example
Reference The boy looks in the pot. HE sees it is empty. Conjunction The boy looked in the pot. BUT it was empty. Substitution The dog chased the bees. But he didn’t catch ONE Lexical They fell into the pond. THERE was a large splash. Ellipsis The boy is sleeping. AND THE DOG [..x..] TOO.
Table 2: Cohesive Ties
11
inadequate for reflecting all of the difficulties that children with DLD face in macrostructure. What’s more, researchers like Pearce, James and McCormack (2010) and Duinmeijer et al. (2012) have shown and explained how story retelling tasks like the Bus Story (Cowley & Glasgow, 1994) are less cognitively demanding than story generation tasks like the Frog Story (Mayer, 1969). Studies like Manhardt and Rescorla (2002) and Olley (1989) used story generation tasks, while Boudreau and Hedberg (1999) and Kaderavek and Sulzby (2000) used retelling tasks; this may have led to differences in macrostructure results, as a less cognitively demanding task may allow children with DLD to produce both more appropriate microstructural narration and more macrostructural narration. It may also be the case, as explained in Part 2.1.1 (p.7), that studies that used only SG components to assess differences between groups instead of (also) employing causal relations, may have inadvertently used a less-sensitive measure.
Yet, while some of these discrepancies might very well be due to methodological differences that exist between certain studies, other studies with opposite results on performance on macrostructure (e.g. Norbury & Bishop, 2003 v. Manhardt & Rescorla, 2002), used very comparable methodologies. How could there still be such inconsistency? Some may hypothesise that more children in Norbury and Bishop’s sample invested their cognitive strength in narrative content, thus decreasing performance on linguistic form, and that Manhardt and Rescorla’s sample did the opposite. That is to say, researchers like Colozzo et al. (2011), Gillam and Johnston (1992), and Lam, Scheper, and Rispens (2015) hypothesise and empirically prove that macrostructural and microstructural ability can be dissociated or imbalanced in children with DLD. A linguistic form-focussed child with DLD may try to excel at a microstructural level, but due to the narrative task’s complexity and processing density, performance at macrostructural level will be negatively affected. The opposite is true for a narrative content-focussed child. This may be a result of the limited processing ability of children with DLD.
2.2 Processing Hypotheses
Studies that focussed on non-narrative processing have found capacity limitations in children with DLD. For example, Leonard et al. (2007) examined 14-year-olds with DLD, and showed how they have much more trouble with a host of verbal and non-verbal processing tasks compared to age-matched controls, both in terms of completion speed and accuracy. However, the precise cause of processing limitations is not yet clear; Gillam and Hoffman (2004) reviewed a large number of processing studies, and found that explanations like
12
processing speed (e.g. Miller, Kail, Leonard & Tomblin, 2001), temporal processing (e.g. Merzenich et al., 1996), working memory abilities (e.g. Gathercole & Baddeley, 1990), and lexical representational quality (e.g. McGregor, Newman, Reilly, & Capone, 2002) could all be candidates for the underlying cause of task capacity restrictions in children with DLD.
Related to this, Bock (1982) and Hargrove, Frerichs, and Heino (1999) were some of the first to suggest that in narration, a finite linguistic capacity exists. This can be fittingly explained with Crystal’s (1987) ‘‘bucket’’ theory for disordered language; in language tasks, there exist trade-offs in language parameters due to processing constraints. That is to say, a child with DLD that produces macrostructurally advanced language might ‘trade-off’ investment in linguistic performance, or the reverse. It seems that telling a story with an elaborate plot line, which is at the same time structurally well-organised, is not an easy task for children with DLD-- as has been attested by several studies (e.g. Baltaxe & D’Angiola, 1992; Copmann & Griffith, 1994; Gillam & Johnston, 1992). That is why it is important to measure both macro- and microstructure in clinical settings: a child that excels at
microstructural level might seem to be faring well in their syntactical, grammatical, and lexical language development, which are often thought of as the hallmarks of language impairment in DLD (e.g. Leonard, 2014), and so may appear to have non-impaired language development. However, if one does concurrently measure macrostructural ability, a very different picture might be painted, namely a picture of a child who struggles very much with juggling the complexities of a narrative task, and thus trading off in story content -- a skill of high importance in later life.
However, telling a relatively uncomplicated story like the Frog, Where Are You? story (Mayer, 1969) does not seem as cognitively demanding for the typically-developing child. Blankenstijn and Scheper (2003) conducted a study into the performance of Dutch TD children (4;0 till 9;0) on the Frog Story test, using narrative data collected by Roelofs (1998), Duinmeijer (2010), and Duinmeijer, de Jong and Scheper (2012). In terms of the
development of TD children on narrative content, 9-year-olds already mentioned 81% of all possible episodes (for reference, adults produced 100%). Children older than 9;0 were not included in this data, as certain studies suggest that narrative development for TD children reaches its peak at around 10;0 (Izard et al., 2001; McKeough, 2000). On the other hand, children with DLD tend to perform below-average on narrative tasks even after the age of 9, and ‘bucketing’ also occurs beyond this age, as Gillam and Johnston (1992) showed in their study on narrative dissociations in children between 9;0 and 12;0.
