You may notice some images loading slow across the Open Collections website. Thank you for your patience as we rebuild the cache to make images load faster.

Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Discourse coherence and atypicality in autistic adults : from corpus analysis to subjective impressions… Geelhand de Merxem, Philippine 2019

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata

Download

Media
24-ubc_2019_november_geelhanddemerxem_philippine.pdf [ 4.56MB ]
Metadata
JSON: 24-1.0384568.json
JSON-LD: 24-1.0384568-ld.json
RDF/XML (Pretty): 24-1.0384568-rdf.xml
RDF/JSON: 24-1.0384568-rdf.json
Turtle: 24-1.0384568-turtle.txt
N-Triples: 24-1.0384568-rdf-ntriples.txt
Original Record: 24-1.0384568-source.json
Full Text
24-1.0384568-fulltext.txt
Citation
24-1.0384568.ris

Full Text

  DISCOURSE COHERENCE AND ATYPICALITY IN AUTISTIC ADULTS FROM CORPUS ANALYSIS TO SUBJECTIVE IMPRESSIONS OF SPOKEN DISCOURSE by PHILIPPINE GEELHAND DE MERXEM B.A., Université de Namur, 2013 M.A., Utrecht University, 2015 A DISSERTATION SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  DOCTOR OF PHILOSOPHY  in  THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (Anthropology) (English) THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) October 2019 © Philippine Geelhand de Merxem, 2019           Discourse coherence and atypicality in autistic adults From corpus analysis to subjective impressions of spoken discourse  Philippine GEELHAND DE MERXEM  with a view to obtaining the PhD Degree in Language, Literature and Communication (ULB - “Langues, Lettres et Communication” and UBC - Department of English Language and Literatures) under the supervision of Professor Mikhail KISSINE (Université libre de Bruxelles) and Professor Jessica de VILLIERS (University of British Columbia)  Academic year 2018-2019 iii  The following individuals certify that they have read, and recommend to the Faculty of Graduate and Postdoctoral Studies for acceptance, the dissertation entitled:  Discourse coherence and atypicality : from corpus analysis to subjective impressions of spoken discourse  submitted by Philippine Geelhand de Merxem  in partial fulfillment of the requirements for the degree of Doctor of Philosophy in English   Examining Committee:  Mikhail Kissine Co-supervisor Jessica de Villiers Co-supervisor  Jürgen Jaspers Supervisory Committee Member Philippe de Brabanter Supervisory Committee Member François Heinderyckx University Examiner Stefan Dollinger University Examiner  iv  Abstract  Adequate and efficient participation in a social interaction requires the ability to produce and interpret coherent discourse. Coherence can be achieved using a diverse set of cognitive and linguistic tools such as topic continuity or connectives. Failing to produce enough coherence cues or to interpret them as such can lead to communication breakdowns, potentially compromising the on-going interaction. In a neurodevelopmental disorder like autism, characterized by impairments in social communication, autistic individuals often fail to produce and understand coherent discourse. Numerous studies have already used discourse analysis to examine discourse (in)coherence and (a)typicality in autistic individuals. However, the studies have been one-sided in terms of methodology, viz. investigating either discourse production or comprehension, but not both. They have also been one-sided in terms of the perspective of the analysis, viz. only considering discourse coherence from the standpoint of neurotypical individuals, without also considering the perspective of autistic individuals. However, social communication is a two-sided dynamic, whereby both communication partners contribute to the coherence of the unfolding discourse. Therefore, the aim of this dissertation is to define the qualities of discourse (in)coherence and (a)typicality in autistic adults by combining both detailed transcript analyses and more subjective interpretation of spoken discourse, by both neurotypical and autistic individuals. To achieve these aims, a corpus of interviews of French-speaking autistic adults and French-speaking neurotypical adults, matched on gender, age and IQ was collected. The data of three interview tasks was annotated and underlie the entire work of this dissertation. On the basis of the annotated data, detailed transcript analyses were performed to examine both the content and the delivery strategies of spoken discourse in autism. In a subsequent step, discourse features identified in transcripts are related with their perception by naïve listeners with and without a diagnosis of autism as well as their contribution to impression formation of the speaker. Taken together, this dissertation shows a consistent difficulty in the production of coherent discourse which transpired both in content and delivery strategy. Crucially, reduced discourse coherence resulted in a one-sided ‘neurotypical’ bias towards autistic individuals, which is likely to further hinder their communication success.    v  Lay Summary  In my dissertation, I sought, on the one hand, to create an inventory of linguistic characteristics that can distinguish between the spoken discourse of autistic individuals and neurotypical individuals. On the other hand, I sought to investigate the real-life implications of these linguistic differences on the perception of (in)coherence and (a)typicality of autistic individuals. The results of my dissertation show that linguistic differences surface both in the content and speech delivery of spoken discourse of autistic individuals; affecting their ability to  produce coherent discourse. Crucially, reduced discourse coherence resulted in a one-sided ‘neurotypical’ bias towards autistic individuals, which is likely to further hinder their communication success. This dissertation highlights that biased perception may reduce social opportunities for autistic individuals, contributing to their social difficulties. Raising awareness to linguistic differences in autism as well as to the existence of biases against autistic individuals can help mitigate negative attitudes towards them.    vi  Preface  This dissertation is formatted in accordance with the regulations of Université libre de Bruxelles and submitted in partial fulfillment of the requirements for a PhD degree awarded jointly by Université libre de Bruxelles and the University of British Columbia. Versions of this dissertation will exist in the institutional repositories of both institutions. The experiments reported in this dissertation are covered by Université libre de Bruxelles (ULB) Ethics Certificate numbers 063/2015 and 060/2017 as well as by University of British Columbia (UBC) Ethics Certificate number H18-00030.   vii  Table of Contents  Abstract ................................................................................................................................................. iv Lay Summary ........................................................................................................................................ v Preface ................................................................................................................................................... vi Table of Contents………………………………………………………………………………………………………………………viii List of Tables ......................................................................................................................................... x List of Figures..................................................................................................................................... xiv Acknowledgments ............................................................................................................................ xvi Introduction ......................................................................................................................................... 1 Chapter 1: Methodology.................................................................................................................... 8 I. Corpus description ................................................................................................................. 8 II. Methodology............................................................................................................................. 8 2.1. Participants .......................................................................................................................... 8 2.2. Procedure ............................................................................................................................ 9 2.3. Material .............................................................................................................................. 10 2.4. Data preparation ............................................................................................................... 12 Chapter 2: Narrative production in autistic adults ................................................................... 29 I. Introduction ........................................................................................................................... 29 1.1. Narrative discourse as a measure of language use in autism ............................................ 29 1.2. Characteristic of narrative discourse in autism ................................................................ 32 1.3. Age and narrative performance......................................................................................... 33 II. Methodology........................................................................................................................... 35 2.1. Participants ........................................................................................................................ 35 2.2. Material .............................................................................................................................. 36 2.3. Procedures ........................................................................................................................ 36 2.4. Linguistic measures ............................................................................................................ 37 2.5. Data preparation ............................................................................................................... 41 2.6. Analysis .............................................................................................................................. 42 III. Results .................................................................................................................................. 43 3.1. Narrative microstructure .................................................................................................. 43 3.2. Narrative macrostructure ................................................................................................. 49 3.3. Internal State Language ...................................................................................................... 59 3.4. Summary of coding results ................................................................................................ 63 IV. Discussion ........................................................................................................................... 64 V. Conclusion .............................................................................................................................. 68  viii  Chapter 3: How do autistic adults use syntactic and prosodic cues to manage spoken discourse? ............................................................................................................................................ 73 I. Introduction ........................................................................................................................... 73 II. Methodology........................................................................................................................... 81 2.1. Participants ........................................................................................................................ 81 2.2. Material .............................................................................................................................. 82 2.3. Procedure .......................................................................................................................... 83 2.4. Analysis .............................................................................................................................. 84 III. Results .................................................................................................................................. 84 3.1. Syntactic coding ................................................................................................................. 84 3.2. Basic discourse units ......................................................................................................... 91 3.3. Summary of results ............................................................................................................ 95 IV. Discussion ........................................................................................................................... 96 V. Conclusion ............................................................................................................................ 101 Chapter 4: Relating corpus findings to their impressions by naïve listeners .................... 103 I. Introduction ......................................................................................................................... 103 II. Pretest ................................................................................................................................... 107 2.1. Methodology .................................................................................................................... 107 2.2. Analysis ............................................................................................................................ 110 2.3. Results ............................................................................................................................. 110 2.4. Summary of results .......................................................................................................... 111 III. Auditory Rating Experiment ........................................................................................ 112 3.1. Methodology .................................................................................................................... 112 3.2. Analyses ........................................................................................................................... 114 IV. Results ................................................................................................................................ 114 4.1. Full Scale .......................................................................................................................... 114 4.2. Subscale on discourse content ........................................................................................ 116 4.3. Subscale on speaker impressions .................................................................................... 118 4.4. Summary of results .......................................................................................................... 121 V. Discussion ............................................................................................................................. 122 VI. Conclusion ........................................................................................................................ 123 Chapter 5: Serendipitous findings ............................................................................................... 