Connectedness in Autistic Adolescents’ Narratives and its Relation to Non-Verbal Cognition
Harriet Reynolds 12102245 26th June 2020 RMA Linguistics Thesis Supervised by Jeannette Schaeffer
ABSTRACT
Previous studies have found that autistic children and adolescents connect events in a narrative to a lesser degree than their typically developing peers (Diehl, Bennetto & Young, 2006; Kelley, Paul, Fein & Naigles, 2006; Losh & Capps, 2003; Peristeri, Andreou & Tsimpli, 2017). The cause of this phenomenon is unknown, however it may be related to non-verbal cognition, including attention shifting between local and global processing, working memory and theory of mind. The present study investigated the connectedness of 18 autistic adolescents’ narratives and measures of the abovementioned non-verbal cognitive abilities. No significant associations were found. A qualitative analysis of the narratives suggests that, although the autistic
participants used the story's gist (the chain of events that causally connect the beginning of the story to the end) to structure their narratives, they demonstrate difficulties in connecting and interpreting events, as well as categorising objects and characters. It is suggested that the Reduced Generalisation Hypothesis (Plaisted, 2001) could explain these observations and warrants further study in relation to narrative structure.
1. INTRODUCTION
An important part of narration is connecting individual events into a causal chain, the story’s ‘gist’, which leads the listener from the beginning of the story to the end (Trabasso & Sperry, 1985). For example, the present narrative task, the Pear Film (Chafe, 1980), starts with a man picking pears, and ends with his (incorrect) suspicion that three boys stole one of his baskets of pears. Thus, the gist of this story is all the events in between that lead him to this incorrect assumption. If a boy had not stolen his pears, and if this boy had not fallen off his bike and been helped by three other boys, and if these boys had not received a handful of pears for their help, and if these boys had not walked past the farmer eating these pears, then the farmer would
have never come to this conclusion. Therefore, all of these events are causally connected and form part of the gist.
Linguistic-pragmatic difficulties are well documented in Autism Spectrum Disorder (ASD) (see Eigsti, de Marchena, Shuh & Kelley, 2011 for a review), and multiple studies suggest that autistic children and adolescents produce less connected narratives than their typically
developing (TD) peers (Diehl, Bennetto & Young, 2006; Kelley, Paul, Fein & Naigles, 2006; Losh & Capps, 2003; Peristeri, Andreou & Tsimpli, 2017). For example, Diehl and colleagues (2006) suggest that autistic children may have the tendency to narrate events out of order and to omit background context.
Several psychological theories have been formulated to explain the autistic cognitive profile, and the three most influential will be investigated in this study. The Weak Central
Coherence Hypothesis (e.g. Frith, 1989; Happé & Frith, 2006) claims that autistic individuals are biased towards local processing. Meanwhile, difficulties with various aspects of executive functioning, including attention shifting, inhibition and working memory, are presumed to explain at least part of the autistic cognitive profile under the Executive Dysfunction Hypothesis (e.g. Pennington & Ozonoff, 1996). Finally, it has been proposed that autistic individuals have Weak Theory of Mind (e.g. Baron-Cohen, Leslie, & Frith, 1985), and therefore struggle to ascribe beliefs and knowledge to others. However, the link between these psychological theories and the pragmatic difficulties observed in autistic individuals remains unclear.
The present study aims to investigate the pragmatic macro-structure of autistic teenager’s narratives by measuring and discussing how they connect events in a story. The study further relates this connection measure to performance in non-verbal attention
shifting/inhibition, as well as measures of non-verbal working memory and non-verbal theory of mind. Thus, the following overarching research question is posed: How do autistic
adolescents connect events in a narrative, and what is the relationship between narrative event-connectedness and non-verbal cognitive mechanisms, namely, attention
shifting/inhibition between local and global processing, working memory and theory of mind? The results will contribute towards our understanding of how autistic adolescents structure narratives, and therefore provide insights into their pragmatic style. Furthermore, by assessing potential links to cognitive mechanisms, leading psychological theories can be evaluated.
2. BACKGROUND
2.1. Narratives
Narratives have been shown to be a successful tool for measuring linguistic and communicative skills in children (Botting, 2002) and they are frequently used for linguistic analyses in clinical populations due to their nature of bringing together multiple linguistic and cognitive factors (Duinmeijer, de Jong & Scheper, 2012). Successfully narrating a story involves describing a series of events in relation to causal principles and goals (Trabasso & Rodkin, 1994), making inferences and adapting to the listener’s needs (Norbury & Bishop, 2003).
A discourse can be split into two semantic levels, referred to as microstructure and macrostructure. The microstructure level involves local processing of individual propositions, while the global whole of a discourse is evaluated on the macrostructure level (Kintsch & van Dijk, 1978). In general, when reconstructing a discourse from memory, people start with the most general schema at the macrostructure level, i.e. the goal. With this goal in mind, they distil all the necessary information into a gist, and then reconstruct the details (Kintsch & van Dijk, 1978). Thus, retelling a story is a global-to-local process. This macrostructure approach to discourse presumably develops gradually as typically developing children age. By investigating the type of false memories pre-schoolers (3-4 years) and fourth-graders (9-10 years) produce when retelling stories, it has been shown that, between these ages, children shift from a
preference for verbatim retelling towards a more gist-orientated approach (Brainerd & Reyna, 1998).
2.2. Language and Narrative Structure in ASD
Language abilities are highly heterogeneous in the autistic population. An estimated 25% of autistic individuals remain non- or minimally-verbal and thus never acquire language to a functional level (Tager-Flusberg, Paul & Lord, 2005). Of those who do acquire functional language, some, but not all, exhibit morphosyntactic impairments similar to those observed in children with Specific Language Impairment (SLI) (e.g. Modyanova, Perovic & Wexler, 2017; Roberts, Rice & Tager-Flusberg, 2004). Pragmatic deficits, on the other hand, are observed across the spectrum and are considered a central feature of ASD (e.g. de Marchena & Eigsti, 2016; Eigsti et al., 2011).
Given the pragmatic difficulties observed in individuals with autism, the question arises as to whether they narrate stories in the same way as their typically developing peers, and if not, how this may differ. In this regard, the study carried out by Diehl, Bennetto and Young (2006) is highly relevant. These authors compared the narrative structure of 17
English-speaking children with ASD, aged 6 to 14, to 17 TD controls matched on age, gender, receptive and expressive language abilities and verbal IQ. They administered the well-established Frog Story task (Meyer, 1969), where the participants were given a few minutes to look at the wordless picture book, and then were later asked to retell the story without the visual prompts. Using the methodology outlined by Trabasso and Sperry (1985), the original story was mapped out and the story’s gist was identified. Then, for each narrative, the number of events mentioned from the story’s gist were counted. Additionally, a list of story details, in terms of characters and objects, was compiled, and the number of these details mentioned was counted per narrative. Comparing these two counts showed that both TD and ASD participants were more likely to
remember gist events than story details, implying that both groups were sensitive to the gist and used it to aid story recall. The researchers further mapped out the connections in each
individual narrative, and divided the number of connections by a measure of story length. The ASD participants scored lower on this measure than their TD peers, meaning that their narratives were less connected. Thus, taken together, these analyses indicate that the ASD group used the gist to structure their narratives, but struggled to make clear connections
between events, which, the authors suggest, may be seen as a tendency to mention events out of order and a failure to provide relevant context.