13
Empirical evidence of high processing demands of narration, comes for instance from studies that look at communicative breakdowns (i.e. dysfluencies) like repairs, fillers, pauses and abandoned utterances in children’s speech. For example, Wagner, Nettelbladt, Sahlén, and Nilholm (2000) analysed instances of such dysfluencies in 5-year-old children with DLD, who were engaged in both a conversational and a narrative discourse. In the narrative task, the children experienced significantly more communicative breakdowns. MacLachlan and Chapman (1988) similarly reported that for children aged 10;0-11;0 with and without DLD, the children with DLD had more dysfluencies in their speech during narration. Wagner et al. (2000) and McLachlan and Chapman (1988) both describe how the children with DLD produced more breakdowns in narration versus conversation. As Colozzo et al. (2011:1612) explain, this provides ‘empirical support for the view that narrative production makes particularly high processing demands’.
As explained before, macro- and microstructural imbalances in a child’s narrative are also thought to occur due to cognitive capacity (i.e. processing) limitations of children with DLD. However, the phenomena of observed imbalances in macro- and macrostructural investment is only hypothesised to be caused by certain processing difficulties. So, the desired support for this hypothesis might be found in instances of dysfluencies. As said, dysfluencies increase when children with DLD are subjected to narrative tasks with a greater processing load. Therefore, the processing hypothesis might be supported if children with imbalances between micro- and macrostructure investment also produced more dysfluencies. However, so far no study has looked at explicit relations between micro- and macrostructural performance and dysfluencies.
Furthermore, more support could be gathered for the processing hypothesis if potential imbalances between micro- and macrostructural performance occurred
bi-directionally. That is, children with DLD are uncontroversially known to perform poorly in microstructural, linguistic form indices like grammaticality (e.g. Leonard, 2014). Therefore, a relatively better score on macro- than microstructure may not be surprising, and does not inherently support a processing theory. However, if one were to also find evidence of decreased macrostrucutral performance combined with relatively better performance on microstructure by the children with DLD, the processing hypothesis would be lent further support. This ‘complimentary pattern’ of imbalances was only found for the first time by Colozzo et al. (2011), while older studies like Gillam and Johnston (1992) have not found evidence for bidirectionality.
14
2.3 Current Study
As has become clear, research is still not conclusive on the question if macrostructural narrative ability is weakened in children with DLD, or not. Some of the incongruous results of previous studies on macro- and microstructure might have to do with the fact that some studies used less-cognitively demanding tasks (e.g. story retelling instead of generation), different metrics, or that the sample included children who mostly ‘traded’ performance at microstructural performance, in order to excel more at macrostructure. Of the myriad of studies laid out in the theoretical framework, only three have explicitly investigated dissociations between macro- and microstructural ability in children with DLD as part of their research- or sub-questions. Therefore, it is not proven whether macro- and
microstructural imbalances could account for some of the inconclusive findings of previous studies on macrostructural ability in children with DLD. Only one study on (narrative) content versus (linguistic) form dissociations has investigated children in the same age range as the current study. Furthermore, all previous studies included a fairly small number of children; this leads a situation where it is not possible to draw any population-wide
conclusions based on the samples and results of these studies. Additionally, the studies were conducted in 1992, 2011, and 2015; since these studies surfaced, new and progressed insights into narrative metrics and narrative components have considerably grown. For example, we now know that story grammar metrics alone are insufficient to truly measure macrostructural ability, and that cohesion also plays a part in microstructural ability. In Table 3 (p.15), an overview is given of the metrics as used by these previous studies.
Additionally, as Manhardt and Rescorla (2002:2) mention, previous studies that did look at story grammar structure, cohesion, and evaluative information, investigated them as ‘individual dimensions’, but very few researchers have examined these dimensions
simultaneously as a macrostructural index. What is more, no studies had examined these three aspects of narrative skills in the narratives of children with (histories of) DLD, before Manhardt and Rescorla (2002) did so. However, Manhardt and Rescorla did not explicitly investigate dissociations between macro- and microstructure in their sample. No study so far has thus included these three macrostructural measures while concurrently investigating dissociations between form and content as their main research objective.
In this study, the macro- and microstructural abilities of sixty-seven children with DLD will again be investigated, employing current procedures and considering the newest metrical insights. That is, macrostructural ability will be measured by not only the inclusion of SG components, but also causal relations, and the inclusion of story evaluation. For
15
microstructural ability, the indices as used by the INMIS will be used (Justice et al., 2006), along with several additional sensitive measures like the child’s MLT-5 (i.e. the mean length of the five longest T-units), the number of times tense was omitted, the percentage use of simple past tense (note: the use of the simple past in narration increases as a child’s narrative abilities grow (Blankenstijn & Scheper, 2003)), the number of ungrammatical T-units, and narrative cohesion.
Furthermore, communicative breakdowns have not before been linked to or
investigated alongside dissociations in form versus content, while both are theorised to signal processing limitations. Therefore, it can also be hypothesised that children who show
dissociations (i.e. imbalances) between macro- or microstructure, and thus are assumed to have more severe capacity limitations, also produce more dysfluencies. This would lend additional support for both of the processing hypotheses. Additionally, more support could be gathered for the processing hypothesis if potential imbalances between micro- and
macrostructural performance occurred bi-directionally (i.e. low micro/high macro and high micro/low macro), as more robust evidence of child’s a true inclination to ‘bucket’ either of the language parameters due to cognitive capacity limitations.