125 I. Introduction ......................................................................................................................... 125 II. Results .................................................................................................................................... 126 2.1. Participants ...................................................................................................................... 126 2.2. Syntactic coding and BDUs ............................................................................................. 126  ix  2.3. Auditory rating experiment ............................................................................................. 128 III. Discussion ......................................................................................................................... 133 Chapter 6: General Discussion .................................................................................................... 135 References ........................................................................................................................................ 143 Appendix 1 : Guidelines for orthographic transcription (French Version) ....................... 155 Appendix 2 : Annotation guidelines for narratives (French Version)................................. 158 Appendix 3 : Examples of narratives .......................................................................................... 167  x  List of Tables  Table 1 Descriptive statistics of participants' characteristics by group ................................... 9 Table 2 ADOS-2 interview questions of the task Friendship, Relationships and Marriage and Solitude ..................................................................................................................................... 11 Table 3 Corpus examples of dependency clause subtypes .................................................... 19 Table 4 Corpus examples of functional sequences subtypes ................................................. 21 Table 5 Corpus examples of categorical sequences subtypes ............................................... 22 Table 6 Corpus examples of adjuncts, discourse-structuring devices, hesitation markers and silent pauses .............................................................................................................................. 24 Table 7 Summary of the different types of Basic Discourse Units ........................................ 28 Table 8 Descriptive statistics of participants’ characteristics per diagnostic group .............. 36 Table 9 Means and standard deviations (in brackets) of IQ scores per diagnostic group .... 36 Table 10 Percentage inter-coder agreement per coding category ........................................ 37 Table 11 Main story elements in the wordless picture book Tuesday ................................. 37 Table 12 Additional categories of story structure ................................................................ 38 Table 13 Coding categories of reference .............................................................................. 40 Table 14 Coding categories of connectives and discourse markers ..................................... 40 Table 15 Coding categories of internal state language .......................................................... 41 Table 16 Regression coefficients of the generalized link model with the additive effect of diagnostic group (ASD diagnosis is the reference level, standard errors is in brackets) ........ 43 Table 17 Means and standard deviations (in brackets) of total words, total syntactic sequences and total syntactic units ............................................................................................................ 43 Table 18 Means and standard deviations (in brackets) of counts and percentage of total dependency clauses, complete dependency clauses and incomplete dependency clauses per diagnostic group ....................................................................................................................... 45 Table 19 Means and standard deviations (in brackets) of counts and percentage of dependency clause subtypes per diagnostic group .................................................................. 46 Table 20 Means and standard deviations (in brackets) of counts and percentage of additional syntactic units per diagnostic group ......................................................................................... 48 Table 21 Means and standard deviations (in brackets) of total counts of main story elements, additional story events and extraneous comments per diagnostic group ............................... 50  xi  Table 22 Means and standard deviations (in brackets) of counts and percentage of discourse-structuring devices per diagnostic group ................................................................................. 52 Table 23 Distribution of the different discourse-structuring devices within diagnostic group .................................................................................................................................................. 52 Table 24 Means and standard deviations (in brackets) of counts and percent of referential expressions per diagnostic group............................................................................................. 55 Table 25 Distribution of the referential expression subtypes within diagnostic group ........ 55 Table 26 Means and standard deviations (in brackets) of counts and percentage of internal state language per diagnostic group ......................................................................................... 60 Table 27 Distribution of the different internal state language subtypes within diagnostic group .................................................................................................................................................. 60 Table 28 Summary of all coding categories and their associated group effect ..................... 63 Table 29 Summary of the different types of Basic Discourse Units and their corresponding strategies ................................................................................................................................... 80 Table 30 Descriptive statistics of participants’ characteristics per diagnostic group (ASD is the reference level) .................................................................................................................. 82 Table 31 Means and standard deviations (in brackets) of the IQ scores per diagnostic group (ASD is ...................................................................................................................................... 82 Table 32 Characteristics of the discourse genres informal conversation, formal conversation, conversational narrative, sociolinguistic interview and semi-structured questions ................ 83 Table 33 Regression coefficients of the generalized link model with the additive effect of diagnostic group (ASD diagnosis is the reference level, standard errors is in brackets) ........ 85 Table 34 Means and standard deviations (in brackets) of total words, total syntactic sequences and total syntactic units ............................................................................................................ 85 Table 35 Means and standard deviations (in brackets) of counts and percentage of total dependency clauses, complete dependency clauses and incomplete dependency clauses per diagnostic group ....................................................................................................................... 87 Table 36 Means and standard deviations (in brackets) of counts and percentage of dependency clause subtypes per diagnostic group .................................................................. 88 Table 37 Means and standard deviations (in brackets) for counts and proportions (in italics) of discourse-structuring devices, adjuncts and hesitation markers per diagnostic group....... 90  xii  Table 38 Means and standard deviations (in brackets) for counts and percent (in italics) of total BDUs, congruent BDUs (bdu-c), silence-bound BDUs (bdu-sil), regulatory BDUs (bdu-r), syntax-bound BDUs (bdu-s) and mixed BDUs (bdu-x) per diagnostic group ......................... 91 Table 39 BDU distribution within diagnostic group .............................................................. 93 Table 40 Distribution of different BDUs summarized over diagnostic group ...................... 94 Table 41 BDU distribution across informal conversation (conv-i), formal conversation(conv-f), conversational narrative(conv-narr) and sociolinguistic interview(int-soc) ........................ 94 Table 42 BDU distribution in the multi-genre corpus LOCAS-F .......................................... 95 Table 43 Summary of all coding categories and their associated group effect ..................... 96 Table 44 Scale Items ............................................................................................................. 108 Table 45 Cumulative link mixed model with additive effects of speaker diagnosis ............ 111 Table 46 Mean and standard deviations (in brackets) ......................................................... 111 Table 47 Descriptive statistics of participants’ characteristics (ASD is the reference level) ................................................................................................................................................ 113 Table 48 Means and standard deviations (in brackets) of IQ scores (ASD is the reference level) ....................................................................................................................................... 113 Table 49 Cumulative link model with additive effects of speaker diagnosis and rater diagnosis ................................................................................................................................................ 114 Table 50 Means and standard deviations (in brackets) of rating scores per speaker diagnosis ................................................................................................................................................ 115 Table 51 Cumulative link model with additive effects of speaker diagnosis and rater ....... 115 Table 52 Means and standard deviations (in brackets) rating scores of ............................. 115 Table 53 Cumulative link model with additive effects of speaker diagnosis and rater diagnosis ................................................................................................................................................ 116 Table 54 Means and standard deviations (in brackets) of the rating scores for the seven items of the discourse subscale per speaker diagnosis.................................................................... 116 Table 55 Cumulative link model with additive effects of speaker diagnosis and rater diagnosis and the interaction effect between speaker diagnosis and rater diagnosis (ASD diagnosis is the reference level for all effects, standard errors are between brackets) ................................. 119 Table 56 Means and standard deviations (in brackets) of the ratings on the four items of the speaker subscale per speaker diagnosis ................................................................................. 119  xiii  Table 57 Means and standard deviations (in brackets) of the rating scores on the four items of the speaker subscale per rater diagnosis ........................................................................... 120 Table 58 Means and standard deviations (in brackets) of the rating scores for the items ease of making oneself be understood, ease of speaker understanding rater and likelihood of becoming friends per speaker diagnosis and rater diagnosis ................................................. 120 Table 59 Regression coefficients of the generalized link model with additive .................... 127 Table 60 Means and standard deviations (in brackets) of total ........................................... 127 Table 61 Regression coefficients of the generalized link model with .................................. 129 Table 62 Means and standard deviations (in brackets) of the rating scores ....................... 129 Table 63 Means and standard deviations (in brackets) of total rating ................................. 129 Table 64 Cumulative link model with additive effects of speaker sex and the interaction 130 Table 65 Means and standard deviations (in brackets) for the rating scores ...................... 130 Table 66 Means and standard deviations (in brackets) for the rating scores of the ........... 131 Table 67 Cumulative link model with additive effect of speaker sex. ................................. 132 Table 68 Means and standard deviations (in brackets) of the rating scores on the items ease of making oneself be understood, ease of speaker understanding rater and likelihood of becoming friends per speaker sex.......................................................................................... 133     xiv  List of Figures  Figure 1 Screenshot of a Praat TextGrid with an orthographic transcription ...................... 13 Figure 2 Screenshot of a Praat TextGrid with a phonetic transcription .............................. 14 Figure 3 Screenshot of a Praat TextGrid with an orthographic and phonetic transcription as well as words and syllable segmentation ................................................................................. 15 Figure 4 Flowchart of the data preparation process ............................................................. 16 Figure 5 Screenshot of a Praat TextGrid with all transcription and segmentation tiers ...... 25 Figure 6 Violin plots of total words (plot A), total syntactic sequences (plot B) and total syntactic units (plot C) per diagnostic group ........................................................................... 44 Figure 7 Violin plots of dependency clauses (plot A), complete (plot B) and incomplete dependency clauses (plot C) per diagnostic group .................................................................. 45 Figure 8 Violin plots of complete verbal (plot A), averbal (plot B) and elliptic dependency clauses (plot C) per diagnostic group ...................................................................................... 47 Figure 9 Violin plots of discourse-structuring devices (plot A), adjuncts (plot B) and hesitation markers (plot C) per diagnostic group .................................................................................... 49 Figure 10 Violin plot for main story elements (plot C), additional story events (plot B) and extraneous comments (plot C) per diagnostic group ............................................................. 51 Figure 11 Violin plots of total discourse-structuring devices (plot A), additive connectives (plot B), temporal connectives (plot C), causal connectives (plot D), contrastive connectives (plot E) and discourse markers (plot F) per diagnostic group ................................................. 53 Figure 12 Violin plots for total referential expressions (plot A), definite nominal expressions (plot B), indefinite nominal expressions (plot C) and pronominal expressions (plot D) per diagnostic group ....................................................................................................................... 56 Figure 13 Violin plots for total counts of references to toads and frogs (plot A), man eating a midnight snack (plot B), lady asleep in front of her television (plot C), frog chased by the dog (plot D), dog (plot E), turtle (plot F), fish (plot G), birds (plot H), cat (plot I), media (plot J) and pigs (plot K) per diagnostic group ..................................................................................... 