Other studies have found a similar trend, in that autistic children’s narratives are less causally connected than those of their TD peers. For example, Peristeri, Andreou and Tsimpli (2017) recruited 30 monolingual Greek-speaking children with ASD (age range: 6;2-12;4) and 15 age-matched TD children for a story retelling task (the Giraffe/Elephant story from the ENNI; Schneider & Hayward, 2005). In this task, the participant first listened to the complete story and then retold it with the aid of pictures. The story consisted of three episodes, each with a clear goal, attempt and outcome, from which a story complexity score was computed, alongside various other measures. The results show that the ASD participants scored significantly lower on story complexity than their TD peers, once again suggesting difficulties with explicitly connecting key story elements into a narrative, even when that information has been verbally given in advance.
In another study, as part of a comprehensive battery of linguistic tasks that included the Frog Story narrative, Kelley, Paul, Fein and Naigles (2006) found that English-speaking
participants with a history of ASD (n=14, aged 5;6-9;1) produced significantly fewer causal explanations for story events than their age- and sex-matched TD peers. Furthermore, the ASD participants were more likely to include misinterpreted information in their narratives than the TD participants.
2.3. Theory of Mind in ASD
Several psychological theories attempt to explain the observed autistic profile, including the weak Theory of Mind account (see e.g. Baron-Cohen, Leslie, & Frith, 1985). Theory of mind is the ability to ascribe beliefs and knowledge, which differ from our own, to others (Baron-Cohen, Leslie, & Frith, 1985; Premack, Woodruff & Guy, 1978). Theory of mind abilities are often measured using ‘false belief’ tasks, initially introduced by Wimmer and Perner (1983), where, to correctly pass the test, the participant must recognise that others will act on their false beliefs rather than real-world truths. Typically developing children learn to resolve such situations between 3 and 4 years of age (Wellman, Cross and Watson, 2001), while children with ASD (n=12, mean age = 8;2) were found to score significantly lower than TD controls (n=15, mean age = 4;8) on a non-verbal false belief task (Colle, Baron-Cohen & Hill, 2007). However, some studies report ASD participants having TD-like performance on false belief tasks (e.g. Schaeffer et al., 2018).
Presuming that weak theory of mind is indeed inherent to ASD, this may lead to
particular difficulties in narrating event pertaining to others’ thoughts, feelings or beliefs. An early study by Baron-Cohen, Leslie and Frith (1986) suggests that this may be the case. The authors recruited 21 autistic participants (aged 6;1-16;11), an age-matched group of 15
children/adolescents with Down’s Syndrome (6;3-17;0) and a younger (unmatched) TD group (3;6-5;11), and asked them to narrate simple sequences of events from pictures. The picture sequences were split into three categories – causal-mechanical, descriptive-behavioural and psychological-intentional. The autistic participants had superior performance in the causal-mechanical category than both other groups, equal performance to the TD group and better performance than the Down’s Syndrome group for descriptive-behavioural sequences, and, crucially, worse performance than both other groups when it came to psychological-intentional events. The authors conclude from this that autistic children/adolescents struggle
disproportionally with narrating events involving intentionality and therefore that weak theory of mind is implicated.
However, as the following study illustrates, the connection between theory of mind measures and narrative structure is not always so clear-cut. Losh and Capps (2003) elicited personal and storybook (the Frog Story; Meyer, 1969) narratives from 28 children with high-functioning ASD, as well as 22 TD children, all aged 8 to 14. The narratives were coded for a number of measures, including one of causality. The results showed that, despite many similarities between the two groups' narratives, the ASD group produced fewer instances of causal language (e.g. because, try to), in both the personal and storybook tasks. As the authors point out, however, it is not clear from these results whether the reduced use of causal
expressions is due to a lack of causal understanding itself, or alternatively a lack of realisation that such causal relationships should be communicated. Along this vein, the researchers further examined the correlation between the narrative measures and scores on a verbal theory of mind task (Strange Stories; Happé, 1994), as well as measures of emotional understanding (a task of defining emotions, plus the Berkeley Empathy Measure; Capps, Ebling & Rasco, 1997).
Emotional understanding scores correlated with the causality measure, as well as descriptions of mental states, while theory of mind was not correlated with any narrative measure, despite the ASD group scoring lower than their TD peers in both theory of mind and emotional
understanding. The authors suggest that the lack of an association between theory of mind and narrative performance may be due to ASD being a spectrum, to which their participants were towards the higher end. They suggest that theory of mind may only be relevant to the narrative abilities of autistic individuals with intellectual impairment, while better emotional knowledge may instead aid high-functioning autistic individuals in their narrative production.
2.4. Executive Functioning in ASD: Working Memory, Attention Shifting and Inhibition
Attention shifting and its counterpart inhibition, plus working memory are components of executive functioning – the set of cognitive mechanisms involved in behavioural control. Under the Executive Dysfunction Hypothesis, it is argued that the traits observed in autism may be due to deficits in executive functioning mechanisms (e.g. Pennington & Ozonoff, 1996).
According to Baddeley (e.g. 2003), the working memory system consists of three parts: a visuospatial sketchpad, a phonological loop and an executive control. As the names suggest, the phonological loop is involved with sounds, and is therefore implicated in verbal processing, while the visuospatial sketchpad is non-verbal. The central executive, common to both the phonological loop and the visuospatial sketchpad, is where attention is managed. Working memory abilities in autism have been studied using a variety of tasks and the results are often mixed, however, a meta-analysis carried out by Wang and colleagues (2017) of 819 ASD participants and 875 neurotypical controls concluded that a general impairment in working memory is present in ASD, and that spatial (i.e. non-verbal) working memory was more impaired than verbal working memory.
Working memory has been shown to affect the structure of autistic children’s narratives in terms of reference (Kuijper, Hartman & Hendriks, 2015). In a novel narrative study with 46 ASD, 37 ADHD and 38 TD children, all aged 6-12 years, Kuijper and colleagues (2015) found that working memory, as measured by a minimally verbal n-back task, was correlated with NP use at the referent reintroduction stage. From this, the authors concluded that working memory is needed to keep track of referents in a discourse, and that observed difficulties with reference in ASD may be the effect of weak working memory.
Working memory may not be the only component of executive functioning to be
implicated in the ASD profile, as there is some evidence that attention switching is additionally impaired. For example, in a visual attention switching task, autistic participants (n=8, mean age
= 13;11) were found to be slower and less accurate in switching than TD controls
(chronological-age-matched: n=8, mean age = 13;10; mental-age-matched: n=10, mean age = 8;7; Courchesne et al., 1994). Inhibition, on the other hand, was found to be TD-like in a negative priming experiment that tested spatial inhibition with 23 autistic participants (aged 7-33) and 23 neurotypical controls (aged 10-35) (Brian, Tipper, Weaver & Bryson, 2003).