Finally, the end goal of this study is to investigate more deeply and provide tangible evidence for or against processing limitations in children with DLD -- by analysing the correlation between macro- and microstructural indices, analysing the potential bi-directionality of imbalances, and analysing the proportionate frequency of dysfluencies among the balanced and imbalanced groups. Finding evidence for or against processing limitations is important in order to recognise what strategy can be adopted to best help children with developmental language impairments.
Study Macrostructural Measures Microstructural Measures Participant Number + Age Gillam and
Johnston (1992) Amount of ………. propositions; plot points; predicate types; dyadic constituents Amount of…….….. morphemes; T-units; complex correct T-units; connectives 10 DLD 9;0-12;0 Colozzo et al. (2011)
Stein and Glenn’s (1979) story grammar system Amount of…….….. grammatical errors; appropriate tense; clear references 13 DLD / 20 DLD Around 9;0 Lam, Scheper and Rispens, (2015) Included plot elements; type-token ratio
MLU + amount of... complex utterances; grammatical mistakes; grammatical mistakes per T-unit 16 DLD 4;0-5;0
16
Chapter 3
3. Research Questions and Hypotheses 3.1 Question and Hypothesis 1
3.2 Question and Hypothesis 2
Is there a relation between micro- and macrostructure in the narratives of children with DLD aged 9;0 to 12;0, while employing all narrative variables which have proven themselves to be sensitive to the presence of disordered language?
Previous research has identified processing limitations in children with DLD (e.g. Gathercole & Baddeley, 1990; Miller, Kail, Leonard & Tomblin, 2001). Due to the processing demands that a narrative generation task like the Frog, Where Are You? story places on the children, children who experience serious capacity limitations will resort to ‘bucketing’ either performance in macro- or microstructural measures. This study hypothesises that there will not any positive correlation between micro- and macrostructure or between micro- and macrostructural variables.
Colozzo et al. (2011) mentioned that in their Texas and Kansas sample, ‘[a]ll but three of the children with SLI [note: Specific Language Impairment is referred to as DLD in this study] […] fell into the extremes of the distribution, obtaining relatively higher scores on either content (n = 9) or form (n = 8)’. Lam, Scheper and Rispens (2015) found that their sample of 16 children with DLD mostly focussed on macrostructural performance, thus excelling in content, but there were some children who performed better on microstructure than macrostructure. In Gillam and Johnston’s (1992) study, no
HYPOTHESIS 1
HYPOTHESIS 2 RESEARCH QUESTION 2 RESEARCH QUESTION 1
17
3.3 Question and Hypothesis 3
children performed relatively better on microstructure than macrostructure. Findings of bi-directionality (i.e. children who show high linguistic form/low narrative content or low linguistic form/high narrative content) give more power to the hypothesis that imbalances occur due to cognitive capacity limitations. As micro- or macrostructure are theoretically just as likely to get ‘bucketed’ due to an increased processing load, this study expects to find evidence of bidirectionality as well.
‘Bucketing’ in language parameters during narration is considered a sign of processing limitations in children with DLD (e.g. Colozzo et al., 2011). Similarly, dysfluencies have been shown to be more present in children with DLD during tasks with high processing demands (e.g. Wagner et al., 2000). Therefore, it is expected that children who show more obvious dissociations or ‘bucketing-behaviour’, who thus are hypothesised to have more severe capacity limitations, also produce more dysfluencies.
HYPOTHESIS 3 RESEARCH QUESTION 3
Is there a relation between instances of communicative breakdowns and imbalances in macro- and microstructural investment?
18
Chapter 4
4. Methods 4.1 Participants
Sixty-seven Dutch children diagnosed with DLD were recruited for this study, in the age range of 9 to 12-years-old (M(SD) = 10.739(0.14)), of which 52 were boys and 15 were girls (male-to-female ratio = 3.46). Note that this sex imbalance is typical of samples of children with DLD (Leonard, 1998). The children were part of a bigger research project investigating efficacy of class-based intervention materials, designed to improve complex
morphosyntactical abilities in older children with DLD. The children had previously been diagnosed with DLD by a speech-language pathologist or clinical linguist, and were receiving therapy for their language difficulties. Only the children of whom the parents handed-in signed informed consent forms were included in this study. The pupils were enrolled in classes five through eight: that is to say, the last half of Dutch primary school education. All of the participants were pupils at special needs education schools, specifically Cluster 2 schools. In the Netherlands, an indication of Cluster 2 service signposts that the clients experience deafness-, hard-of-hearing and/or speech- or language problems.
The inclusionary criterium was a diagnosis of DLD, as diagnosed with diagnostic tests and observations by the particular school’s speech-language therapist or clinical linguist. A diagnosis of DLD is given if a child performs at least 1.25 standard deviations below the norm score on several clinical and diagnostic tests. Exclusionary criteria besides total deafness or significantly impaired vision were few; the main difficulty of the children with DLD who attend Cluster 2 schools is their language development. In case their language developmental issues might be linked to, or are overpowered by behavioural or psychological problems, or mental or motoric disablement, they will typically get referred to Cluster 3 and 4 schools instead (Smeeds, 2007). As such, children with (mild) attention deficit hyperactivity disorder, autism spectrum disorder, developmental dyslexia, hearing problems, etc., or who wore a cochlear implant, were also included. Bilingual children with DLD were also included; studies like Kupersmitt and Armon-Lotem (2019) and Tsimpli et al. (2016) show no differences in relative narrative complexity and form between bilingual children with DLD and monolingual children with DLD, therefore there is no ground for exclusion.