57 Figure 14 Violin plots of  total internal state language (plot A), cognition terms (plot B), emotions terms (plot C), physiology terms (plot D), modal terms (plot E) and evaluative terms (plot F) per diagnostic group .................................................................................................... 61 Figure 15 Violin plots of total words (plot A), syntactic sequences (plot B) and syntactic units (plot C) per diagnostic group ................................................................................................... 86  xv  Figure 16 Violin plots for total dependency clauses (plot A), complete dependency clauses (plot B) and incomplete dependency clauses (C) per diagnostic group .................................. 87 Figure 17 Violin plots for complete verbal (plot A), averbal (plot B) and elliptic (plot C) dependency clauses per diagnostic group ................................................................................ 89 Figure 18 Violin plots for discourse-structuring devices (plot A), adjuncts (plot B) and hesitation markers (plot C) per diagnostic group ................................................................... 90 Figure 19 Violin plots for total BDUs (plot A), congruent BDUs (plot B), silence-bound BDUs (plot C), regulatory BDUs (plot D), syntax-bound BDUs (plot E) and mixed BDUs (plot F) per diagnostic group ....................................................................................................................... 92 Figure 20 Diverging stacked bar charts representing the ratings of the speakers for the seven items of the discourse subscale .............................................................................................. 117 Figure 21 Diverging stacked bar charts representing the ratings on the four items of the speaker subscale per rater diagnosis and speaker diagnosis ................................................. 121 Figure 22 Boxplots of discourse-structuring devices per participants’ diagnosis and sex .. 128 Figure 23 Diverging stacked bar charts of the ratings on the items relevance (plot A), coherence (plot B) and pedantic style (plot C) per speaker diagnosis and sex .................... 131     xvi  Acknowledgments  Writing this dissertation was a four-year journey which would have been impossible without the support of many people.   First and foremost, I would like to express my deep gratitude to my two supervisors Prof. Mikhail Kissine and Prof. Jessica de Villiers. I am very thankful to Mikhail for being enthusiastic about my research ideas and interests. Your guidance, support and readiness during these four years have helped me greatly to develop these ideas and become a better researcher. I am also very thankful to Jessica who welcomed me in Vancouver. I have really appreciated your availability (which was not always easy with a nine-hour time difference!) and the time you have given me. I have really enjoyed all our discussions and I’m very grateful for the support and advice you have provided me at different points in my PhD.  I am also very grateful to the members of my jury who have accepted to take part in the final step of my PhD.   I would also like to thank the foundation Jean-François Peterbroeck for funding my PhD, giving me the opportunity to bring to fruition my research project.    I cannot thank enough all the wonderful team members of ACTE who have helped me in so many ways during these four years. I’m very thankful to Ekaterina Ostashchenko who has helped me since the first days of my PhD. I have really appreciated all our discussions and shared research experiences. I am also very thankful to Pauline Maes and Fanny Papastamou for their precious help in the final stretch. You helped me overcome hurdles of the writing process, and all this with a smile and good humor. I’m also grateful to Adeline Hanzir for being such a great office mate, your cheerfulness has been of great help during more stressful periods. I am also very thankful to Gaétane Deliens who was always ready to help me with my research. I am also very grateful to Elise Clin, Fanny Stercq, Vanessa Demeuldre, Nicolas Ruytenbeek, Jonathan Noël, Marielle Weyland, Eleanor Miller and Morgane Colin. Thank you for your interest in my research, your attentive listening and your encouragements.    xvii  Creating a new corpus of annotated data was an ambitious task and would have been impossible without the help of linguistics students:  Hélène Gygax, Marie Bélenger, Alba Leszczynski, Chloé Martinache, Céline Lemonne & Samy Ialy.  I am also very thankful to all my participants. Thank you for your interest in my research and the time you dedicated for me.  During these four years, I have also received amazing support from all my friends.  In particular, I would like to thank Cora. Your support throughout the years has been invaluable. I am very grateful for your attentive listening, good advice and mental support. You have been a great source of inspiration. I am also very grateful to Arthur who has always been there for me. Thank you for finding the right words to motivate me when I needed reassurance. Edo, thank you for your good spirit and reminding me to relativize in stressful situations. I would also like to thank Merel and Nadya who despite the distance have continued to encourage me during these four years.   I am also very grateful to all the people I met during my stay in Vancouver. Thank you to my first friends at UBC, Lucas and Boris, your friendship made the rainy months in Vancouver less intimidating. Thank you to Anna for being such a great roommate and friend; our discussions and your good advice were of great help during my PhD candidacy. A special thank you to Greg, Mike and Hannah for making sure I took enough breaks from my dissertation!   I am extremely grateful to my boyfriend Wouter. Thank you for believing in me and pushing me to give the best of myself, especially in stressful periods. You have been an incredible source of motivation and helped me get through the final stretch.  Last but (definitely) not least, I am extremely grateful to my family. Manoëlle, Arnould, Aurélien & Dorian, thank you for your constant support, enthusiasm and attentive listening (despite the occasional very long monologues about my research). Your support throughout the years have been really precious.    xviii  Most of all, thank you to my parents, Carine & Stéphane, who have never doubted me and have supported me in all my endeavors. I could not have made it without your love and encouragements.   1  Introduction  The aim of my dissertation is to better understand how adults with a diagnosis of autism produce and perceive spoken discourse by combining both detailed transcript analyses and more subjective impressions of spoken language. The research in this dissertation thus concerns individuals with a clinical condition. Therefore, I will first define the terminology that will be used throughout this thesis to refer to these individuals as the impact of the words we use to describe, define and refer to them (and their condition) is not trivial (Dunn & Andrews, 2015). In accordance with the expressed preferences of the community of autistic adults (Kenny et al., 2016), I will not use ‘person-first’ language (e.g., adults with autism) but will use ‘identity-first’ phrasing (e.g., autistic adults). Comparison participants without a diagnosis of autism will be referred to as neurotypicals (NT).  If you were to judge a speaker based on the following utterance: “uh I as long as the shared house uh but in the end with my ex-partner it was it was a bit like that except that we were one room short1” What would be your first impression? Probably not a very positive one. Humans are extremely rapid at forming first impressions of another person, even based on brief glimpses of the person’s verbal and non-verbal behavior. If an individual displays atypical social behavior, the first impressions of these behaviors as odd or peculiar can have clinical value. This is the case for Autism Spectrum Disorder (ASD), a neurodevelopmental disorder characterized by core impairments in social and communication behaviors (American Psychiatric Association, 2013). Currently, an ASD diagnosis is ascertained based on expert clinical impressions. For example, the Autism Diagnostic Observation Schedule (ADOS, Lord et al., 2000) is a tool for assessing and diagnosing ASD which is based on the observation of a series of verbal and non-verbal behaviors of the individual being assessed such as communication, social interaction, and play (or imaginative use of materials). The quality/typicality of these different behaviors is rated by the clinicians, leading to a diagnosis of autism (or not). In such a setting, many autistic individuals will display distinct behaviors that will be recognized within minutes (de Marchena & Miller, 2017). For autistic adults, discourse competence, viz. producing and understanding coherent discourse, is one of those recognizable behaviors. However, while it might be easy to determine at first glance that the quality of communication is atypical or incoherent, it is much more difficult to pin-point exactly  1 Original utterance : « euh bah moi du moment que la coloc euh mais en fin de compte avec mon ex conjoint c’était c’était un peu ça à part qu’il manquait une pièce »  2  what speech characteristics are leading to this perception of atypicality and/or incoherence. Speech abnormalities are one of the criteria in the ADOS and are coded by the clinician during the test administration, but they are not considered in the diagnostic algorithm due to a lack of agreement between clinicians (Bone, Black, Ramakrishna, Grossman, & Narayanan, 2015).  Applied to clinical populations, discourse analysis is a theoretical model and methodological approach that may help unveil subtle patterns of speech abnormalities that characterize disorders or population with impairments (Müller, Guendouzi & Wilson, 2008). From a theoretical point of view, a main focus of discourse analysis is to examine how coherence is established between the utterances composing the discourse. Indeed, establishing coherence is an important process that allows us to communicate as well as understand more meaning than that which is merely contained in the individual utterances within the discourse (Kehler, 2002). While coherence qualifies the mental representation of the discourse rather than of the discourse itself, its reconstruction often relies on linguistic elements of the discourse itself (Sanders & Spooren, 2009). As such, discourse provides an excellent window on the relationship between linguistic elements in the discourse and quality of the discourse representation, viz. how the presence/absence of certain linguistic forms impacts the mental representation of the discourse. While discourse analysis is a useful approach for many different clinical disorders such as dementia (e.g., Dijkstra, Bourgeois, Allen, & Burgio, 2004) or traumatic brain injury (e.g., Coelho, Liles, & Duffy, 1991), it is particularly well suited to describe and understand the communicative profiles of individuals who have acquired average phonetic, lexical and grammatical skills but continue to experience communication difficulties. Indeed, discourse analysis can capture language difficulties extending beyond the sentence-level into other domains of competence such as difficulties with social and cognitive skills. From a methodological point of view, discourse analysis usually relies on orthographic transcriptions of speech, which can be subsequently coded for various linguistic criteria. Transcriptions allow detailed a posteriori analyses of fluent speech, bypassing the limitations entailed by the temporality and volatility of speech.  One type of discourse that has been extensively studied in autism is narrative. Narratives are a fundamental part of human experience and play a key role in various spheres of our lives. Most of our daily communication occurs in the form of story narratives (Lê, Coelho, Mozeiko, & Grafman, 2011), from telling a bedtime story to your children to sharing the latest gossip with your friends on the playground to complaining about your boss to your  3  partner after a long day at work. Narrative competence is thus a key characteristic of our personal identity and social functioning of human societies (e.g., Bruner, 1991, 2004). As such, narratives serves as context to study how language is organized and used to verbalize complex ideas, allowing for a multitude of analyses within-and between utterances of the story (Lê et al., 2011). Considering the distinct characteristics of this discourse genre, narratives have provided a unique insight into autism-specific problems associated with language and social difficulties. Recent results suggest that autistic individuals perform less well than their neurotypical peers on all three major dimensions of narrative production (to be discussed in much more detail in Chapter 2): (i) microstructure (productivity and grammar), (ii) macrostructure (coherence and cohesive adequacy), and (iii) internal state language (Baixauli, Colomer, Rosello, & Miranda, 2016). The most salient difficulty is related to the global coherence of the narrative production, viz. difficulties with the overall conceptual organization of the narrative. Specifically, autistic individuals experience difficulties referring to meaningful events - whether fictitious (related to story characters) or personal (related to the self) - to organize the structure and content of the narrative discourse, as well as reduced use of causal and internal state language. In addition to providing key discourse features differentiating the speech of autistic and neurotypical individuals, some studies have highlighted the relationship between narrative and social competence in autism (Volden et al., 2017). As a result, narrative production is becoming an important target of intervention treatments (Gillam, Hartzheim, Studenka, Simonsmeier, & Gillam, 2015; Petersen et al., 2014).  While transposing speech from its original interaction context onto a written transcription has allowed fine-grained analyses of speech patterns in clinical contexts, these transcript analyses typically lack prosodic information. Yet, atypical prosody is very common in autistic individuals (McCann & Peppé, 2003). These abnormalities tend to persist well into adulthood (DePape, Hall, Tillmann, & Trainor, 2012; Fusaroli, Lambrechts, Bang, Bowler, & Gaigg, 2017; Shriberg, 2001). Even those autistic adults whose average linguistic and cognitive skills are in the typical range will often display atypical prosody, deviating from the expected norm in a wide range of ways. That is, their prosody has been described as unusually flat or monotone, variable, sing-songy, pedantic, machine-like, stilted, bizarre or exaggerated (Baltaxe & Simmons, 1985; Catherine Lord, Rutter, & Le Couteur, 1994). Atypical prosody contributes to the social communication difficulties experienced by autistic individuals (e.g., DePape, Chen,  4  Hall, & Trainor, 2012) and can have a negative impact on the quality of their social interactions, impeding the development of socio-communicative abilities.   