Problems with attention shifting in ASD may therefore be linked to weak planning and cognitive flexibility rather than inhibition, unless the individual has comorbid ADHD (Craig et al., 2016).
Of particular interest to the present paper is the study carried out by Peristeri and Tsimpli (2020), who investigated attention shifting/inhibition in combination with global and local levels of processing, and correlated this measure with linguistic referencing expressions. In their task, adapted from the 1977 Navon Task, participants are presented with three blocks of trials – local, global and switch (for a full explanation of this task see §4.3.). Using this task with 15 autistic, 15 TD and 15 Specific Language Impairment (SLI) Greek-speaking children aged 9;1-12;5, the researchers calculated the global and local 'costs' for each group. In this way, the 'global cost' was the difference in accuracy and reaction times between the global trials in the switch block and the non-switch block, while the 'local cost' was calculated in the same way with the local trials. The researchers found that the ASD group had a significantly higher global cost than both the TD and SLI groups, meaning that they had more difficulties switching from local to global trials. On the other hand, in terms of local costs, they performed TD-like. These results suggest that the picture is more complicated than a simple attention switching/inhibition impairment, and that levels of processing are highly relevant. Peristeri and Tsimpli further elicited narratives from these children in a retelling task, and analysed their speech for referencing expressions at three stages (Introduction, Maintenance and Reintroduction). The observed global costs were
correlated with the use of less adequate referencing expressions at the Reintroduction stage and partially (RT measure only) at the Introduction stage, with ASD participants showing a tendency for referent underspecification. The SLI group, on the other hand, had higher local
costs and tended to overspecify referents. These results suggest that autistic children’s
difficulties switching between local and global processing lead to problems selecting appropriate referencing expressions during narration. Presumably this is due to the nature of narratives, where both the global discourse (macrostructure) and local linguistic expressions
(microstructure) must be taken into consideration.
2.5. Central Coherence and Local versus Global Processing in ASD
The Weak Central Coherence Hypothesis, first introduced by Frith (1989) and further developed with Happé (see Happé & Frith, 2006 for a revised version and a review of related studies), posits that ASD is characterised by a default preference for local over global processing. One of the main advantages of this theory, unlike the Theory of Mind (e.g. Baron-Cohen et al., 1985) or Executive Dysfunctioning (e.g. Pennington & Ozonoff, 1996) accounts, is that it takes into consideration autistic individuals’ strengths as well as their weaknesses. Reports of unusual and highly specialised abilities, such as identifying the brand of a vacuum cleaner from the noise it makes (Happé & Frith, 2006), are not uncommon in autism, which deficit-centred approaches fail to explain. Instead, a preference for local processing, usually referred to as weak central coherence, is considered a cognitive style, which can be advantageous in certain situations or tasks in which a detail-focused approach is preferable to a holistic approach.
The weak central coherence style of autism is considered to be one end of a spectrum, spanning across the entire population (Happé & Frith, 2006). Individual differences are therefore expected within the neurotypical population, with some of us defaulting to the weak side,
exhibiting a preference for detail-oriented thinking, and others to the strong side, preferring more holistic, ‘gist’ thinking. Booth and Happé (2010) present empirical findings to support this claim in the form of a sentence completion task that highlights local or global processing tendencies. In this task, participants were required to complete sentences such as “you can go hunting with
a knife and…”. Responses such as “catch a bear” were indicative of strong central coherence, while an answer such as “fork” indicated a focus on local processing, and therefore weak central coherence. Administering this task to 176 TD individuals, aged 8-25 years, revealed individual differences that were not modulated by IQ measures, supporting the claim that central
coherence style varies throughout the population. Additionally, Happé, Briskman and Frith (2001) found evidence that weak central coherence may be heritable. In a study of 22 families with an autistic son, 15 with a dyslexic son and 10 with a TD son, the parents of autistic boys, particularly the fathers, were found to differ significantly from the control parents in their cognitive approach to 4 tasks testing central coherence, exhibiting a detail-focused approach.
Happé and Frith (2006) further emphasise that weak central coherence represents a preference for local processing, and crucially not an inability to process information globally. Therefore, if autistic participants are cued to attend to holistic information, they are able to do so. Indeed, in a study by López, Donnelly, Hadwin and Leekam (2004), autistic participants (n=17, mean age 13;1) attended to faces holistically when cued, but did not use a holistic approach in uncued trials. On the other hand, an age-matched TD comparison group of the same size displayed a holistic approach in both cued and uncued trials. However, contrary to what would be expected from the Weak Central Coherence Hypothesis, the autistic participants did not perform significantly better than their TD peers on trials that targeted parts of faces, where a local-processing bias would be predicted to enhance performance. It should be noted though, as Happé and Frith (2006) point out, that results from facial processing tasks may not be generalisable to visual processing, as two separate mechanisms may have developed independently from one another due to the evolutionary importance of facial processing (see e.g. Suzuki & Cavanagh, 1995).
Further evidence that autistic individuals have weak, as opposed to no central
coherence comes from a pair of studies carried out by Plaisted, Swettenham and Rees (1999) with 17 autistic participants (aged 6;7-16;7) and 17 TD participants (English-speaking, aged
6;1-14;4). The first task used was a version of the Navon task (Navon, 1977), dubbed the ‘divided attention’ task, where participants were required to signal if the letter ‘A’ was present, either locally or globally, or not. In the second task, the ‘selective attention’ task, participants indicated whether an ‘H’ or ‘S’ was present. Trials were either congruent (e.g. big ‘H’ made of small ‘H’s) or incongruent (e.g. big ‘H’ made of small ‘X’s). The researchers found a global bias in the TD group for both tasks, but interestingly found that while the autistic group demonstrated a local bias in the divided attention task, they showed a global bias in the selective attention task. These results may indicate that the autistic individuals voluntarily chose to attend to the local level first in the divided attention task, or alternatively that they had difficulties inhibiting local information. Some of the results may also be explained by a problem with attention switching, or alternatively inhibition, although this would not explain any processing advantages. The exact relationship between processing preference, attention switching and/or inhibition is therefore still unclear and requires further investigation. However, these results are important in showing that autistic individuals can process information on the global level.