19
4.2 Materials
In this retrospective study, narratives were elicited with the help of Mayer’s (1969) wordless booklet Frog, Where Are You?. This is a psycholinguistic tool that can be used to elicit narratives (Scheper & Blankenstijn, 2014). The Frog Story counts twenty-four pages of pictures without text. The theme in the book is the frantic search of a boy and his dog to find his escaped pet frog. The eliciting procedure takes place in spoken Dutch. The book is unsuitable for children with visual impairments only, due to its graphic nature (Hoiting, 2009). As Berman and Slobin (1994) explain, this book is suitable for use with children aged four-years-and-up, adolescents and even adults; as these authors argue, the story is
representational of stories that are typically told in first and second world countries – that is, the story has a chronological order, its style is comic-like, it has a happy ending, and it has a fairy tale-like atmosphere.
This booklet wasadditionally selected as stimulus for the following reasons. Firstly, story generation tasks like the Frog Story have been proven to put higher processing demands on children (e.g. Pearce, James & McCormack, 2010; Duinmeijer et al., 2012), which
increases chances of finding ‘trade-offs’ in narrative parameter performance due to cognitive capacity limitations. Secondly, while the Frog Story is an English booklet, due to its wordless contents it has been shown to be helpful in tapping narrative abilities of both TD children from diverse language backgrounds (Berman & Slobin, 1994), and various developmentally disordered populations (e.g. Reilly et al., 2004; Norbury & Bishop, 2003). Thirdly, many previous studies that employed a variety of the same metrics as this study and investigated children with DLD, have used this booklet (e.g. Manhardt & Rescorla, 2002;Norbury & Bishop, 2003; Boudreau & Hedberg, 1999); this allows for a better comparison of results.
As part of a bigger research project, children were tested with four other written, explicit, structured tasks before performing this final task, the Frog Story narration. Results from the written tasks are not used in this study. Respectively, the written tasks tested: 1) the children’s passive knowledge of linking words like ‘but’ (maar), ‘because’ (omdat), and ‘for’ (want); 2) the children’s ability to use certain linking words and verbs to create sentences relating to visual stimuli; 3) the children’s ability to explicitly name word classes (e.g. noun, verb, etc.) related to highlighted words in written sentences; 4) a type of grammaticality-judgment task, where children had to underline one faulty word per sentence in a series of either correct or incorrect sentences, and explain to the researcher why they believed that word to not fit in that sentence.
20
4.3 Procedure
4.3.1 Task Administration
For eliciting a narrative, the eliciting-procedure as set out by Scheper and Blankenstijn (2014) was used – that is, information about the contents was kept minimal, to reduce priming
effects. Narratives were consistently coaxed by the researcher in a following fashion: ‘I know this is a book about a frog, a dog and a boy, but I don’t know the story yet. Can you tell me? There are only pictures inside, no text. It is best if you first have a little look at the book yourself, make a little story in your head, and then tell that story to me. But you can keep the book open at all times. Okay?’. Once the child had finished looking at all the pictures, they were allowed to tell the story to the researcher. There were no children who refused to tell the story, or who did not sufficiently understand these instructions. Storytelling and inspecting the booklet took around ten to twenty minutes for each child.
The solo researcher sat diagonally or across from the child, and made sure that they themselves could not see the pictures clearly, in order to prevent the child from pointing to pictures without verbal explanation. The three characters were introduced to the child up-front before reading, during the eliciting procedure. Importantly, the researcher emphasised that they did not yet know the plot or story arc well – children will be more complacent in their storytelling effort if they believe the listening party is already anticipating, or aware of, certain events or episodes (Berman & Slobin, 1994). During narration the researcher
encouraged the child with as little verbal feedback as possible. In particular, no verbal feedback was given that might have encouraged a certain word choice or grammatical tense; oral encouragement existed as far as possible only of words like hm, ohh, or yes.
Furthermore, the children were filmed while telling the narrative, and were made aware of the fact they were being recorded by video; none of the children objected to this, or seemed nervous as a result, so this is not expected to have affected narrative performance. The narratives were collected in quiet areas of the schools, in order to reduce drops in attention and for the recording equipment to capture all of the child’s speech.
4.3.2 Story Transcription
The children’s narratives were transcribed verbatim from the video recording, while adhering to the transcription procedures of the Human Analysis of Transcripts (CHAT) of the Child Language Data Exchange System (CHILDES; MacWhinney, 2000), and according to guidelines as set out by STAP (Van den Dungen & Verbeek, 1999). The story was divided
21
into T-units, where a T-unit is comprised of a main clause and its dependent constituents, which has been shown to be particularly useful for transcribing oral production (e.g. Larsen-Freeman, 1983).
4.3 Data Analysis 4.3.1 Macrostructure
STORY GRAMMAR (SG): The pictures in Frog, Where Are You? depict a story that can be told through a comprehensive structure (Mayer, 1969). This comprehensive structure can be told through setting components, initiation components, components of planning, episodes, and outcome components (Trabasso & Rodkin, 1994). In the Frog Story, these elements can be expressed by including nineteen story grammar episodes. Therefore, children can receive at most 19 points for this macrostructural index, where each of the 19 categories was assigned a score of 1 to indicate the presence of that category, or a 0 for absence. Then a proportion score is calculated by dividing the total number of mentioned episodes by nineteen.