Recent studies have tried to identify more precisely which prosodic and acoustic features guide the perception of atypical prosody. For example, Grossman (2015) examined first impressions of autistic and neurotypical children by neurotypical adults. More specifically, the author examined whether autistic children would be perceived as being more socially awkward than their neurotypical peers based on very brief exposure to static (still images) and dynamic information (audio-visual information). Blind to the diagnostic status of the children, the adult participants in this study were instructed to judge whether the person in the clip was socially awkward. Results show that typical adults evaluated autistic children as socially awkward significantly more frequently than their typical peers. These results remained consistent for both static and dynamic information. Typical adults seem to make use of various subtle cues in the prosody and/or facial expressions displayed by autistic children to form judgments of social awkwardness. Bone et al. (2015) sought to gain further insight into the perception of social awkwardness by focusing more specifically on the auditory modality. The authors examined awkward prosody by exploring the relationship between objective acoustic measures of prosody and subjective perceptions of prosodic awkwardness. Naïve raters, i.e., blind to the diagnostic status of the participants, were instructed to rate awkwardness along three prosodic components (rate/rhythm, volume and intonation/stress) as well as expressivity. The analysis of the subjective perceptions suggests that more awkward speech can be characterized as less expressive (more monotone) and more often involves perceived awkward rate/rhythm, volume, and intonation. The analysis of objective acoustic characteristics suggests that cues related to timing such as speaking rate and rhythm are highly predictive of perceived awkwardness. Using a classification task, the authors also show that acoustic-prosodic features can significantly distinguish autistic participants from neurotypical participants. Taken together, the results of this study expose a crucial point: naïve non-experts (and not only clinicians) are sensitive to features at a very fine-grain level of speech analysis, leading to perceptions of atypicality. Not only are naïve listeners able to detect and judge features related to atypical speech in autistic individuals, their future behaviors and attitudes towards these individuals can also be influenced by these features. This is especially problematic when the impression of the person’s speech is negative, as it can become a barrier to the social integration of autistic individuals. The potential repercussions of these judgments of awkwardness on the on-going  5  social interaction were highlighted in a study by Sasson et al. (2017), who explored the association between impressions of social awkwardness of autistic individuals and subsequent behavioral intentions towards them such as the likelihood to start a conversation with them. Across different experimental settings, the authors found that the first impressions of typical peers (adults and adolescents) of the social behaviors of autistic adolescents were more negative than those of their matched typical peers. Furthermore, these first impressions remained less favorable across different modalities (audio, visual and static images) and across short and longer examples of social behavior. These impressions also remained constant with repeated exposure to the same stimuli. Most importantly, negative first impressions of autistic adolescents were associated with reduced intentions to initiate or pursue social interaction with these individuals. In other words, neurotypical adolescents and adults were more reluctant to interact with autistic adolescents than other neurotypical peers. The authors suggest that this latter finding provides insight into a previously overlooked source of difficulty in the social interactions of autistic individuals. Furthermore, the authors propose that the social interaction difficulties experienced by autistic individuals are not only an individual impairment (specific to them), but also a relational one, viz. an impairment between them and their interaction partners. This conclusion echoes theoretical (re)conceptualizations of the socio-communicative difficulties in autism, as not only stemming from autistic people themselves, but also from the behaviors, perceptions and judgments of their conversational partners (Milton, 2012; Perkins, 2010; Sterponi & de Kirby, 2016).   Research Questions My dissertation focuses on language production of autistic adults with the specific objective of identifying linguistic characteristics contributing to negative impressions of spoken discourse of autistic adults. By linking concrete linguistic criteria to subjective impressions, we can address two pressing issues associated with the socio-communicative difficulties of autistic individuals: 1) define more accurately which aspects of language require remediation and 2) raise awareness of the biases autistic people may endure. I tackled these two issues in three steps. In a first step, described in Chapter 1, I created an annotated corpus of spoken data from 24 French-speaking autistic adults and 24 French-speaking neurotypical adults, matched on gender, age and IQ. The spoken data was collected during the Autism Diagnostic Observation Schedule, Second Edition (ADOS-2; Lord, Rutter, DiLavore, Risi, Gotham, & Bishop, 2012). My dissertation is based on a part of this corpus, viz. data from three tasks (six  6  and a half hours of speech in total). Chapter 1 provides an overview of the methodology of these three tasks, viz. data collection, transcription, segmentation and annotation. In a second step, I conducted two detailed transcript analyses of this corpus to identify linguistic features associated with the often-reported incoherence and atypicality of the speech of autistic adults (e.g., Baixauli et al., 2016; Stirling, Douglas, Leekam, & Carey, 2014). The first study on discourse production, presented in Chapter 2, targets more specifically the quality of (in)coherence, and investigates a widely studied discourse genre, viz. storybook narratives. On the one hand, this study was conducted to validate my new corpus as representative of language in ASD. On the other hand, the narrative study aimed at extending our knowledge of narrative ability in autism by systematically analyzing features from all three major narrative dimensions identified in previous studies. Chapter 2 provides a set of linguistic features typically associated with discourse (in)coherence in a new sample of storybook narratives. I hypothesized that autistic adults would perform less well on all three dimensions of narrative production, resulting in less coherent narratives. The second study on discourse production, presented in Chapter 3, targets the quality of (a)typicality by investigating different discourse strategies to deliver speech during a dyadic conversation on relationships and solitude. This study aims at addressing a methodological gap in transcript analyses, namely the lack of prosodic information, and thus advancing our understanding of spoken discourse in autism. To fulfill this aim, the transcriptions of six Experimenter-Autistic Speaker dyads and six Experimenter-Neurotypical Speaker dyads were segmented by modeling an innovative segmentation method which generates discourse units by combining syntactic and prosodic information (Degand & Simon, 2009a). This study provides a set of discourse units reflecting different strategies of managing and regulating information using syntax and prosody. I hypothesized that autistic adults would display more atypical strategies to convey speech than neurotypical adults. Finally, in a third step described in Chapter 4, the linguistic features identified in the two production studies were related to their perception by naïve listeners. A rating scale was developed to reflect the features identified in Chapter 2 and Chapter 3. To consider both the perspective of neurotypical adults and autistic adults themselves, I designed a rating experiment with a crossed-group design in which both autistic and neurotypical raters evaluated the speech of both autistic and neurotypical speakers. Of course, it is important to investigate whether naïve neurotypical individuals would be sensitive and influenced by the speech characteristics of autistic individuals. However, it is equally important to determine whether  7  autistic adults will attend to similar linguistic cues and interpret them in a similar fashion as neurotypical adults. If autistic and neurotypical people’s perception of discourse (in)coherence and (a)typicality is driven by different linguistic features, any attempt to remedy communication difficulties between these two groups of people will be futile as they do not use the same ‘language’ to begin with. Chapter 4 provides subjective ratings from two new groups of autistic and neurotypical participants of objective linguistic features produced by autistic and neurotypical individuals. I hypothesized that autistic speakers would be rated more negatively than neurotypical speakers resulting in more negative impressions of the speaker, implying there is a bias against them even before they are given a chance to take part in social interactions.   Outline of this dissertation This dissertation provides novel insights into the linguistic features distinguishing the speech of autistic and neurotypical adults and shows that the subjective impressions of these differences count against the autistic adults. My dissertation starts with the description of a new annotated corpus of spoken data from French-speaking autistic adults and neurotypical adults (Chapter 1). To confirm the validity of this corpus, I first investigated a widely studied discourse genre, viz. storybook narratives, confirming the typical cognitive and language profile of high-functioning autistic individuals (Chapter 2). I then investigated discourse skills beyond the narrative genre and provided a description of discourse strategies to manage information through semi-structured interview questions about relationships and solitude (Chapter 3). Finally, I related discourse features identified in Chapter 2 and Chapter 3 to their perception by naïve autistic and neurotypical adults (Chapter 4). A small chapter is dedicated to unexpected gender effects in the data of Chapter 3 and Chapter 4 (Chapter 5). Finally, Chapter 6 is devoted to a general discussion of the main findings of my dissertation, limitations and future research directions.     8  Chapter 1: Methodology  Data collection, transcription, annotation and segmentation  I. Corpus description  The first step of this dissertation was to create an annotated corpus of spoken data. In this chapter, I describe how I collected the data as well as how I prepared them for analysis, viz. how I transcribed, annotated and segmented the data. Speech samples were collected during the administration of the Autism Diagnostic Observation Schedule-2 (ADOS-2; Lord et al., 2012), a standardized instrument for diagnosing and assessing autism, by accredited ADOS assessors. The ADOS-2 includes 5 Modules, of which only one is administered to the individual, chosen on the basis of the individuals’ expressive language level and chronological age. The participants in the studies of this dissertation were administered either Module 3 (for verbally fluent children and young adolescents) or Module 4 (verbally fluent older adolescents and adults).  The ADOS-2 was not only used to validate the diagnosis of the autistic participants. More importantly, it was also considered a valid source of conversational data to examine the discourse competence of our participants, as Modules 3 and 4 are built almost entirely on conversation, and the cues of the different tasks approximate natural conversational situations. For example, interview questions covered topics of everyday conversations, such as work, hobbies or relationships. Furthermore, the semi-structured quality of these modules creates comparable situations across participants, giving them the same opportunities to demonstrate patterns of conversational strengths and weaknesses. The three studies of this dissertation are based on data from this corpus. In the following sections, I describe the procedure of data collection, preparation and annotation.   II. Methodology   2.1. Participants  The corpus comprises conversational data of 24 French-speaking autistic adults and 24 French-speaking neurotypical individuals, matched on chronological age, gender and Intellectual Quotient (IQ). Specifically, participants were pairwise matched on age (plus or minus one year)  9  and gender, and group matched on IQ. Age was calculated on the basis of the date of the first testing session and the participant’s birth date. IQ was assessed using the full version of the Wechsler Adult Intelligence Scale (WAIS-IV; Wechsler, 2008). The WAIS-IV is a widely used IQ test for adults and comprises ten subtests which yield four individual index scores of the major components of intelligence, viz. verbal comprehension, perceptual reasoning, working memory and processing speech, as well as a broader index score (Full-Scale IQ). Finally, the ADOS-2 was administered to both groups of participants to address two research objectives: 1) confirm the diagnosis of autistic participants and 2) collect comparable speech samples in the two groups. As can be seen from Table 1, the ADOS scores of participants in the autism group were significantly higher than those of the participants in the comparison group. Autistic and neurotypical participants did not differ in Full-scale IQ (FSIQ), Verbal Comprehension (VIQ) or Perceptual reasoning (PIQ). Finally, there was no group difference in chronological age.   Table 1 Descriptive statistics of participants' characteristics by group  ASD  NT t df p N (M:F) 16:8 16:8    Age (SD) 27.42 (11.63) 27.62 (11.47) -0.061 45.992 0.9516 Age range 15-52.09 15-53.04    ADOS Score 10.54 (3.27) 0.86 (1.36) 13.369 30.739 <. 0001 FIQ (SD) 107.13 (18.70) 107.96 (10.84) -0.189 36.896 0.8513 VIQ (SD) 111.42 (20.58) 111.33 (11.98) 0.017 36.99 0.9864 PIQ (SD) 107.96 (14.86) 103.54 (10.91) 1.174 42.226 0.2471  2.2. Procedure  Autistic participants were recruited via the Autism in Context: Theory and Experiment (ACTE) register of volunteers and (for some adolescent participants) in a secondary school specialized in autism. Participants in the comparison group were recruited via advertisements on the internet. The study was conducted in three sessions, either at the university or at participants’ homes. The Wechsler Adult Intelligence Scale and the Counting Stroop task were administered during the first session by the ACTE team’s neuropsychologist. The former was administered to verify that autistic participants met inclusion criteria of an IQ above 70 and subsequently served to match participants in the comparison group. The latter was administered to measure conflict resolution and inhibitory control abilities. During the second  10  session, participants completed four tasks on a computer. First, they performed two eye-tracking tasks, which are described in detail in Deliens, Papastamou, Ruytenbeek, Geelhand, & Kissine (2018). Participants then performed a computerized version of the Wisconsin Sorting Task in order to measure their cognitive flexibility and the Tower of London Task, to measure their planning ability. All computerized tasks were run in 64-bit Windows 7 with a 16.5 in. monitor (resolution: 1920 x 1080 pixels). Finally, the third session consisted of the administration of the ADOS-2. According to the ADOS-2 guidelines, the entire interview was video-taped, resulting in a total of 40 hours of interview (23 hours for the autistic group and 17 hours for the neurotypical group).  