Few studies have directly tested the Weak Central Coherence Hypothesis alongside narrative analyses in autism. However, the abovementioned study by Peristeri and colleagues (2017) goes part way towards this, as it was specifically the global switch cost that was
correlated with referencing, implying a local-processing bias. Narrative comprehension and inferencing abilities in autism have also been studied from the perspective of central coherence by Nuske and Bavin (2011) with 14 English-speaking autistic children (aged 4;6-7;11) and 14 TD children (aged 4;2-5;4) matched on receptive vocabulary and a picture-completion task (WPPSI-3; Wechsler 2002). In this study, the children were given a block design task (WPPSI-3; Wechsler 2002)as a measure of central coherence, as well as a comprehension task on spoken paragraphs (CELF-4; Semel, Wiig & Secord, 2003)with main idea and detail questions, and finally, an inferential processing task that required factual, script and propositional inferences to be made (based on Slackman & Nelson, 1984). Local processing was therefore tested by the
detail questions and propositional inferences, while the main idea questions and script
inferences tested global processing. No evidence was found for enhanced local processing in the autistic group, and the autistic participants were able to answer the main idea questions just as well as their typically developing peers (although this may have be due to priming from the title). However, the autistic participants were less able to make script inferences, suggesting a difficulty in creating a global schema from the context to aid understanding. Furthermore, the researchers found that correlations between tasks differed for the TD and autistic groups. For example, for the TD group successfully making propositional inferences was correlated with good performance on the main idea questions, suggesting that making inferences was easier for those with good global processing. On the other hand, in the autistic group, propositional inference scores were correlated with performance on the block design task, in which good local processing is necessary. The authors therefore suggest that, although the two groups scored similarly on some tasks, they may be employing different processing strategies.
3. THE PRESENT STUDY
Based on the aforementioned literature on narrative structure in autism, it is hypothesised that autistic individuals struggle to produce causally connected narratives (Diehl et al., 2006; Kelley et al., 2006; Losh & Capps, 2003; Peristeri et al., 2017). Using a different narrative measure (see §4.2.) than in previous studies, the present study aims to investigate the association between narrative structure (causal connectedness) and non-verbal cognitive measures. As typically developing children are expected to shift to a gist-oriented approach towards story retelling by at least age 10 (Brainerd & Reyna, 1998), testing children below this age may be problematic. Therefore, the present study focuses on the narratives of adolescents.
Following a combination of the Weak Central Coherence Hypothesis (e.g. Frith & Happé, 2006) and the Executive Dysfunctioning Hypothesis (e.g. Pennington & Ozonoff, 1996), it is
hypothesised that autistic individuals have a local processing bias and that shifting attention from local to global stimuli (or, alternatively, inhibiting local stimuli) is challenging, which in turn negatively impacts their ability to organise linguistic information into a global discourse (Peristeri & Tsimpli, 2020). It is further hypothesised that theory of mind, specifically belief reasoning, is weak in autistic individuals (e.g. Baron-Cohen et al., 1985) and that this will lead to difficulties in understanding the characters’ thoughts and feelings and thus drawing inferences. An
understanding of others’ beliefs would also presumably allow the narrator to comprehend the listener’s point of view, therefore ensuring that story events are conveyed in a way that allows the listener to fully understand the relationship between events and construct a global picture. Due to the nature of the theory of mind task employed in this study (Forgeot d’Arc & Ramus, 2011), it is also possible to measure the participants’ understanding of physical causation. While physical causation would be presumed to be intact by the Weak Theory of Mind Hypothesis and some studies have shown this to be the case (e.g. Baron-Cohen et al., 1986), other findings presented above (e.g. Diehl et al., 2006; Kelley et al., 2006; Losh & Capps, 2003) show a more generalised difficulty with expressing causality in narratives. Therefore, contrary to the Weak Theory of Mind Hypothesis, it is hypothesised that some autistic individuals will struggle with the physical causation test items. An understanding of cause and effect would equip the storyteller with the necessary skills to comprehend how one event or detail may influence another, as well as what information a listener would require to fully understand the causal relationship between events, therefore, poor understanding of physical causation is hypothesised to result in less causally connected narratives. Finally, it is presumed that non-verbal (visuospatial) working memory is required to successfully narrate the story events, as the task is presented visually. Hypothesising that working memory, especially spatial working memory, is impaired in autism (Wang et al., 2017), weak non-verbal working memory is expected to negatively influence narrative connectedness.
With all the above in mind, the hypothesis tested in the current study is as follows:
Low causal connectedness in autistic adolescents’ narratives is associated with a) a local-processing bias
b) difficulties understanding others’ beliefs c) difficulties understanding physical causation d) low non-verbal working memory capacity.
4. METHODOLOGY
4.1. Participants
A total of 23 autistic participants (6 female) aged 13;2-17;9 (mean age: 15;4, SD: 19.6 months) were recruited from a HAVO/VWO special education high school in the centre of the
Netherlands devoted to students with a diagnosis of ASD. All participants were native Dutch speakers learning English as a second language except one, who was a native English speaker fluent in Dutch. Narrative recordings were not available for 4 of the participants, and so their data was excluded. One other participant was excluded for incorrectly completing the Global-Local task, resulting in analysable data for 18 participants in total. All the participants are presumed to have normal intelligence, since they have fluent speech and follow a regular high-school curriculum. Testing was conducted in the participants’ high-school, individually, in a quiet classroom by one of two experimenters. All tests were conducted in one session, which lasted approximately one hour.
Although English was the second language of the majority of our participants, they had a high level of English proficiency1. 14 of the 18 participants reported using English on a daily
basis, with the remaining 4 using it at least a couple of times a week. Multiple participants anecdotally expressed a keen interest in using English online as well as with family and friends, with some even saying they preferred to communicate in English than Dutch. Furthermore, English and Dutch express causal connectedness in very similar ways. Therefore, the fact that testing was conducted in English is not considered to be an issue.
The study aimed to additionally recruit a typically developing group of adolescents as a control group. However, due to the restrictions and safety concerns surrounding Covid-19, it was not possible to test this group.
4.2. Narrative task
Materials & Procedure
Narratives were elicited by means of The Pear Film (Chafe, 1980). The Pear Film is a 5-minute wordless video, which has previously been used to compare how participants from different cultural backgrounds choose a narrative strategy, and was originally used with American and Greek women (Tannen, 1979), and later compared with data from Spanish speakers (Blackwell, 2009) and written Spanish accounts (Orero, 2008). The film can be split into three main scenes. In the first scene, the viewer sees a man picking pears, another man walk past with a goat – an event that is interesting precisely because it is irrelevant to the rest of the story (Orero, 2008:6) – and a boy arrive on a bike, who then steals a basket of pears. In the second scene, a
sequence of events is presented, where a girl cycles past the boy, causing the his hat to fly off,
1A shortened version of the PPVT-III (Dunn, 1997) was administered to all participants from sets 9-14 of the original test. Maximum score = 72, group mean = 62.6, SD = 7.4, range = 44– 72.
which in turn leads to his distraction, hitting a rock, falling off his bike and spilling the pears. Three boys arrive on the scene, help the boy pick up the pears and, as they are leaving, one boy finds the hat on the ground. He gets the boy’s attention, returns the hat and in exchange receives three pears for him and his friends. This second scene therefore prompts the narrator to consider the causal relationships between events (Orero, 2008:6). In the final scene, we return to the farmer, who has just discovered his basket of pears is missing. He then looks up and sees the three boys, each eating a pear. This final scene invites the narrator to connect the previous events into a final conclusion and to describe the farmer’s thoughts and/or feelings (Orero, 2008:6). The Pear Film thus provides an excellent opportunity to evaluate how
participants narrate a series of interconnected events, how they deal with cause and effect and how they draw the story elements together into a final conclusion, involving inferences, thoughts and emotion.