Though in the original study Trabasso and Rodkin (1994) identified seventeen SG components, nineteen points are used as a metric in this study, following Roelofs (1998) and Blankenstijn and Scheper’s (2003) adaptation of the framework to also include an extra internal response (number 19) and an additional initiating event. Appendix A gives more information about what story-arc boxes children needed to tick exactly, in order to classify parts of their narrative as an SG element.
CAUSAL RELATIONS (CR): In this study, Fichman et al.’s (2017) method of analysing narratives is used to measure causal relation performance, by identifying Enabling, Physical, Motivational, and Psychological causal relations as based on Trabasso and Sperry’s (1985) indices of causal chains and causal connections. As Fichman et al. (2017:80) state, “[c]ausal relations reflect the connections between basic SG categories or between SG
Causal Relation Definition Example
Enabling Connects an attempt to a goal or to
another attempt in the following episode The dog tried to catch the bees (attempt). The boy saw it and decided to run away (goal). Physical Connects attempts and outcomes within
episodes The dog tried to swim away (attempt). But he got stuck on the boy’s head (outcome).
Motivational Connects goals and attempts within
episodes He said shhh (attempt). Because he wanted to hear the frog (goal). Psychological Establishes relations between internal
responses (IR) and attempts or outcomes. The boy was happy (IR). Because they found the frog (outcome). Table 4: Causal Relations
22
categories and characters’ internal responses”. See Table 4 (p.21) and Figure 1 (p.8) for further definitions and examples. The densities for causal relation use are obtained by dividing each story’s total number of causal relations by the total number of T-units.
EVALUATION (EVA): Previous studies that measure evaluation in narration, like Manhart and Rescorla (2002), Norbury and Bishop (2003), and Torng and Sah (2020), have identified the following evaluative devices: a) frames of mind, like happy or sad, referring to an affective or cognitive stage; b) evaluative words, like funny boy or ugly frog; c) negative comments, like ‘the boy almost caught the frog’; d) hedges, which indicate uncertainty, like ‘the boy was probably angry at the dog’; e) character speech, including direct (‘he said shhh’) and indirect speech (‘the boy told the dog to be quiet’); f) causal conjunctions, like because and for; g) attention getters, like sound effects or exclamations.
As Reilly et al. (2004) suggest, proportion scores can be used to report the use of evaluation. These can be obtained by dividing the total number of the previously mentioned evaluative devices by the number of T-units in the narrative.
4.3.2 Microstructure
PRODUCTIVITY, COMPLEXITY: Eleven variables will be analysed, namely the mean length of the T-unit in words (MLT), the mean length of the five longest utterances (MLT5), the lexical diversity (NDW), the total number of words (TNW), and the length of the story in T-units (LENGTH) as productivity measures, and the proportion of complex T-units
(COMPLEX), the proportionate amount of coordination and subordination (COORD; SUBORD), the proportionate amount of omitted tense (OMIT), the proportionate amount of the use of simple past (PAST), and the proportionate amount of ungrammatical T-units (UNGRAM) as complexity measures. Proportion scores are all based on the total score for each variable divided by the number of T-units.
COHESION (COH): Liles’s (1985) modification and application of Halliday and Hasan’s (1976) cohesive tie-framework will be used in this study (see Table 5). Each cohesive tie will be identified and categorized by its type. The total number of cohesive ties
Cohesive Tie Example
Reference The boy looks in the pot. HE sees it is empty. Conjunction The boy looked in the pot. BUT it was empty. Substitution The dog chased the bees. But he didn’t catch ONE Lexical They fell into the pond. THERE was a large splash. Ellipsis The boy is sleeping. AND THE DOG TOO.
Table 5: Cohesive Ties
23
will get divided by the number of T-units, which gives the average proportion of cohesive ties per T-unit per child.
4.3.3 Communicative Breakdowns
Pauses, fillers, and repairs will be counted as communicative breakdowns in this study. Pauses are defined as a mid-sentence ‘needless’ stall. Fillers are defined as interruptions in the speech flow that do not result in syntactic changes or the addition of new information; as such, semantically empty sounds like urn, err, and ehh are considered fillers, but not
semantic words like you know. Finally, repairs are defined as an attempt to correct an utterance. For example, they <fall> fell into the pond is considered a repair. Each type of communicative breakdown is divided by the number of T-units in order to calculate the breakdown density.
Important to note is that, when adhering to CHILDES transcription guidelines (MacWhinney, 2000), these sorts of dysfluencies are embedded in brackets in the transcription, and are thus not included in microstructural measures of productivity.
24
Chapter 5
5. Results
5.1 Analysis of Output Data
The output data from this study is a purely numerical outcome. Therefore, correlation matrices using Pearson correlation are used for in-depth analysis of associations and
dissociations between different macrostructural, microstructural and dysfluencies measures, and for calculating the correlation between combined macro- and microstructural measures. To combine data, feature scaling (i.e. scaling the variable to have a values between 0 and 1) is first carried out on all variables to normalise the data distribution. This is done by making use of the following formula: 𝑥 𝑛𝑒𝑤 =
, where x is a specific data point of the
variable, x min is the variable’s minimum score, x max the variable’s maximum score, and x new is the normalised data point. Colozzo et al.’s (2011) Relative-Strength-of-Form Index (RSF-TNL) is additionally used in order to calculate how many children show ‘bucketing’ behaviour, and if imbalances occur bidirectionally. Colozzo et al.’s RSF formula calculates the relative strength of form (i.e. microstructure), compared to the strength of content, where 𝑅𝑆𝐹 = . Finally, RStudio will be employed for the statistical analysis (R Core Team, 2013).