2.3. Material  In order to conduct more detailed analysis of discourse, the video recordings of the ADOS-2 were converted to audio files. For ease of analysis, the entire audio file was subdivided into the various tasks of the ADOS-2. Study 1 (Chapter 2) is based on the narrative task of the ADOS-2, namely Telling a Story from a Book. Study 2 (Chapter 3) and Study 3 (Chapter 4) are based on the semi-structured questions of the tasks Friendships, Relationships and Marriage and Solitude. The three tasks amount to a total of six and a half hours of transcribed, segmented and annotated data.   2.3.1. Narrative task: Telling a Story from a Book For the task Telling a Story from a Book, the 29-page picture book Tuesday by Wiesner (1991) was used to elicit a narrative from each participant. The book is considered wordless with only four sentences providing temporal indications (“Tuesday evening, around eight” at the beginning of the story; “11:21 pm” and “4:38 am” during the story; and “Next Tuesday, 7:58 pm). Tuesday is about frogs which are suddenly able to fly on their lily pads, and start exploring the neighborhood, surprising those still awake. However, as the sun rises, they start to lose their capacity to fly. Policemen find the water lily leaves lying on the streets and are confused. Next Tuesday, pigs are able to fly. The experimenter interviewing the participants introduced the task by saying: “This is a wordless picture book, I will start telling you the story, and you will finish telling it”. In practice, the first two pictures of the story were told by the experimenter and the remaining pictures were told by the participant. When necessary, the experimenter provided some encouragement to pursue the story with prompts in the form  11  of backchannelling (‘mhm’) or general comments (‘Tell me more’, ‘And here, what’s happening?’).  2.3.2. Semi-structured interview questions: Friendships, Relationships and Marriage and Solitude Following the ADOS-2 guidelines of these two tasks, the experimenter asked about the participant’s understanding of and opinion on the topics of friendship, relationship and marriage. Table 2 summarizes the different questions.  Table 2 ADOS-2 interview questions of the task Friendship, Relationships and Marriage and Solitude Friendship, Relationships and Marriage Do you have friends? Can you tell me more about them? What do you like to do together? How did you meet them? How often do you see each other? What does it mean for you “to be friends”? How do you know a person is your friend?  How do you distinguish between a friend and someone you only work with or go to school with? Do you have a girlfriend/boyfriend? What is his/her name? How old is he/she? When did you last see him/her? What is s/he like? What do you like doing together? How do you know he/she is your boyfriend/girlfriend? Where do you want to live when you are older? In what kind of accommodation (apartment, house, shared house)? Who would you like to live with? Your family? Roommates? Alone? Have you ever thought of having a long-term relationship, getting married (when you are older)? Why do you think some people get married or live with their partner when they grow up? In what ways is that a good thing? What could be difficult in being married or living with one’s partner? Or living with roommates? Solitude Do you sometimes feel lonely? Do you think other people of your age sometimes feel lonely? Are there things that you do to make you feel better? What are things that others might do to make themselves feel better when they feel lonely?     12  2.4. Data preparation  2.4.1. Transcription  All participants’ audio recordings were analyzed in Praat (Boersma & Weenink, 2017). First, the speech output of the participant and the experimenter was orthographically transcribed in a tier by trained master students in linguistics. During a first meeting, the general aim of the project and the guidelines of the orthographic transcriptions were explained verbally to the students. The students were then given a written version of the guidelines (see Appendix 1). They were instructed to read it at home and train on a few audios. They were also instructed to write their questions or comments in a Praat tier. A second meeting was then set up to make sure all transcription guidelines had been understood and clarify any potential misunderstandings. A point that often required clarification was the rendering of phonetic variations into the orthographic transcription. For example, short words like the personal pronouns ‘je’ (I) or ‘se’ (self), ‘je’, formally pronounced as [ʒə] and [sə] were often reduced phonetically to [ʃ] and [s]. Students’ misunderstanding lay in representing this phonetic reduction in the orthographic transcription by writing it out as ‘j’ or ‘s’. Transcribed orthographically in this form, it would mean that the participant produced [ʃi] and [es], the letters of the alphabet. The correct way to transcribe such cases is to orthographically transcribe these two words as ‘je’ and ‘se’ and render the phonetic reduction in the phonetic transcription as [ʃ] and [s]. Next, students were instructed to work independently but told they could ask questions at any time, if they encountered new difficulties. Figure 1 depicts a screenshot of a Praat TextGrid with an orthographic transcription in tier 1.      13  Figure 1 Screenshot of a Praat TextGrid with an orthographic transcription   Ten percent of all transcriptions were double-transcribed. Most disagreements were related to perceiving different words or rendering phonetic variation in the orthographic transcription. In the case of the former, the disagreement was resolved by instructing both coders to listening to the audio again and determining together which of the two words is the most probable to have been said. In the case of the latter, disagreements were resolved by referring to the instructions of the transcription protocol, viz. the guidelines instructed clearly not to invent new written forms to simulate pronunciations and to follow the orthographic norms of French. In a second step, the Praat plug-in EasyAlign (Goldman, 2011)a semi-automatic speech alignment tool, was used to create three new tiers on the basis of the orthographic transcription. First, the phonetization function was applied, which duplicates the tier with the orthographic transcription in a new tier and converts automatically the transcription into a phonetic one (using the SAMPA alphabet2). This transcription was then manually checked to correct any transcription errors or pronunciation errors. Figure 2 depicts a screenshot of a Praat TextGrid with an orthographic transcription in tier 2 and its phonetic counterpart in tier 1.   2 For a description of the SAMPA alphabet, see the EasyAlign tutorial : http://latlcui.unige.ch/phonetique/easyalign/tutorial_easyAlign_english.pdf  14  Figure 2 Screenshot of a Praat TextGrid with a phonetic transcription   Using the phone segmentation function, the orthographic transcription was segmented into words and the phonetic transcription was segmented into syllables. To detect the temporal boundaries of syllables and words, the phone segmentation function uses the speech recognition engine HTK (HMM Tool Kit, Goldman, 2011)). The outcome of this procedure was a TextGrid file for each participant’s audio recording with four tiers: 1) an orthographic transcription, 2) a phonetic transcription, 3) a word segmentation and 4) a syllable segmentation. See Figure 3 for an example of a TextGrid with all four tiers of speech transcription and segmentation.     15  Figure 3 Screenshot of a Praat TextGrid with an orthographic and phonetic transcription as well as words and syllable segmentation   Figure 4 below depicts the different steps of the data preparation process, viz. orthographic transcription, phonetic transcription, word segmentation and syllable segmentation.    16  Figure 4 Flowchart of the data preparation process        Audio of narrative production Manual transcription by trained master students in linguistics.   Orthographic transcription ortho tier Phonetic transcription Semi-automatic transcription with the phonetization function in Praat phono tier Manual checking of phonetic transcription Word segmentation Semi-automatic segmentation with the phone function in Praat Syllable segmentation Manual checking of word and syllable segmentations word tier syllable tier 1) Ortho tier 2) Phono tier 3) Word tier 4) Syll tier Final TextGrid  17  2.4.2. Segmentation 2.4.2.1. Theoretical rationale Segmentation units are crucial to explore how discourse is realized and organized. Deciding how to segment discourse, viz. which linguistic criteria will be used to ‘chunk’ the discourse into smaller units, is not a trivial question. In fact, many researchers have sought to define what are the nature and role of these units in the unfolding discourse (e.g., Chafe, 1987; Degand & Simon, 2005; Schegloff, 2000; Selting, 2000). In this dissertation, I modeled the approach of speech segmentation in Basic Discourse Units (BDUs) developed by Liesbeth Degand and Anne-Catherine Simon (Degand & Simon, 2005, 2009a, 2009b). Specifically, their segmentation model takes on the paradox of spoken discourse by combining linguistic features that represent discourse as a product, viz. a syntactically structured and organized object as well as discourse as a process, viz. a temporal and dynamic object. The former gives priority to syntax, while the latter gives priority to prosody. Although syntax and prosody can be conceived as two distinct levels of surface segmentation, they are also interdependent and come together at a distinct organizational level (Degand & Simon, 2009a, 2009b). Specifically, a BDU originates when there is a match between a syntactic boundary and a prosodic boundary. Considering the known prosodic impairments in ASD (e.g., Diehl, Watson, Bennetto, McDonough, & Gunlogson, 2009; Paul, Augustyn, Klin, & Volkmar, 2005; Peppé, McCann, Gibbon, O’Hare, & Rutherford, 2006), it is important to incorporate this dimension when representing speech, as their communication difficulties might not only be restricted to the syntactic organization of discourse content, but in their prosodic strategies to deliver speech. The segmentation method developed by Degand and Simon seemed particularly suitable to gain insight into these difficulties. In practical terms, the segmentation of speech into BDUs results from a segmentation of the transcript and its corresponding audio in two independent phases, a syntactic and a prosodic one. The syntactic segmentation was performed manually following the coding protocol developed by Tanguy, Van Damme, Degand, & Simon (2012). The prosodic segmentation was also performed manually based on perception. I did not include features identified semi-automatically by Prosogram, a software performing semi-automatic transcriptions of prosody (Mertens, 2004). In this respect, my prosodic segmentation did not include the full criteria proposed by Degand and colleagues and thus cannot be compared directly to theirs. In the following paragraphs, I will describe both segmentation procedures,  18  highlighting when necessary any divergences from the segmentation procedure by Degand and colleagues.   2.4.2.2. Segmentation procedure Both segmentations were performed by master students in linguistics. During a first meeting, the general aim of the project and the segmentation procedure was first explained verbally. The students were then given a written version of the segmentation protocols and instructed to read them at home. A second meeting was then organized to answer any questions based on the reading. After clarification of any potential misunderstandings, the students were instructed to code a few training audio recordings to familiarize themselves with Praat and the coding guidelines. A third meeting was organized to go over the coding and resolve any ambiguities or coding difficulties. Next, the students were instructed to work independently but told they could ask questions at any time, if they encountered new difficulties. Ten percent of all segmentation was double-coded. For clarity reasons, the guidelines for the syntactic segmentations are firs laid out (Section 2.4.2.3.), then coding difficulties and resolution of coding disagreement related to the segmentation guidelines are discussed in Section 2.4.2.5.   2.4.2.3. Guidelines for the syntactic segmentation Dependency clauses The coding protocol developed by Tanguy et al. (2012) relies on the theoretical principles of dependency syntax (e.g., Blanche-Benveniste et al., 1990). Specifically, orthographic transcriptions were first segmented into dependency clauses, which are subsequently segmented into smaller sequences. There are three types of dependency clauses: 1) verbal (organized around a tensed verb), 2) averbal (organized around an averbal constituent such as a noun, a pronoun or an adverb) and 3) elliptical (units that are incomplete but can be interpreted as a verbal dependency clause on the basis of the context). Following the annotation convention of Tanguy et al. (2012), dependency clauses are represented between squared brackets [ ]. Table 3 summarizes the types of dependency clauses and provides examples.    19  Table 3 Corpus examples of dependency clause subtypes Type Example verbal dependency clause [elles volent loin de leur marécage en direction d’une d’un petit village] they fly far away from their swamp in the direction of a small village Autistic participant (female, 22 years old) averbal dependency clause  [d’accord] all right Comparison participant (female, 18 years old) elliptic dependency clause [on retrouve le chat de chez mémé] <euh> [le chien] we rediscover the cat from granny uh the dog Autistic participant (female, 34 years old)  For each type of dependency clauses, if at least one of its obligatory complements was missing, it was coded as incomplete. Below is an example of an incomplete verbal dependency clause as the complement object of the main clause (I assume it’s) is missing.   (1)  [je suppose que c’est] I assume it’s Autistic participant (male, 32 years old)  Sequences  Verbal dependency clauses were further segmented into functional sequences, while averbal dependency clauses were further segmented into categorical sequences. Depending on their function, elliptical clauses were either segmented into functional or categorical sequences. Following the annotation convention of Tanguy et al. (2012) sequences are represented between parentheses ( ).  Functional Sequences Functional sequences were segmented according to Bilger & Campione (2002). They define a functional sequence as an utmost sequence representing functional constituents without taking into account the details of its composition. Verbal and elliptical dependency clauses were segmented into functional sequences. There are four types of functional sequences: subject  20  sequence, verbal sequence, object sequence and governed sequence. Table 4 below provides examples of the different categories. - Subject sequence A subject will be coded as a subject sequence if it appears as a full lexical form. Hence, clitic subjects are not coded as a subject sequence and are integrated into the verbal sequence.  - Verbal sequence Verbal sequences include the verb and its different clitic and pronominalized agents. Verbal sequences can also include the following elements: verbal markers of negation and restriction, modals, infinitives selected by the verb, certain forms related to the verb (fixed or not) adjectival attributes and indefinite nominal attributes. - Object sequence Objects are coded as independent object sequences when they appear in a full lexical form. Furthermore, object sequences include the following elements: direct objects, indirect objects, such as locatives, agents, attributes of nominal subjects (definite and indefinite nominal sequence) and impersonal constructions. - Governed sequence A governed sequence is governed by the verb but is not part of the verbal valency, viz. it is not syntactically required by the verb. Governed sequences were further specified whether they appeared on the right or left side of the verbal sequence.     