The participants were shown this video on a laptop computer screen, which was tilted towards them so that the experimenter could not see. Prior to viewing the film, the participants were instructed to watch the video carefully and were made aware that they would be required to retell the story. After the film had finished the experimenter prompted the participant to tell the story in English and their narration was audiotaped.
Data Analysis
The audio recordings were transcribed by one of the two experimenters in CLAN, where CHILDES conventions were adhered to, including false starts, repetitions and filled pauses (MacWhinney, 2000). All transcriptions were checked by the second experimenter and any discrepancies were discussed until a resolution was agreed upon.
Following Diehl and colleagues (2006), the methodology of Trabasso and Sperry (1985) was used as a starting point for the coding, scoring and analysis. First, the Pear Film's plot was
mapped out by splitting it into discrete events and marking the causal connections (table 1). A causal connection between events indicates that the first event is necessary for the second event to occur (Trabasso & Sperry, 1985). For example, for the event 'he returns the hat' (event 18 in table 1) to occur, it is first necessary for the events 'his hat is knocked off’ (8; table 1) and 'one finds the boy’s hat' (16; table 1) to have occurred. These causal connections are thus represented by shaded cells in Table 1. The mapped events were limited to those that
constituted the gist storyline, which is defined as the sequence of events that causally connect the opening and closing events (Trabasso & Sperry, 1985), as, due to the nature of this or any film, the total number of events would be very high. From this gist storyline, a connectedness score was created, where one point was available for every successful causal connection, resulting in a total of 34 available points. The number of communication-units (c-units), where one c-unit comprises a verb and its arguments (Strong, 1998), were additionally counted for each participant's narrative as a measure of narrative length (Diehl et al., 2006). The final connectedness measure for each narrative was then calculated as the number of successful causal connections per c-unit.
Connection
(1) a farmer is picking pears
x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x (2) he puts the pears in a basket
(3) a boy arrives x x
(4) the farmer is in the tree (/not looking)
(5) he steals a basket of pears
(6) he cycles away
(7) a girl cycles past the boy
(8) his hat is knocked off
(9) the boy is distracted
(10) he hits a rock
(11) he falls
(12) the pears spill
(13) 3 boys help him up
(14) they help put the pears back in the
basket
(15) they walk away
(16) one finds the boy's hat
(17) he gets the boy's attention
(18) he returns the hat
(19) the boy gives him (3) pears
(20) as a thank you
(21) he shares the pears with his friends
(22) the farmer comes down from the tree
(23) he notices his basket is gone
(24) the boys walk past
(25) eating their pears
(26) the farmer watches them pass
(27) he is confused/he thinks they stole the
pears
4.3. Global-Local Task
Materials & Procedure
Attention switching/inhibition between global/local processing was tested using the Global-Local task (Peristeri & Tsimpli, 2020; Peristeri et al., 2019; based on Navon, 1977). The task
consisted of 3 blocks: local, global and switch. Each block contained 64 test items and was preceded by 12 practice items per block. Each test item was a shape (circle, X, triangle or square) made from smaller shapes (circle, X, triangle or square). Each trial was either
congruent, where the large shape and smaller shapes were the same, or incongruent, where the small shapes and large shape were different (see figure 1). Half the participants were presented with the ‘local first’ version, where they completed the blocks in the order local, global, local-global switch, while the other half completed the ‘global first’ order – global, local, global-local switch. Before each block, the participant first viewed written (Dutch) instructions about how to complete the task and were told which shapes to concentrate on (small (local), big (global) or small-big/big-small (local-global/global-local) switch). For each test item, participants were required to press the key corresponding to the number of lines needed to draw the target shape (1 for circle, 2 for X, 3 for triangle, 4 for square). Examples of the different conditions are given in figures 2-4.
Figure 1: Congruent (left) and incongruent (right) trials
Figure 2: Local condition
Target shape: X Target response: 2
Figure 3: Global condition Target shape: square Target response: 4
Figure 4: Local-global condition
Target shapes: square (local), square (global), triangle (local) Target responses: 4, 4, 3
Data Analysis
A slightly different approach to scoring the Global-Local task was taken compared with previous studies (Peristeri & Tsimpli, 2020; Peristeri et al., 2019) so that the direction (global or local) and extent of any processing bias would be incorporated into one measure. First, global switch costs in terms of reaction times (RTs) were calculated as the percentage increase from the average RT on the incongruent global non-switch trials to the average RT on the incongruent global switch trials. Local costs were calculated in the same way using the local trials. The two costs were then merged into one measure by subtracting the global RT cost from the local RT cost. This final measure, 'global/local RT cost', thus represents whether a participant shows a global bias (positive value) or a local bias (negative value) and how strong this bias is (distance from 0). Global switch costs for accuracy were computed for each participant by subtracting the accuracy score (% correct) on the incongruent global switch trials from the accuracy score on the incongruent non-switch global trials. Local switch costs were calculated in the same way for the incongruent local trials and once again a merged measure was calculated – 'global/local accuracy cost' – by subtracting the local cost from the global cost.
Linear models were computed to determine if the order of blocks (global-first or local-first) influenced the global/local RT cost and global/local accuracy cost scores. These analyses revealed that the order of blocks had a significant effect on the global/local RT cost at the p<0.05 level (mean = 74.58%, SE = 28.64%, p = 0.0192) and a significant effect on the global/local accuracy cost at the p<0.1 level (mean = 12.19%, SE = 6.85%, p = 0.941). As the observed local/global biases were influenced by the task design, the global/local switch cost scores for RT and accuracy were excluded from further analysis.
An alternative, exploratory measure was thus computed, which did not differentiate between local and global trials. A general 'attention switching/inhibition RT cost' was calculated as the RT percentage increase from non-switch to switch trials, and likewise a general 'attention
switching/inhibition accuracy cost' was calculated as the percentage difference between accuracy on the non-switch and switch trials. These measures therefore quantify each participant's performance (in terms of accuracy and RTs) on the switch trials, relative to their performance on the non-switch trials.
Prior to calculating these measures, the data was first checked and any inconsistencies were resolved. During testing, 3 participants alerted the investigator to the fact that they had accidentally reversed the order of the first block i.e. they had attended to the global level during the local block or vice-versa. This was noticed when they were presented with the instructions for the second block. These participants were encouraged to continue the task and to reverse the second block accordingly, so that one full block of local trials and one full block of global trials were still completed, followed by a switch block. Upon later inspecting the data, these participants were easily identified and their accuracy scores were adjusted according to the correct responses for the reversed condition.
RT outliers were removed from the data set prior to analysis. Following Peristeri et al. (2019), a value was considered to be an outlier if it was less than 250ms or more than two standard deviations above the mean for that condition (global non-switch, local non-switch, global switch, local switch). Outliers were then replaced with the mean value for that condition. This procedure resulted in 4.43% of the data being classified as outliers and replaced.