The specific research questions are thus as follows:
Using Pearson correlation, which indices will be correlated either significantly positively or negatively with each other (significance: p <0.05*, p <0.01**)? Are grand (combined) macro- and microstructural variables correlated?
Using Colozzo et al.’s (2011) RSF-TNL scoring system, how many children will show dissociations between form and content? Is there a bidirectional result? Using null models, how does this sample of children perform on all indices
respectively? What can we learn from this? 5.2 Macrostructure
Table 6 gives an overview of the macrostructural performance by the whole group of 9 to 12-year-olds (n = 67). Despite the fact that no control group was present, it can be tentatively stated that performance on the Story Grammar index is low, with children mentioning only
25
55% of all possible episodes (M = 0.55; SD = 0.02).1 Compared with the Dutch nine-year-old
children as tested with the Frog Story by Roelofs (1998), Duinmeijer (2010), and Duinmeijer, de Jong and Scheper (2012), who mentioned on average 81% of all possible episodes, the children with DLD in the current sample perform significantly worse (Fisher's Exact Test: odds ratio = 0.289; 95% CI = 0.143…0.565; p = 0.0001). Moreover, as becomes clear from the columns Minimum and Maximum, there exist huge differences in macrostructural performance in this singular group, with the lowest performer mentioning only 11% of all possible episodes, while the maximum performer mentions 95% of all possible episodes. The same applies for the indices of mentioning of causal relations and the use of evaluation (see Table 6). Plus, see Appendix B for the detailed results on evaluative devices; evaluative conjunction was used the most in proportion to the narrative’s total T-units, with thereafter character speech and frames of minds. Negative words were used the least.
5.3 Microstructure
Table 7 gives an overview of the children’s performance on microstructural measures. Again, from the microstructural indexes it also becomes clear that there is some dissonance in the group’s performance, with some children having an MLT-W of 3.4 and others of 10.4, and some children having a lexical diversity index of 34 different words, and others of 248. Similarly, some children do not omit tense at all, while others omit it 86.8% of the time, and some produce zero complex T-units, while for others 27% of all T-units consisted of more than one finite verb. Also, see Appendix C for the detailed results on cohesive devices; reference was used the most in proportion to the narrative’s total T-unit count, and ellipsis and substitution the least.
1 Due to the COVID-19 outbreak in the Netherlands in March 2020, when data collection would have taken
place, safely testing a control group using the same methodology as used with the children with DLD proved impossible. To guard against differences in conditions between groups, it was decided not to test TD children.
Macrostructural Measures Mean(SD) 1st Qu. 3rd Qu. Min. Max.
Story Grammar
(Proportion of all episodes)
0.55(0.02) 0.4210 0.6316 0.1053 0.9474 Causal Relations (Proportion of T-units) 0.24(0.01) 0.1600 0.3196 0.0256 0.4615 Evaluation (Proportion of T-units) 0.23(0.02) 0.1446 0.2718 0.0500 0.7692 Table 6: Macrostructural Measures
26
5.4 Communicative Breakdowns
Table 8 gives an overview of the average instance of communicative breakdowns in the children’s narratives. Again, children differed wildly on total occurrences of communicative breakdowns, with some children not producing repairs in their speech at all, and others producing word repairs in 151% of their T-units. Pauses occurred the least in narratives.
5.5 Dissociation
One explanation for the observed large variances in investment of this sample on one variable, is connected to the hypothesis that children with DLD have a tendency to perform well on either macro- or microstructure due to processing limitations. This hypothesis could
Dysfluencies Measures Mean(SD) 1st Qu. 3rd Qu. Min. Max.
Pauses (Proportion of T-units) 0.150(0.017) 0.04826 0.20781 0 0.78261 Fillers (Proportion of T-units) 0.248(0.028) 0.09091 0.33013 0 1.18868 Repairs (Proportion of T-units) 0.481(0.039) 0.2336 0.6476 0 1.5111 Table 8: Dysfluencies/Communicative Breakdown Measures
Microstructural Measures Mean(SD) 1st Qu. 3rd Qu. Min. Max.