21  Table 4 Corpus examples of functional sequences subtypes Functional Sequence Example Subject sequence (les crapauds) (continuent à s’élever) (the toads) (continue to rise) Autistic participant (male, 19 years old) Verbal sequence <et> <du coup> (elles volent) <and> <so> (they fly) Comparison participant (female, 18 years old) Object sequence (elles visitent) (les lieux) (they visit) (the place) Comparison participant (female, 26 years old) Governed sequence (ils entraient) (par les fenêtres) (par les cheminées) (they enter) (through the windows) (through the chimneys) Autistic participant (male, 23 years old)  Categorical sequences Averbal dependency clauses cannot be ‘functionalized’ and were therefore not segmented into functional sequences but into categorical sequences. Table 5 below summarizes the types of categorical sequences with examples.    22  Table 5 Corpus examples of categorical sequences subtypes Categorical Sequence Example Nominal sequence (petit dejeuner) (breakfast) Autistic participant (female, 34 years old) Pronominal sequence <et> (toi) <and> (you) Autistic participant (male, 19 years old) Adjectival sequence (parfait) (perfect) Experimenter 2 (female, 23 years old) Adverbial sequence (oui) (yes) Autistic participant (male, 20 years old) Prepositional sequence (à toi) (your turn) Experimenter 2 (female, 23 years old) Subordinated sequence (alors) (qu’une personne mangeait une tartine et un et un buvait un verre de lait) (while a person was eating a sandwich and drinking a and a glass of milk) Comparison participant (male, 24 years old) Interjective sequence <et> (hop) <and> (hop) Comparison participant (female, 20 years old)  Participial sequence (bien collé) (nicely stuck) Comparison participant (male, 24 years old)  Additional coding categories In addition to dependency clauses, adjuncts and discourse markers were coded. Adjuncts are elements which are not governed by the verbal head but are nevertheless associated to the verb. These elements are located on the periphery of verbal dependency clauses. Similar to the coding of governed sequences, the adjunct was specified according to whether it appeared on the left or right of a dependency clause. Discourse markers fell into the following categories: 1) connectives, 2) conjunctions and complementizers and 3) discourse markers in  23  a narrower sense, i.e., lexemes that serve a structuring or meta-discursive function (e.g., ‘bah’(well)).  Considering both the heterogeneity of terms used to refer to the linguistic phenomenon of ‘discourse markers’ and the heterogeneity of linguistic forms that can be included under this label (Crible, 2017; Maschler & Schiffrin, 2015), I will follow the Textlink3 network’s advice to use the single and more general label ‘discourse-structuring devices’ (Crible, 2017). This terminological choice allows for more clarity and flexibility when referring to subtypes of discourse-structuring devices. Specifically, the annotation protocol of the narrative study codes for two specific types of discourse-structuring devices: connectives and discourse markers. On the one hand, the label ‘connectives’ emphasizes the ‘connecting’ function of the linguistic device, viz. relating two utterances in a specific manner. In the narrative study, four connectives will be investigated, viz. additive (e.g., and), temporal (e.g., then), causal (e.g., because) and contrastive (e.g., but). On the other hand, the label ‘discourse markers’ is used refer to the linguistic devices that serve to negotiate relations between speaker and hearer and/or between speaker and text (Maschler & Schiffrin, 2015). Certain silent pauses, hesitation markers and paraverbal elements were also segmented. Silent pauses lasting longer than 250 milliseconds (ms) were coded as a separate segment. If a silent pause lasted less than 250 ms, it was attached to the next coded segment. Filled pauses such as ‘euh’ (uh) and ‘euhm’ (uhm) were also coded as independent segments if they appeared between dependency clauses. Following the annotation convention of Tanguy et al. (2012), adjuncts, discourse-structuring devices and hesitation markers are represented between angle brackets < >. See Table 6 for examples.     3 TextLink is a portal that aims at bringing together linguistic resources on discourse structure: http://textlink.ii.metu.edu.tr/  24  Table 6 Corpus examples of adjuncts, discourse-structuring devices, hesitation markers and silent pauses Type Example Adjunct <alors> <évidemment> [à l’intérieur y’a des personnes] <so> <of course> [inside there are people] Comparison participant(male, 52 years old) Discourse-structuring devices  <bah> [ils ont ils ont l’air de basculer dans leur truc] <well> [they look like they’re flipping into their thing] Autistic participant (female, 27 years old) Hesitation markers  [elles s’accaparent les objets des gens] <euhm> [elles s’abrutissent devant la télé] [they snatch the objects of people] <uhm> [they become numb in front of the tv] Autistic participant (female, 34 years old) Silent pauses  <donc> [elles utilisent les feuilles pour faire un peu pareil] (1.43) <et> <du coup>[elles] <euh> (1.12) <et>  <du coup> [elles volent] <so> [they use the leaves to do a bit the same] (1.43) <and> <so> [they] <uh> (1.12) <and> <so> [they fly] Comparison participant (female, 26 years old)  2.4.2.4. Guidelines for the prosodic segmentation The prosodic segmentation is based on the segmentation in syllables and involves the identification of major intonation boundaries, as well as the assignment of an intonation contour to each of these boundaries. The identification of a major intonation boundary was manual and based on perception according to the summary guidelines of Simon & Christodoulides (personal communication). In this dissertation, a major intonation boundary was identified according to two criteria: 1) end of speaker turn and 2) silent pauses, viz. when final syllable was followed by a silent pause of more than 200 ms, not preceded by a hesitation marker. It is important to note that Degand and colleagues performed a more detailed prosodic segmentation, relying also on acoustic correlates of major intonation boundaries identified semi-automatically with the software Prosogram (Mertens, 2004) in Praat (Boersma & Weenink, 2017). Hence, the prosodic segmentation applied in this dissertation is a simplified segmentation.  Following the annotation convention used by Degand and Simon, major intonation boundaries are represented as three backward slashes ///. Each boundary is followed by a letter representing the type of contour carried by the last syllable of the unit. The letter T indicates  25  a final contour (descending), the letter C indicates a continuing-rising contour (intrasyllabic rise or positive interval with respect to the preceding syllable), the letter S indicates a flat contour (sometimes lengthened, with no pitch difference with respect to the preceding syllable) and the letter F indicates a focalization contour (descending moving departing from a high target).  Figure 5 below depicts a TextGrid with the segmentation units of the syntactic and prosodic segmentation. Tier 5 contains the coding of dependency clauses and Tier 6 contains the coding of sequences. Tier 7 contains the coding of major prosodic boundaries and Tier 8 contains the contours.   Figure 5 Screenshot of a Praat TextGrid with all transcription and segmentation tiers   2.4.2.5. Coding difficulties and disagreement When familiarizing themselves with the segmentation procedure, the students experienced coding difficulties due to overlapping speech. Overlapping speech is a common phenomenon during social interaction but was not analyzed in this dissertation. Therefore, for reasons of convenience and efficiency, the students were instructed to delimit the overlapping section with boundaries in the dependency and/or sequence tier and to code it with the symbol ‘%’. Similarly, in a very few cases, speech had not been transcribed as it was incomprehensible and  26  had been coded as ‘xxx’ in the orthographic transcriptions. In these cases, the students were also instructed to code these parts as ‘%’.4 The majority of coding disagreements resulted from coding ambiguities between the following categories: object sequence, governed sequence or adjunct. Ambiguity between the coding of governed sequences and adjuncts was addressed by applying the two tests put forward by Tanguy et al. (2012, p.13) to distinguish between these two categories, viz. test of pronominalization and extraction. The following example taken from Tanguy et al. (2012) illustrates these two tests.  (2) [(dans le voyage au bout de la nuit)SRg (il commence)SV (une ligne)SO (en disant)SRd]urv [(in the journey to the end of the night)SRg (he starts)SV (a line)SO (by saying)SRd]urv Extraction test :  c’est dans le voyage au bout de la nuit qu’il commence une ligne en disant it is in a journey to the end of the night that he starts a line by saying  Pronominalisation test :  il y commence une ligne en disant he starts there a line by saying  Coding disagreements resulting from ambiguity between a governed sequence and an object sequence of location was resolved through discussion between the two coders. For example, in the example (3) the sequence ‘vers une petite maison située dans la zone urbaine’ was coded by coder one as a governed sequence and by coder two as an object sequence. The first coder argued for a coding as a governed sequence because it could be subject to the extraction rule ‘après avoir franchi les oiseaux c’est vers une petite maison située dans la zone urbaine que les grenouilles se dirigèrent’. The second coder argued for an object sequence because the verb ‘se diriger’ is transitive and thus requires a direct object. Supporting, this argument is the manual’s definition of a governed sequence that it is governed by the verb but not part of the verb’s valence. On basis of the arguments of coder 2 and the manual’s definition,  4 For the narrative task, instances coded as % amounted to 2 percent of the for the autism group and 1 percent for the neurotypical group. For the open-questions about relationships and solitude, instances coded as % amounted to less than 1 percent for the autism and neurotypical group (0.002 and 0.004, respectively).  27  the disagreement was resolved by deciding on the coding of a sequence following the verb ‘se diriger’ as an object sequence.  (3) [(après avoir franchi les oiseaux)SRg (les grenouilles)SS (se dirigèrent)SV (vers une petite maison située (dans la zone urbaine)SRd/SO]urv  [(after passing the birds) SRg (the frogs)SS (headed)SV (for a small house in the urban area)SRd/SO]urv  There were also instances of disagreement due to ‘errors’ (e.g., inattention, forgetting the guidelines of the protocol). For example, one such mistake was to code a sequence of discourse-structuring devices (and then so) as one unit (example 2), whereas the protocol specifically specifies to code each discourse-structuring device separately (example 3).   (4) <et puis donc>md [le linge se transforme en capes pour euh certaines grenouilles]urv <and then so>md [the cloth transforms into capes for uh some frogs]urv (5) <et>md<puis>md<donc>md [le linge se transforme en capes pour euh certaines grenouilles]urv <and>md <then>md <so>md [the cloth transforms into capes for uh some frogs]urv Such disagreements were easily solved by referring back to the explicit guidelines of the coding protocol. These instances of disagreement were very rare, as the protocol was thoroughly explained, and mistakes were addressed and corrected during the second and third meeting (see Section 2.4.2.2).  2.4.2.6. Basic Discourse Units The identification of Basic Discourse Units (BDUs) is performed by matching the syntactic and prosodic segmentations. A BDU is identified each time the boundary of a dependency clause coincides with a major intonation boundary. Degand & Simon (2009a) identified four different ways boundaries coincide. Table 7 summarizes the different types of BDUs.     28  Table 7 Summary of the different types of Basic Discourse Units BDU Description Congruent BDU one dependency clause = one major prosodic unit One-to-one mapping of syntactic and prosodic unit  Syntax-bound BDU several major prosodic units = one dependency clause One-to-many mapping, i.e., one syntactic unit produced through several successive prosodic units  Intonation-bound BDU several dependency clauses = one major prosodic unit Many-to-one mapping, i.e., several syntactic units are grouped into one major prosodic unit  Regulatory BDU Adjunct/discourse marker = one major prosodic unit A non-governed element is ‘autonomized’ in a major prosodic unit Mixed BDU Mismatch between syntactic and prosodic boundaries.  In this dissertation, the major intonation boundaries corresponded to a silent pause of more than 200 ms, not preceded by a hesitation marker and my syntactic boundaries corresponded to the boundaries of a syntactic clause. Furthermore, I identified BDUs manually, by extracting the TextGrid tiers of the syntactic and prosodic segmentations from Praat into an excel sheet, while Degand and colleagues match the BDUs directly in Praat according to a script. While the same guidelines were applied in the current procedure, it is not a direct replication of their method. Accordingly, the BDUs of my dissertation cannot be compared directly to those identified by Degand & Simon (2008, 2009a, 2009b). The implications of this methodological difference on the results of the BDU distribution are discussed in more details in Chapter 3.     29  Chapter 2: Narrative production in autistic adults  A systematic analysis of the microstructure, macrostructure and internal state language  I. Introduction  The second step of this dissertation is to identify the linguistic features that distinguish the speech of autistic and neurotypical adults. Language in Autism Spectrum Disorder (ASD) is characterized by very heterogeneous linguistic profiles, ranging from individuals who will never develop functional speech and will remain non-verbal to individuals who will acquire average or even above average verbal skills (e.g., Eigsti, de Marchena, Schuh, & Kelley, 2011; Tager-Flusberg, 2000). At the same time, these hugely variable linguistic profiles consistently share difficulties related to the domain of pragmatics which continue to be present even in individuals who have achieved average syntactic, lexical and phonological skills as well as average IQ (Volden, Coolican, Garon, White, & Bryson, 2009). This suggests two different areas of difficulties in verbal autistic individuals. On the one hand, the communication difficulties they experience are (mostly) associated with the use of language rather than with the structural linguistic properties. On the other hand, their communication difficulties are not restricted to the linguistic domain but also surface in the social and cognitive domains. As already mentioned in the introduction, discourse analysis proved a useful theoretical and methodological framework to study language use in ASD. The following section discusses in more detail narratives, a type of discourse that has been extensively studied in autism research; I also outline the main features that seem to characterize narrative discourse in autism in the existing literature.   