Looking at the accuracy scores in the switch block, some participants exhibited almost floor performance in a section of the block. This was attributed to persistent errors, where a participant made a switching error and then continued the pattern, e.g. responding global-local-local-global-local and so on. An adjustment procedure was therefore created to account for such errors. Adjustments were made if a switching error occurred and was directly followed by errors in the next 3 incongruent trials. In such a case, the original error and the first following error were maintained, while all following trials were re-coded and re-calculated according to the new pattern. This procedure significantly reduced the unfair effect of persistent errors, while also
ensuring an equal number of trials were coded as ‘global’ or ‘local’. 6 such adjustments were made across the data of 5 participants (out of 18). One other participant used the reverse pattern from the beginning of the block and their data was re-scored without penalisation.
4.4. Odd One Out Task
Materials & Procedure
A touchscreen version (Janssen, 2016) of the Odd One Out task (Henry, 2001) was
administered to test non-verbal working memory. The test consisted of 6 rounds, each with 4 trials, which became progressively harder by introducing more sets of test items to remember – one set in the first round, two sets in the second round and so on. A test item set consisted of three shapes or diagrams, two of which were the same and one of which was different – the odd one out. Memory for the location of each odd one out was tested using an empty grid. The participant was first introduced to the task using two practice items, which they were guided through with the help of (Dutch) audio recordings. During testing, the participant first viewed the test items individually, and selected the odd one out for each. The participant was then
presented with an empty grid and was required to select the locations of the odd ones out. If more than one trial contained errors, testing stopped at the end of that round, otherwise the participant progressed to the next round. Figures 5-7 give an example of the testing procedure in the second round.
Figure 5: The participant is presented with the first set of trial items and selects the odd one out.
Figure 6: The first set of trial items disappears and the participant is presented with the second set of trial items. They select the odd one out.
Figure 7: The second set of trial items disappears and the participant is presented with an empty grid. The participant selects where the odd ones out were located.
Data Analysis
A non-verbal working memory score was calculated for each participant, which corresponded to the total number of correctly remembered locations. If a participant incorrectly identified the odd one out, this was taken as the correct location and so they were not penalised for incorrect perception.
4.5. Theory of Mind Task
Materials & Procedure
The first half of a non-verbal theory of mind task developed by Forgeot D’Arc and Ramus (2011) was administered. The task consisted of a series of videos and alternative endings, which were presented on a computer screen. In the testing phase, there were 4 scenarios in total, each with 5 conditions, resulting in a total of 20 trials. For each test scenario there was one beginning scene (identical across conditions), directly followed by one of three change scenes (seen
change, unseen change, no change). These first two parts were played in the centre of the screen. At the end of the change scene, the video stopped, displaying a suspense frame, and two alternative endings played – one on the bottom left and one on the bottom right. The different combinations of change scene and end scenes resulted in 5 different conditions, shown in figure 8. Paths 4 and 5 were used for calculating physical causation scores, while paths 1 and 2 were used to calculate the false belief score. Path 2 was therefore a filler path. Prior to the testing phase, the participants completed 3 practice trials to familiarise them with the task set-up. The participant indicated the appropriate ending by pressing one of two keys.
Figure 8: The 5 conditions for one scenario in the Theory of Mind Task (Forgeot D’Arc & Ramus, 2011: 979).
Data Analysis
The participants’ responses were first marked as correct or incorrect. Then, following the methodology of Forgeot D’Arc and Ramus (2011), two scores were calculated, one for false belief reasoning, and one for physical causation reasoning. For the false belief reasoning score, the participant was required to choose the correct endings for both the mentalistic unseen change condition (path 1, end 1; figure 8) and the mentalistic seen change condition (path 2, end 2; figure 8). For the physical causation score, the participant was required to choose the correct ending for both the mechanistic (unseen change and no change) conditions (path 4, end 3 and path 5, end 4 respectively; figure 8). As there were 4 scenarios, this resulted in a
maximum 4 points for physical causation and false belief reasoning.
4.6. Predictions
From this methodology, the following prediction is made:
It is predicted that low narrative coherence scores correlate with
a) Higher attention shifting/inhibition costs, in terms of RTs and accuracy b) lower false belief scores on the theory of mind task
c) lower physical causation scores on the theory of mind task
d) lower non-verbal working memory scores on the Odd One Out task
4.7. Statistical analysis
Statistical analyses were conducted in R (R Core Team, 2017). The main model used was a linear model with the narrative connectedness score as the dependent variable and 5
independent variables: attention switching/inhibition RT cost, attention switching/inhibition accuracy cost, non-verbal working memory score, false belief score and physical causation score.
Prior to computing the model, a series of Pearson's product-moment correlation tests were carried out in order to confirm that none of the independent variables were highly correlated. No significant correlations were found between variables (p > 0.1; r < 0.5).
4.8. Individual and Qualitative Narrative Analysis
A further qualitative analysis was carried out, where individual narratives were assessed for their connectedness with regards to the hat storyline (events 8,16,18,19 and 20 in table 1) and the story’s conclusion (events 19,23,24,25,26 and 27 in table 1), focusing on where connections and events were missed, simplified or misunderstood as well as how characters and objects were (mis)categorised. As the quantitative analysis only considers individual connections between two events, this qualitative analysis therefore adds more depth and considers whether relevant background information is communicated to the listener, how individual events,
characters and objects are (mis)interpreted and how multiple events are drawn together into a coherent global story.
5. RESULTS
5.1. Descriptive Statistics
Measure Mean SD Range
Connections (max. = 34) 12.94 5.18 7–22
C-units 29.11 10.75 12–46
Connectedness (connections per c-units) 0.45 0.10 0.30–0.59 RT attention shifting cost (% increase) 80.08 20.00 55.34–125.63 Accuracy attention shifting cost (%
difference)
7.73 10.35 -1.56–36.72
Physical causation score (max = 4) 3.06 0.97 0–4
Belief score (max = 4) 2.72 1.48 0–4
Non-verbal working memory score (% correct)
85.78 21.94 11.90–100
Table 2: Summary of data across tasks for the group
Narrative Coherence: Figure 9 shows how many participants made each of the causal connections from the gist storyline of the narrative task.
Figure 9: Causal connections successfully made in the Pear Film Task (narrative).
Global-Local Task: Figures 10 and 11 show the average RTs and accuracy per condition. Figures 12 and 13 then show the individual switch costs in RTs and accuracy.
Figure 10: Average RTs per condition. Error bars represent the standard deviation from the mean.
Figure 11: Average accuracy per condition. Error bars represent the standard deviation from the mean.
Figure 12: Individual RT switch costs, in terms of % increase between non-switch and switch trials.
Figure 13: Individual accuracy switch costs, in terms of % difference between non-switch and switch trials.
Theory of Mind Task: Figure 14 shows the average scores for the Physical Causation and False Belief scores calculated from the Theory of Mind task. A linear mixed effects model was
computed to determine if there was a significant difference between scores for the two indices, with the participant as a random effect. No significant difference between scores was found (mean = 0.333 points, SE = 0.361 points, p = 0.369).
Figure 14: Average Physical Causation and False Belief indices from the theory of mind task (maximum score = 4). Error bars represent the standard deviation from the mean.
Non-verbal working memory: Figure 15 shows the individual scores, as a percentage, on the Odd One Out task. As the graph shows, the majority of the participants were at or close to ceiling performance.