Mean Length of T-unit (In words)
6.68(0.1360) 6.095 7.240 3.42 10.40 Lexical Diversity
(In words)
101.7(4.360) 75.5 120 34.0 248.0 Total Number of Words
(In words) 285.6(14.88) 205.5 351.0 82.0 766.0 Coordination (Proportion of T-units) 0.060(0.009) 0.01245 0.07093 0.00 0.316 Subordination (Proportion of T-units) 0.050(0.007) 0.0 0.08340 0.00 0.222 Story Length (In T-units) 42.358(1.72) 33 49.5 18.0 84.00 Complex T-units (Proportion of T-units) 0.065(0.007) 0.0 0.11 0.00 0.270 MLT-5
(Mean of 5 Longest T-units)
11.28(0.261) 9.9 12.6 6.40 18.40 Omitted Tense
(Proportion of T-units)
0.103(0.015) 0.0385 0.1310 0.00 0.868 Use of Past Tense
(Proportion of T-units) 0.477(0.038) 0.1490 0.7555 0.00 0.966 Ungrammaticality (Proportion of T-units) 0.390(0.022) 0.2440 0.4955 0.09 0.875 Use of Cohesion (Proportion of T-units) 0.567(0.030) 0.44696 0.73750 0.042 1.111 Table 7: Microstructural Measures
27
be invalidated by a significant positive correlation between macro- and microstructure scores. Nevertheless, with all normalised macro- and microstructural measures respectively
combined, results show no significant correlation. Pearson's product-moment correlation shows the estimate of the correlation coefficient between microstructural and macrostructural measures to be 0.0912, with a 95% confidence interval (CI) running from -0.1523 to 0.324 (t[65] = 0.738, p = 0.463).
The reason for a correlation of r = 0.0912 gets clarified when using Colozzo et al.’s (2011) RSF-TNL scoring system. In total, 38 children scored below 0.45 (i.e. high narrative content/low linguistic form), 18 children scored between 0.45 and 0.55 (i.e. form and content performance are balanced), and 11 children scored above 0.55 (i.e. low content/high
form)(see Table 9). The fact that there is a sizable number of children in the balanced group may have caused the correlation coefficient to be rather neutral, rather than negative which would indicate that microstructure reliably goes down when macrostructure goes up, or the other way around. Note that the cut-off values of 0.45 and 0.55 for the RSF-TNL
classification were selected because they are symmetrical and analysis confirmed that nearly homogenous subclasses of absolute differences in micro- and macrostructure scores could be formed between the balanced and dissociated groups when using these cut-off points.
5.6 Specific Correlations
To investigate the hypothesis that individual macrostructural measures are not correlated with comparable microstructural measures, a correlation matrix was designed (e.g. adequate use of conjunctions at microstructural level does not correlate with the ability to construct causal relations at macrostructural level). Figure 2 (p.29) depicts a correlation matrix of all
measures; the bigger the circle, the more significant the correlation. Bluer circles indicate an increasingly positive correlation, while redder circles indicate an increasingly negative one. Please note that the variables OMIT (omitting tense) and UNGRAM (ungrammatical T-units) are ‘negative’, i.e. a low score on them indicates ‘better’ language performance. Table 10 (p.30) shows the full correlation matrix with p-values and correlation coefficients.
RSF Score n = Mean(SD) Min. Max. Low Micro High Macro 38 0.367(0.01) 0.1956 0.4498 Balanced Micro and Macro 18 0.513(0.01) 0.4745 0.5435 High Micro Low Macro 11 0.621(0.02) 0.5650 0.7340 Table 9: Breakdown of RSF Scores
28
In terms of associations, the measures of microstructural productivity are all
positively correlated with each other. For example, lexical diversity (NDW) increases when the mean length of utterance (MLT-W) increases (r = 0.617; 95% CI = 0.443…0.747; p = 2.641*10-8). Measures of microstructural complexity are correlated with each other and with
productivity. For instance, the use of subordination (SUBORD) is understandably correlated with the child’s production of complex T-units (COMPLEX) (r =0.907; 95% CI =
0.852…0.942; p = 2.2*10-16), and the production of ungrammatical T-units (UNGRAM) is
negatively correlated with the use of the past tense (PAST) (r =-0.261; 95% CI = -0.472…-0.022; p = 0.033). That is, when the proportionate production of ungrammatical T-units increases, the proportionate use of the past tense goes down, and the same is true the other way around. Positive macrostructural correlations exist between the use of Story Grammar and causal relations (r = 0.233; 95% CI = -0.008…0.448; p = 0.058), and the use of causal relations with the use of evaluation (r = 0.259; 95% CI = 0.020…0.470; p = 0.034). This confirms that performance is correlated within macro- and microstructural indices.
However, when considering correlations between macro- and microstructural measures, the use of causal relations is very reliably correlated in a negative trend with subordination (r = 0.30), complex Tunits (r = 0.34), story length (r = 0.30), MLTW (r = -0.33), lexical diversity (r = -0.32), total number of words (r = -0.31), and MLT-5 (r = -0.32). Concretely, this means that all these microstructural measures go down when the production of causal relations increases, and the other way around.
29
Correlations between macro- and microstructure and communicative breakdowns are also analysed. For macrostructure, almost all measures of dysfluency are not correlated to performance on narrative content. In fact, dysfluencies are much more correlated with measures of form. For example, cohesion and mid-T-unit pauses are nearly-significantly negatively linked with each other (r = -0.228; 95% CI = -0.444…0.013; p = 0.0636); use of cohesive devices goes down when use of pauses goes up, and the other way around. The use of fillers is positively correlated with omitting tense (r = 0.309; 95% CI = -0.074…0.511; p = 0.0109); more fillers means more T-units with omitted tense. Furthermore, the use of repairs is very significantly linked in a positive manner with all of the microstructural indices – except for coordination, subordination and ungrammaticality (see table 7).