1.1. Narrative discourse as a measure of language use in autism  Narrative can be broadly defined as a discourse genre in which a series of events are presented as unfolding over time, and are related to each other (Stirling et al., 2014). The production of a coherent narrative relies on various cognitive and linguistic skills. For example, stories typically have a reliable and consistent structure. Hence, storytelling will require the speaker to adhere to a story schema of the essential story components, such as setting, initiating event, consequences, resolution and ending (Hughes, McGillivray & Schmidek, 1997). Failing to  30  include these essential story elements will usually affect the coherence of the overall organization. Coherence can be also be achieved through the use of linguistic devices, which create cohesive ties within the narrative structure. For instance, ties can be established between story events with temporal (e.g., then, after) or causal (e.g., because, therefore) connectives. Ties can also be established through references chains, viz. by using different referential expressions to reidentify a story protagonist across the narrative (e.g., definite expressions such as ‘the boy’ or a pronoun ‘he’). Further contributing to the coherence of a narrative is the ability to express the point of view of the different story characters by verbalizing their internal states. This is reflected by the inclusion of linguistic devices to communicate about feelings, desires, beliefs, intentions and other internal states (Stirling et al., 2014). Linguistic terms that refer to the internal and mental states of the story protagonists or narrator are usually referred to as ‘internal state language’ in the literature (Bretherton & Beeghly, 1982).  The following excerpt is taken from the narrative of a neurotypical participant in my corpus, exemplifying how coherence can be achieved using all the aforementioned cognitive and linguistic devices. Referential expressions are highlighted in orange, connectives are highlighted in purple and internal state language is highlighted in green.   (1) […] la tortue je pense que ce qu’elle a vu c’est euh les crapauds volants (1.89) et du coup elle semble avoir quand même euh un peu peur et j’ai l’impression qu’elle est en train d’essayer de rentrer dans sa (0.84) dans sa carapace (0.44) et même les poissons les ont vus parce qu’ils ont la tête en l’air […] […] the turtle I think that what she saw is uh the flying toads (1.89) and so she seems to be indeed uh frightened and I’m under the impression that she is trying to retract in her (0.84) shell (0.44) and even the fish saw them because they’re looking at the sky […]  Comparison participant5 (female, 39 years old)  In this excerpt, the participant initiates the narrative’s structure by verbalizing one of the main events of the story, viz. a turtle sees flying toads. At this point of the narrative, the experimenter had just introduced the character of the turtle (with an indefinite expression, a turtle) and the participant carries on with the story by keeping the referential link to the story  5 The examples included in this dissertation come from my corpus. The comparison group was neurotypical participants.  31  protagonists by re-introducing the turtle with a definite expression (the turtle) and subsequently maintains reference by using a pronoun (she). Relations between the utterances describing the events are established by using the causal connectives therefore and because, as well as the additive connective and. Finally, the participant also refers to the internal states of the turtle by referring to emotions (fear) and intentions (is trying to). Producing a coherent narrative is therefore a complex task, which draws on multiple linguistic (e.g., grammar and lexicon), cognitive (e.g., deriving a story structure) and social skills (e.g., monitoring listeners’ knowledge and interest) (De Marchena & Eigsti, 2016; Diehl, Bennetto, & Young, 2006; Volden et al., 2017). For this reason, narrative production is a particularly suitable measure of language use in autistic individuals who continue to experience communication difficulties, despite performing well on standardized tests of structural language skills  (Manolitsi & Botting, 2011).  Furthermore, narratives can provide grounds for relating communication difficulties with deficits in cognitive mechanisms, such as deficits in Theory of Mind abilities (Baron-Cohen, Leslie, & Frith, 1985). For example, Siller, Swanson, Serlin, & Teachworth (2014) examined ToM abilities by looking at the use of internal state language, viz. the inclusion (or not) of terms reflecting the psychological states of the story protagonists, in the storybook narratives of 5-8-year-old autistic children and neurotypical controls, as well as their performance on a series of ToM tasks. The authors found a difference in emotional terms, with autistic participants producing significantly fewer emotion terms than their neurotypical peers. Furthermore, they found a significant relationship between emotional state descriptions and ToM abilities (as measured by a composite score of the different ToM tasks), which could not be attributed to other group differences such as differences in narrative length or variables such as chronological age and language abilities. The authors conclude that difficulties with inferring other’s internal states can at least partially explain the poor narrative skills of autistic children.  Narratives also provide a context in which Weak Central Coherence (Frith & Happé, 1994) can be tested by examining participants’ ability to relate a coherent global structure and content (e.g., organizing the story structure around the gist of the story rather than secondary details). For example, in a story recall task, Diehl et al. (2006) found that high-functioning autistic children performed like their typical peers when remembering story gist and were equally aware of the importance of gist story events. Nevertheless, they produced less coherent narratives because they were less likely to use these gist events to organize their narrative. Such difficulty with combining pieces of information into a coherent whole suggests weak central coherence (Happé & Frith, 2006). Finally, difficulties in piecing together a  32  coherent whole can also suggest more general difficulties with planning and organization (executive functioning, Ozonoff, Pennington, & Rogers, 1991).  Narrative production also presents several methodological advantages. It is a particularly useful tool for the elicitation of sequences of utterances from a relatively naturalistic and fluent speech sample (Stirling et al., 2014). On the basis of audio recordings of the participant’s performance, narratives can be transcribed verbatim, converting them into a format that can be used for subsequent assessment. Specifically, the transcripts can then be coded for different features of interest such as use of reference, inclusion of story grammar elements, etc. Another methodological advantage is that narrative elicitation tasks can be structured in a way that allows collection of comparable speech samples across participants (Stirling et al., 2014). For example, wordless picture books are often used to elicit narratives, with all participants seeing the same pictures and having to interpret them.  1.2. Characteristic of narrative discourse in autism  Considering the potential of narrative production to provide unique insight into communication in autism, a large number of studies have been published on narrative discourse in autism. Three major dimensions of narrative production have been investigated in these studies: (i) the microstructure, (ii) the macrostructure and (iii) internal state language (Baixauli et al., 2016). The most important discursive difficulties experienced by autistic individuals are situated at the level of the macrostructure; autistic participants seem less capable to maintain global coherence in their narratives relative to their neurotypical peers. These difficulties surface especially as a reduced use of cohesive linguistic devices such as referential links (Norbury & Bishop, 2003) and causal conjunctions, as well as poor organization of the story around the gist events (Capps, Losh, & Thurber, 2000; Diehl et al., 2006; Losh & Capps, 2003; Tager-Flusberg, 1995). Regarding the microstructure, results remain inconclusive with only some studies finding group differences. For example, some authors report that autistic children  produced shorter and syntactically less complex narratives than comparison groups (Capps et al., 2000; Tager-Flusberg, 1995), while others find no such group difference (Losh & Capps, 2003; Norbury & Bishop, 2003). However, in their meta-analysis, Baixauli et al. (2016) suggest worse performance of autistic children and adolescents on productivity (number of words and utterances), lexical diversity (number of different words) and syntactic complexity (mean length of utterance). Hence, at least some autistic children and adolescents (viz. those  33  represented in the meta-analysis) do seem to display delayed development in at least some features of morphosyntactic acquisition (Park, Yelland, Taffe, & Gray, 2012).  Finally, regarding variables related to internal state language, autistic individuals include significantly fewer mental state terms than control groups (Baixauli et al., 2016). Baixauli et al. (2016) found that autistic participants with no IQ delay seem to have more difficulties in recognizing and expressing internal states than their neurotypical peers. Although the opposite tendency might be expected, the authors suggest that cognitive and linguistic elements associated with socioemotional meaning such as mental state terms might follow a particular developmental path, resulting in an uneven development of these skills relative to the development of other cognitive and linguistic skills (such as syntactic skills).  1.3. Age and narrative performance  Most studies on narrative production in ASD have targeted children and young adolescents. In their meta-analysis, Baixauli et al. (2016) did not find chronological age to be a significant moderator of narrative competence. However, the age range of participants in all the studies these authors reviewed did not extend beyond fifteen, so it is not clear whether older adolescents and adults continue to display difficulties in narrative production or whether they would perform similarly to their neurotypical peers. The few studies that did focus on storytelling in autistic adults report results that are quite similar to those found for children, suggesting that autistic adults continue to experience difficulties in narrative production. That said, results on narrative length and complexity in adults are as inconsistent as those in the literature on children. For example, when narrative production was supported by sequences of pictures (Beaumont & Newcombe, 2006) or a wordless picture book (Colle, Baron-Cohen, Wheelwright, & Van Der Lely, 2008), no group differences surfaced in narrative length and syntactic complexity. However, using a more complex task (the Social Attribution Task designed by Heier and Simmel (1944) with dynamic stimuli, Klin (2000) did find that autistic adults produced shorter narratives than their typical peers.  Regarding the overall quality of their narratives, autistic adults do seem to display similar profiles of strength and weaknesses to autistic children. For example, while autistic adults have been found to produce internal state language in their narratives with the same frequency as their typical peers, they seemed to be less able than their typical peers to provide causal explanations for the mental states of the protagonist (Beaumont & Newcombe, 2006,  34  Colle et al. 2008). Older autistic individuals also continue to show subtle difficulties with referential cohesion. For example, Colle et al. (2008) report that while the total number of referential expressions dedicated to the maintenance of reference does not differ per group, the type of expression does. The two groups also differed in their use of referential expressions to maintain reference. Specifically, neurotypical participants used pronominal expressions more often than autistic participants to maintain reference to a character that was being talked about. This had the effect of a more cohesive and forward-moving story than if full NPs were used. In contrast, autistic participants used more full NPs and less pronominal expressions to maintain reference. In other words, it is not the case that adult autistic participants could not maintain reference across a narrative, but they did so in a less efficient way, suggesting subtle but persistent pragmatic difficulties.   Outlook To sum up the discussion so far, narrative production has been evaluated overwhelmingly in children and young adolescents. While narrative abilities appear early in development, they continue to develop well into adulthood (e.g., Bamberg, 1997). During adulthood, expressing oneself with clarity, precision and efficiency is crucial for success in various circumstances, from personal (e.g., building relationships) to educational (e.g., attending university) and employment (e.g., securing employment) settings (Nippold, Frantz-Kaspar, & Vigeland, 2017). Studies on neurotypical adults have suggested that language samples, such as narrative production, can serve as a good measure of spoken language proficiency (Nippold et al., 2017). However, to this end one should systematically include measures of microstructure, macrostructure and internal state language, which was not the case in most of the existing studies on narrative production. Furthermore, adding analyses of narrative production from adults to the existing literature on children and young adolescents will provide a more comprehensive overview of narrative development, which could then be used to determine areas requiring language intervention.  The overarching aim of the present study is to contribute to a clearer linguistic and communicative profile of ASD in adulthood by providing a systematic description of narrative performance in autistic adults with linguistic and cognitive functioning in a typical range. A specific annotation scheme was developed to code narrative production, in order to be able to compare the production of autistic participants to pairwise matched neurotypical adults relative to microstructure (syntactic complexity), macrostructure (overall story structure and  35  cohesive ties) and internal state language of the corpus’ narratives. I hypothesized that older autistic adolescents and adults would perform worse than their neurotypical peers on all three dimensions of narrative production, resulting in less coherent narratives overall for autistic individuals in comparison to neurotypical individuals.  II. Methodology  This study received ethical clearance from the Ethics Committee of the Faculty of Psychology and Education at Université libre de Bruxelles and the Behavioural Research Ethics Board of the University of British Columbia. Written consent was obtained from all participants or their parents.  