Figure 15: Individual scores on the non-verbal working memory task
5.2. Associations between Narrative Coherence and Cognitive Abilities
A linear model was computed to test the effect of attention switching costs in RTs and accuracy, non-verbal working memory scores, false belief scores and physical causation scores on the connectedness score of the narrative task (connections per c-unit, abbreviated here as cpc). None of these variables were found to have a significant effect on the narrative connectedness score (attention switching/inhibition RT cost: mean = 0.002 cpc, SE = 0.001 cpc, p = 0.207; attention switching/inhibition accuracy cost: mean = -0.003 cpc, SE = 0.002 cpc, p = 0.196; non-verbal working memory: mean = 0.003 cpc; SE = 0.001 cpc; p = 0.805; belief score: mean = 0.025 cpc; SE = 0.017 cpc; p = 0.162; physical causation score: mean = -0.031 cpc, SE = 0.026 cpc, p = 0.254).
5.3. Individual and Qualitative Narrative Results
To complement the quantitative results above, individual narratives were additionally analysed qualitatively. This analysis concentrated on how multiple events were connected throughout the story, focusing on how the importance of the hat is conveyed and the farmer’s final reaction. Missed, simplified or misunderstood connections and events were recorded, as well as how objects and characters were (mis)categorised.
One striking pattern across multiple narratives is the removal or underdevelopment of the hat storyline. The boy's hat is taken from his head by the wind when a girl passes him2, also on a bicycle, leading to his distraction, hitting a rock and falling. 10/18 participants mentioned that the hat was lost at this point in the story, while 6/10 link it to the girl passing and 5/10 to him being distracted. Thus, we see that only roughly half of the participants mention this key event in the first place, and roughly half again manage to integrate it into the surrounding events. The hat becomes more relevant to the story as it progresses, as, after a group of boys help the boy who fell, one finds and returns the hat, gaining three pears in return. 7/18 participants mention the hat being found, while 12/18 mention its return. However, these are not necessarily the same participants as those who narrated its original loss. 3 out of the 10 participants who mention the hat being lost do not mention it being found, nor returned, while 6 of the 8
participants who failed to mention the hat being lost describe how it was found or returned. Of these latter participants, 5/6 refer to the hat, when it is found, as if its loss had already been narrated, for example through the use of an anaphoric reference without an antecedent, e.g. (1), or by saying that it was 'still lost', without previous mention of this event (2). The sixth participant mentioned the hat's loss directly before its return, but without any temporal anchoring or context for the listener to understand when this happened (3).
2 It is however difficult to perceive from the video that it is the wind that caused this and not the
girl herself. Therefore, any link between the girl passing and the hat being knocked off was accepted as correct.
(1) and &-uh then &-uh they bring him back his hat which he lost (participant 106) (2) and eventually &-um <he had> [/] <had> [/] he <had> [/] had still lost his hat and that
<they> [/] &+g &+g they gave that back &+g as well (participant 114)
(3) he also lost his hat, so the three kids also brought him that back (participant 112)
Linking all these connected events into a coherent storyline was clearly a challenge for most, as only 3/18 participants mention all three hat events – that it is lost, found and returned – and only 2 successfully complete the sequence, by additionally recognising that the pears are gained as a reward for this act. Meanwhile, 4 other participants make the immediate link between the return of the hat and the pears as a reward, but fail to mention at least one other event in the chain. Another 3 recognise that the pears are a reward, but for helping in general rather than for returning the hat, an event which does not feature in their narratives at all.
Another section of the story that requires making multiple connections is the end
sequence, where the farmer realises his basket is gone and sees the three boys, each eating a pear. For the listener to fully understand the farmer's thoughts, they need to know three things: that the farmer has realised his basket is missing, that the boys have pears and that the farmer has seen the boys. 15/18 participants mention that the farmer has seen the basket is missing, 8/18 say that the boys have pears (one of which however never mentions that they received them) and 9/18 that the farmer sees the boys, with only 6/18 mentioning all three. A total of 6 participants explicitly give the farmer's reaction, for example that he is confused, 3 of which do so after having mentioned all three pieces of necessary information. The other 3 describe his reaction after describing only one or two of these key events, and none connect the farmer's reaction to the boys having pears, which it certainly hinges on.
Actions and events were also misunderstood or caused confusion in some instances, for example the pears as a reward were instead interpreted as a possible theft by some, e.g. (4)
and (5). Participant 101's version of events (4) involves connecting some crucial elements – that they got some pears, that this was related to the return of the hat, that one boy got them and that he shared them – but misses the information that shows this was an amicable exchange. Participant 108's response (5), on the other hand, implies that he is aware of the importance of the boys having pears, that they must have come from somewhere and that this should be explained to the listener, but is clearly uncertain about how this happened.
(4) and <while> [/] while bringing it back he grabbed three pears for him and his friends (participant 101)
(5) &-um &-uh <they> [/] they have pears. I think they got them from <the> [/] the woman, dunno, stole them from the woman, no idea. (participant 108)
Another observation is that some struggled with categorising objects and characters and/or misinterpreted events. For example, two participants could not tell if the story was about apples or pears and repeatedly referred to both, e.g. (6). Both participants who struggled with identifying the fruit were also unsure about the goat that passed by, again giving multiple options (7), or qualifying the (wrong) interpretation with 'I think' (8). Likewise, the baskets of pears became 'boxes' (114), 'bags' (112) and 'buckets' (116) and multiple participants
categorised the characters in somewhat unusual ways. The main character, a boy of roughly 10 years old, is first referred to most frequently as a 'kid' (7 participants), then a 'boy' (6
participants) and a 'woman', 'guy', 'girl' 'person' and 'someone' once each. The girl that cycles past was mostly correctly categorised as a 'girl' (9 participants, 2 of whom showed uncertainty, e.g. 'a girl, I think'), but also referred to as a 'child', 'boy', 'woman', 'someone' and even a 'man'.
(7) when he was doing that there walked &-uh another man with a goat or a sheep or something like that (participant 106)
(8) and then someone with a mule, I think, passed by (participant 114)
6. DISCUSSION
6.1. Global/Local Attention Switching/Inhibition and Narrative connectedness
As mentioned previously (§4.3.) the original measures of global/local cost were not used, due to the significant effect of the stimuli order. Almost half the participants (8/18) displayed a global-processing bias in terms of RTs, rather than the predicted local-global-processing bias. However, for 15/18 participants the observed bias was in the opposite direction as the first stimuli of the switch block – i.e. if the first trial required global attention, then the participant exhibited a local-processing bias, as evidenced by larger global than local RT costs. The stimuli order also had some effect on accuracy costs, although to a lesser degree. This presumably indicates that many participants were so focused on the first type of switch that switching back became more effortful, and they then retained this pattern throughout the block. This in itself is a valuable finding that future studies employing this task should be aware of, and seek to overcome. It would also be useful to investigate this phenomenon further and determine if the order of stimuli has an effect on TD adolescents' performance as well. It should be noted that no such effect was reported in previous studies using this task (Peristeri & Tsimpli, 2020; Peristeri et al., 2019), which may be due to methodological differences, particularly if the effect is specific to autism, and if, in the study containing autistic participants (Peristeri & Tsimpli, 2020) the effect of stimuli order was computed with all three groups (autistic, SLI and TD) pooled. The fact that the task design had such an impact on local- and global-processing biases among the participants suggests that any underlying local-processing bias, if indeed present at all, can be easily
overcome by directing the participant's attention during the task. This would be in line with similar results showing that cueing has an effect on whether autistic individuals attend to a task globally or locally (e.g. López et al., 2004).