The relation between communicative breakdowns and the sample’s RSF-TNL score was also calculated. In Table 11 it becomes clear that instances of pauses and fillers do not seem to get higher in the groups experiencing dissociations. However, the instances of repairs do increase for the low macrostructure/high microstructure group compared to the balanced and low RSF group. This gets confirmed by a correlation test between the respective dysfluencies and the RSF score: pauses and fillers are non-significantly correlated with the RSF score (r = -0.189 and r =0.02) – which means that a change in RSF score cannot predict changes in the use of pauses or fillers. Butm repairs are significantly positively correlated with the RSF score (r = 0.296; 95% CI = 0.06…0.501; p = 0.015) – which means that when the RSF score increases (i.e. showing a relative dissociation with low macrostructure and high microstructure), the use of repairs also increases.
Lastly, the balanced group does produce more fillers than both other groups, more pauses than the children with high microstructure and low macrostructure results, and more repairs than the children with low microstructure and high macrostructure results.
Dysfluencies Low Micro/High Macro Balanced Low Macro/High Micro Pauses (div. T-units) 0.21029 0.18175 0.14120 Fillers (div. T-units) 0.18658 0.26860 0.18750 Repairs (div. T-units) 0.28128 0.31220 0.64850
Table 11: Dysfluencies in the Different Groups
30
31
Chapter 6
6. Discussion
6.1 General Discussion
The goal of this study was to consolidate, and re-investigate with updated methods, a handful of previous findings that indicate a (sometimes bidirectional) dissociation in narrative micro- and macrostructural performance of children with DLD, which supports the theory of a processing and/or executive functions deficit at play in DLD, and also to link this with instances of communicative breakdowns, which could lend further support for the processing theory. To this end, the following questions were asked:
Is there a relation between micro- and macrostructure in the narratives of children with DLD aged 9;0 to 12;0, while employing all narrative variables which have proven themselves to be sensitive to the presence of disordered language?
Do potential imbalances between macro- and microstructure occur bi-directionally? Is there a relation between instances of communicative breakdowns and imbalances in
macro- and microstructural investment?
Results obtained with Colozzo et al.’s (2011) RSF-TNL scoring system indicate that a sizable number of children (49 out of 67) experienced dissociations. Furthermore, a Pearson's product-moment correlation shows the estimate of the correlation coefficient between
normalised microstructural and macrostructural measures in the current study’s sample to be 0.0912, with a 95% confidence interval running from -0.1523 to 0.324 (p = 0.463). This indicates that increased performance on narrative content does not predict increased
performance on linguistic form, and the other way around, for children with DLD aged 9;0 to 12;0 – however, due to the null-result finding this cannot be stated conclusively.
Additionally, when considering correlations between individual macro- and microstructural measures, the use of causal relations was very reliably correlated in a negative trend with subordination (r = 0.30), complex Tunits (r = 0.34), story length (r = 0.30), MLTW (r = -0.33), lexical diversity (r = -0.32), total number of words (r = -0.31), and MLT-5 (r = -0.32), which concretely means that all these microstructural measures go down when the production of causal relations increases, and the other way around. These findings are in line with
32
(2011), Gillam and Johnston (1992), and Lam, Scheper and Rispens (2015), and confirms Hypothesis 1 (p.16).
What does this consolidated finding tell us about children with DLD and narration? For this, it is vital to consider either end of the narrative scale. That is, the results that show a low RSF-TNL score (high content, low form) are consistent with findings of previous
researchers like Gillam and Johnston (1992) -- it has long since been known that children with DLD tend to have impaired linguistic skills, especially concerning lexical and
syntactical skills (e.g. Leonard, 2014). Since such skills are considered part of a narrative’s microstructure, this may lead children with DLD to perform relatively better on
macrostructure than microstructure. So, the fact that 38 children in the current study scored in this category may not strictly indicate a processing deficit. Yet, Colozzo et al. (2011:1621) were the first to find evidence of a ‘complimentary pattern’, where children with DLD showed poor content with better linguistic form. Notably, Colozzo et al. found that their sample was distributed evenly between the ‘dissociations subgroups’. On the other hand, Lam, Scheper and Rispens (2015) found that their sample of 16 children with DLD mostly focussed on macrostructural performance, though there were some children who performed better on microstructure than macrostructure. Either way, ‘complimentary pattern’ findings provide more robust evidence of a true inclination to ‘bucket’ either of the language
parameters due to cognitive capacity limitations. Therefore, this study also looked at the distribution of children on the RSF-TNL scale. Though fewer children in this sample showed the ‘low content/high form’ pattern (n = 11) than the ‘high content/low form’ pattern (n = 38), quite a large group still had RSF scores of >0.55, >0.6 and even >0.7. It can be
concluded that children with DLD (aged 9 to 12) have a tendency to ‘trade-off’ performance on either microstructure or macrostructure, though microstructure does appear more
vulnerable to ‘bucketing’, perhaps due to the fact that lexical, morphological and syntactic skills are significantly affected in children with DLD (e.g. Leonard, 2014). This confirms Hypothesis 2 (p.16/17).
The subgroup with a low RSF-TNL score generally produced stories that had a well-developed plot line, unified by causal relations, and an interesting story for the listener by making use of -- among others -- sound effects, character speech, and an individual appraisal of the story. However, producing a more developed, causally connected plot also requires more complex syntax in the form of, for example, microstructural coordination and subordination. Results showed that 38 children failed to juggle both of these narrative demands, and had a disbalance in their content and form investment. On the other side of the