The measure used in the statistical analysis instead was one of general
switching/inhibition ability between processing levels, where performance on non-switch and switch trials were compared. This measure was not found to have a significant effect on the narrative connectedness measure. However, any results from this task should be approached with caution, considering that the task design was found to be so influential. Besides issues with the Global-Local task design, this null result could additionally or alternatively be due to the narrative measure not being nuanced enough. For example, the measure considers individual connections to be equal in weight and does not therefore fully quantify the complex reality of how these events and connections build on one another to make a global narrative. This idea is explored further in the qualitative analysis (§5.3. and §6.4.) and the considerations presented here might therefore be useful in creating new quantitative measures for future studies. Finally, another possibility is that autistic individuals in fact do not struggle with switching attention and/or inhibition between global and local levels of processing, or that, even if they do, this is not the cause of their difficulties with connecting events in discourse.
6.2. False Belief Reasoning, Physical Causation Reasoning and Narrative Connectedness
No association was found between narrative connectedness and false belief, nor physical causation reasoning. An interesting point worth mentioning, however, is the finding that belief scores did not significantly differ from physical causation scores, contrary to the Weak Theory of Mind Hypothesis (e.g. Baron-Cohen, Leslie, & Frith, 1985), which would predict intact physical causation reasoning alongside impaired false belief reasoning. Perhaps this dissociation would appear with more trials, as indeed only half the task was administered in this study. However,
this would come at a methodological cost, in terms of time and attention. The 20-trial version took our participants approximately 20 minutes each to complete, and most unfortunately found the task repetitive and expressed feelings of boredom. Increasing the number of trials would therefore be even more time-consuming and would potentially risk losing the participants' attention.
Alternatively, it may in fact be the case that the mentalistic and mechanistic trials were equally challenging for our autistic participants. This would imply that causal connections themselves are the problem, regardless of whether they relate to mentalising or not. This could explain why producing causally connected narratives is generally difficult for autistic individuals (e.g. Diehl et al., 2006; Kelley et al., 2006; Losh & Capps, 2003; Peristeri et al., 2017) and could stem from problems integrating information, potentially related to problems with forming
generalisations, i.e. the Reduced Generalisation Hypothesis (Plaisted, 2001; see §6.4). One way to investigate this, and to generally learn more about how autistic participants approach this task, would be to elicit self-reports after each trial. In this way, we could open a direct window into their reasoning abilities and techniques.
6.3. Non-verbal Working Memory
Non-verbal working memory scores were not found to have a significant effect on the narrative connectedness measure, and so this part of the hypothesis can be neither confirmed nor rejected. One reason for this is likely to be that many participants scored at or close to ceiling level and so there was little variation in the dataset. Further studies that make use of the Odd One Out task with adolescent or adult participants should therefore increase the number of levels beyond 6. The high rate of ceiling performance may also indicate that this task was not particularly challenging for the autistic participants recruited in this study. However, this of course cannot be known without a TD control group, and even in the case that comparisons are
possible in the future, the ceiling effect would likely prevent any meaningful conclusions from being drawn, unless the control group were to perform worse. Looking at Schaeffer and colleagues' (2018) study, which also employed the Odd One Out task and found no significant difference between ASD and TD participants in performance, it might well be the case that this task is not sensitive to the specific working memory difficulties experienced by autistic
individuals. Visuospatial working memory is not a unitary mechanism, but rather draws together different components – 'what', 'where', executive control and possibly image rehearsal have been so far identified (Baddeley in Carter & Frith, 2010: 310) – and so this opens up questions regarding which component or components are affected in autism, and additionally which are required, and to what degree, in various working memory tasks.
6.4. Individual and Qualitative Analyses of Narratives
This study is based on the hypothesis that autistic individuals struggle with connecting events in a narrative (Diehl et al., 2006; Kelley et al., 2006; Losh & Capps, 2003; Peristeri et al., 2017). While the lack of a TD comparison group means that this cannot be confirmed in the present data, a qualitative look at the narratives produced for this study reveals some interesting patterns and suggests that connecting story events is difficult for adolescents with ASD.
Fully connecting multiple events (e.g. those relating to the hat), as well as drawing conclusions (e.g. the farmer’s reaction) was clearly difficult for most of the participants, with only 2/18 successfully narrating all 4 of the crucial hat events (it is lost, found, returned and a reward is given) and similarly only 3 participants gave the farmer’s reaction in combination with the 3 key points that make sense of this (he realises his basket is gone, plus he sees the boys walking past and that they have pears). However, despite these difficulties there is evidence that the participants attempted to use the gist to structure their stories, just as would be
For example, 3/18 participants recognise that the pears were a reward, but connect this to helping in general, rather than to the specific act of returning the hat. This suggests a simplified storyline constructed using a broad global schema, consistent with Kintsch and van Dijk’s (1978) model of (neurotypical) discourse comprehension. Likewise, where one participant (5, §5.3) expressed confusion regarding the origin of the pears, his attempt to explain this event suggests a global to local approach, where he is aware of the importance of this event to the gist and is trying to reconstruct the details, just as typically developing children learn to do (Brainerd & Reyna, 1998). Furthermore, it was found that 6 (out of 18) participants failed to mention the hat being lost then later referred to this event in their narrative. This suggests that establishing background information in order to refer back to it at a later stage was difficult for at least some participants, similar to the observation made by Diehl and colleagues (2006).
However, despite not fully anchoring the loss of the hat in the storyline, by mentioning it when they did they at least showed some understanding of the hat's importance to the story gist.
It was further observed that some participants struggled to correctly identify characters and objects, as evidenced by the production of categorisation errors, multiple options,
expressions of uncertainty or broad categories. The examples of miscategorisation point to difficulties in placing certain stimuli into categories based on the integration of various cues (shape, material, size etc.), while the preference of broader categories ('kid', 'person', 'someone' etc.) used by some may be a strategy for avoiding errors. Interestingly, the use of broad
categories is in opposition to the findings of Alderson-Day and McGonigle-Chalmers (2010), who found their autistic participants (n=14, mean age = 13;4) used more narrow categories than their age-matched TD peers while completing the Twenty Questions Task (Mosher & Hornsby, 1966). The difference in findings may be due to the interaction of memory, as unlike in our narrative task, Alderson-Day and McGonigle-Chalmer’s (2010) participants were producing categories while viewing the stimuli. The observed categorisation patterns in this study are therefore perhaps a product of or compensation for weak memory. In this case, this would