2.1. Participants  From the initial 48 participants of our corpus, data from 18 autistic participants (ASD) and 18 neurotypical participants (NT) was available for narrative analysis6. Inclusion criteria for both groups included: 1) age between 15 and 60 years, 2) a Full-Scale IQ (FIQ) score above 70, 3) Verbal IQ (VIQ) score above 70 and 4) normal or corrected-to-normal vision and audition. For the control group, there was the extra inclusion criterion of no known psychiatric, developmental or neurological disorder. Participants were pairwise matched on age (plus or minus one year) and gender. Autistic participants had previously obtained a clinical diagnosis of autism from a multi-disciplinary team assessment external to our research group, based on criteria of the Autism Diagnostic Observation Schedule 2 (ADOS-2; Lord et al., 2012) and the Autism Diagnostic Interview-Revised (ADI-R; LeCouteur et al., 2003). For our study, clinical diagnosis of ASD was confirmed for all participants by a research-accredited ADOS assessor using Module 3 or 4 of the ADOS-2 (Lord et al., 2012)(C Lord, Rutter, DiLavore, Risi, & Gotham, 2012). Neurotypical participants were also administered Module 3 or 4 of the ADOS-2 and all scored below the autism cut-off.  All participants received and signed an informed consent form, which included an authorization to be filmed during the ADOS-2. Furthermore, as advised by Baron-Cohen et al., the Empathy Quotient (EQ; Baron-Cohen & Wheelwright, 2004) was administered  6 Due to technical errors or experimental errors, we had to reject the data of 6 participants with ASD. When data were not available for a participant with ASD, we also excluded from the analyses data for the corresponding, pairwise matched NT participant.   36  conjointly with the Autism Quotient (AQ; Baron-Cohen, Wheelwright, Skinner, Martin, & Clubley, 2001), which provides an estimate of autistic-like traits presented by an individual, and allows for them to be situated on the continuum from autism to neuro-typicality. As can be seen from Table 8, the ADOS-2 and AQ scores of participants in the autism group were significantly higher than those of the participants in the comparison group. Participants in the comparison group scored significantly higher on the EQ.  Table 8 Descriptive statistics of participants’ characteristics per diagnostic group  ASD  NT t df p N (M:F) 18 (11:7) 18 (11:7)    Age (SD) Age-range 28.90 (11.80) 15.2 - 52.9 28.79 (11.84) 15.0 -53.04 0.028 34 0.98 ADOS Total Score (SD) 10.06 (2.86) 0.88 (1.32) 12.35 23.97 <.0001 AQ (SD) 35.94 (7.67) 9.92(4.54) 11.34 24.92 <.0001 EQ (SD) 23.19 (10.75) 42.58(10.19) -4.87 24.506 <.0001  Participants’ Intellectual Quotient (IQ) was assessed using the full version of the Wechsler Adult Intelligence Scale (WAIS-IV; Wechsler, 2008). As can be seen from Table 9, autistic and neurotypical participants did not differ in Full-scale IQ (FIQ), Verbal IQ (IQ) or Perceptual IQ (PIQ).   Table 9 Means and standard deviations (in brackets) of IQ scores per diagnostic group  ASD NT t df p FIQ (SD) 110.67 (19.31) 108.17 (11.99) 0.467 28.41 0.64 VIQ (SD) 114.83 (21.11) 111.44 (12.41) 0.59 27.50 0.56 PIQ (SD) 108.72 (16.58) 103.50 (11.36) 1.10 30.08 0.28  2.2. Material  The material of this study comes from storybook narratives of the wordless picture book Tuesday (Wiesner, 1991). A full description of the material is given in Chapter 1.    2.3. Procedures   37  The narrative task was administered during the standard ADOS-2 procedure. Based on the word tier of the TextGrids, narrative coding was performed by research and student assistants, all blind to participant diagnosis. Reliability of coding was measured by double-coding 10% of the transcripts. All disagreements were discussed and resolved7. See Table 10 for agreement scores.   Table 10 Percentage inter-coder agreement per coding category Category Agreement (%) Disagreement (%) Story Grammar 72 28 Reference 71 29 Discourse-structuring devices 98 2 Internal State Language 82 18  2.4. Linguistic measures  2.4.1. Syntactic coding Participants’ narratives were segmented into syntactic units following the coding protocol of (Tanguy et al. (2012). The full protocol is described in Chapter 1.   2.4.2. Story structure To examine the organization of the produced narratives, our coding scheme included a measure of story grammar. The story ‘setting’ was not included in the analysis, as it was always introduced by the experimenter (according to the instructions of the ADOS-2 and the protocol of this study). According to Kauschke et al. (2016) and Rumpf et al. (2012), there are two main events in the story: a triggering event, the frogs’ sudden ability to fly, and a turning point, their loss of this ability. In between these two pivotal events, a series of main events happen. Table 11 summarizes the different story elements coded for this category.   Table 11 Main story elements in the wordless picture book Tuesday Story elements Description  7 Here is an example of how disagreement was solved. There was a disagreement on the coding of the word ‘sleep’ (or any expressions referring to this state) as a term of cognition. This disagreement was resolved based on the argument that a word or expression could not be coded as internal if the protagonist was not conscient.  38  Sequence of events  1 The turtle and the fish watch the frogs fly away. 2 The frogs discover the surrounding neighborhood.  3 They fly among crows. 4 They fly by a man who is having a midnight snack.  5 The frogs pursue their adventure in a garden and fly into some drying laundry.  6 They visit the house of an old lady  Part 1: the old lady fell asleep watching television Part 2: she doesn’t notice the frogs Part 3: One frog took control over the TV remote.  7 Meeting with a dog Part 1: One of the frogs runs into a dog. Part 2: The other frogs come to provide reinforcement to the frog.  Decisive event The frogs lose their ability to fly and lose their lily pads. Conclusion They fall on the ground and find their way back to the pond Coda 1 Police, detectives and the media are in the streets and are puzzled to find all the lily pads in the streets. Coda 2 The pigs start to fly.  To capture whether and how participants added elements not pertaining to the traditional story schema, the following categories were also coded: Additional story event, Image description, Additional elements and Extraneous comments. Any event that was represented in the pictures but could not be coded as a main event was coded as Additional story event. The category Image description was applied when participants provided simple descriptions of elements present in the picture, without mentioning any actions or psychological states of the characters. Such descriptions often take the form of simple nominal expressions (e.g., a cat). Additional element was coded when participants mentioned events or elements that were not present in the picture. Extraneous comments was coded when participants produced comments that were related to the narrative task rather than the events depicted in the pictures. Additional elements and Extraneous comments were included in the coding to examine whether and how participants interrupted the structure of the narrative or the storytelling task itself. These two elements were hypothesized to hinder narrative coherence. See Table 12 for detailed examples of the four additional categories of story structure. Table 12 Additional categories of story structure Category Example  39  Additional story event Le chat regard avec beaucoup de curiosité The cat looks with a lot of curiosity Comparison participant (male, 24 years old) Image description Il y a une ferme   There is a farm Autistic participant (female, 27 years old) Additional elements Les grenouilles c’est des aliens en fait   actually the frogs are aliens  Autistic participant (male, 28 years old) Extraneous comments  Le genre de truc qui me fait chier The kind of stuff that pisses me off Autistic participant (female, 34 years old)  Coding of the story structure of entire narratives from the corpus are provided in Appendix 3.   2.4.3. Referential expressions The coding category of reference concerns references to the story’s characters. To measure how participants referred to these story entities, their syntactic form was coded. Story entities could be identified either in a definite or indefinite manner. Table 13 summarizes the coding categories of reference.     40  Table 13 Coding categories of reference Type of reference Examples Definite referential expressions Definite nominal expression les grenouilles (the frogs), le crapaud (the toad)  Pronominal expression ils, elles (they)  Adjective demonstrative cette grenouille (this frog), ces tortues (these turtles)  Demonstrative pronoun celle-là (that one)  Indefinite referential expressions Indefinite nominal expression une grenouille (a frog)  Indefinite adjectival expression certains (some)  Indefinite pronominal expression quelqu’un (somebody)  2.4.4. Discourse-structuring devices Two specific types of discourse-structuring devices were coded: connectives and discourse markers. Connectives were coded into the following subcategories: additive, temporal, causal and contrastive connectives. The coding of discourse markers included lexemes serving a structuring or meta-discursive function (e.g., ‘bah’(well), Tanguy et al., 2012). One important characteristics of discourse-structuring devices is that they are multifunctional (Crible, 2017). In the present corpus, connectives such as ‘donc’ (so) or ‘alors’ (then) were sometimes used as a discourse marker, viz. served a meta-discursive function, rather than relating two events together with the original meaning of the connective. In these cases, they were not coded as a connective but as a discourse marker. Table 14 summarizes the different types discourse-structuring devices.  Table 14 Coding categories of connectives and discourse markers Category Example Temporal après (after), ensuite (then), puis (then), avant (before) Additive et (and), de plus (moreover) Causal parce que (because), car (because), puisque (because), donc (so) Contrastive mais (but), cependant (however), néanmoins (nevertheless), bien que (although) Discourse marker bah (well), tu vois (you see)   41  2.4.5. Internal State Language The coding category Internal State Language (ISL) consists of references made to the internal states of the story’s characters. The following table covers the ISL categories (adopted from Kauschke & Klann-Delius, 1997, Kauschke, van der Beek, & Kamp-Becker, 2016) coded in the storybook narratives. Table 15 summarizes the different ISL categories with examples.  Table 15 Coding categories of internal state language Category Definition Example Emotion Terms referring to distinct emotions or expressive behaviors of emotions Avoir peur (to be afraid), être content (to be happy) Cognition Terms referring to mental/cognitive states and expressions of knowledge/beliefs/memories Croire (believe), dire (say), s’interroger (to question oneself) Physiology Terms referring to subjective biological and physical perception and sensations Voir (see), entendre (hear) Modality Terms of volition, obligations and intentions Vouloir (want), desirer (desire) Evaluation Terms expressing a moral judgment or an evaluation of people/events Apparemment (apparently), évidemment (obviously)  Both individual words and phrases could be coded as internal state language. For a phrase to be coded as such, it should be clearly paraphrased into a mental state term. Words or phrases were not coded as internal state language if they involved personality traits (e.g., mischievous; see Bang, Burns, & Nadig, 2013). The annotation guidelines for the linguistic measures in participants’ narrative productions are reproduced in Appendix 2.   2.5. Data preparation  All coding was performed in Praat (Boersma & Weenink, 2017). For each participant, a TextGrid was created containing all coding categories. Using the script ‘tierextraction.praat’ created by Oliver Ehmer (http://www.oliverehmer.de/transformer/), each tier was extracted into a Comma Separated Value (csv) file. The speech of the experimenter was manually removed to keep only the speech belonging to participants’ narrative production.   42  2.6. Analysis  All statistical analyses were conducted in R (R Core Team, 2016). General logistic models (Poisson family) were performed on all coding categories using the glm function. For each category, viz. syntactic coding, story structure, reference, connective and internal state language, a model was created with feature type as dependent variable and group diagnostic entered as fixed effect. When necessary, the variability in the length of narratives — as measured by number of syntactic units — was controlled for by including the total number of syntactic units per narrative as a fixed effect in the models. The significance of the model was determined by comparing it to a model without the fixed effect of diagnostic group using the anova function from the ‘stats’ package. To examine which specific variables differed per group, Tukey post-hoc analyses were conducted using the emmeans function from the ‘emmeans’ package. All significant effects (p ≥ 0.05) reported in this chapter remained so, when controlling for total number of syntactic units.  Considering the variability in measurements (count data), percentage scores were also calculated for a more homogeneous representation of the data. To visualize the proportion of a given feature within the entire narrative, percentage scores were calculated as the total count of a given feature divided by the total number of syntactic units. These percentage scores are presented in the summary tables alongside the raw counts. Models and plots were created using the raw scores of the dependent variables. In this chapter and the following, violin plots are used to illustrate the results. Violin plots represent the distribution and probability density of the data. A kernel density estimation is used to show the distribution shape of the data. The wider parts the shape indicate a higher probability that the data of participants will take on a given value while the narrower parts of the shape represent a lower probability that the data of participants will take on a given value. In this respect, compared to simple boxplots which only display summary statistics such as mean, median and interquartile ranges, violin plots are more informative as they show the full distribution of the data. For this reason, violin plots are useful when the data distribution is multimodal. This latter advantage makes them particularly suitable to describe heterogeneous data — which is often the case in autism studies.    43  III. Results  3.1. Narrative microstructure  3.1.1. Narrative productivity: total counts of words, syntactic sequences and syntactic units  Narrative productivity was measured by examining total counts of words8, syntactic sequences9 and syntactic units10. Neurotypical participants produced more words, syntactic sequences and syntactic units than the autistic participants, χ2(1) = 124.42, p < .0001; χ2(1) = 6.494; p = 0.01 and χ2(1) = 6.9519, p = 0.008. See Table 16 and 17.  Table 16 Regression coefficients of the generalized link model with the additive effect of diagnostic group (ASD diagnosis is the reference level, standard errors is in brackets)  Total words Total syntactic sequences Total syntactic units NT 0.165 (0.01)*** 0.054 (0.02)* 0.085 (0.03)** Signif. codes :  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1  Table 17 Means and standard deviations (in brackets) of total words, total syntactic sequences and total syntactic units  ASD NT Total words 467.67 (276.66) 551.56 (205.35) Total syntactic sequences 239.72 (146.78) 253.06 (79.92) Total syntactic units 101.67 (61.52) 110.72 (36.36)  In the following figures, the median is indicated by the red dot and the mean is indicated by the diamond shape. Figure 6 contains violin plots depicting the data distribution of the variables in the coding category ‘narrative productivity’ per diagnostic group, viz. total counts of words, syntactic sequences and syntactic units.     8 The count of total words is the total of the words produced during the narrative, except the repletion of the same type in a row (e.g., ‘he he he called me’), interrupted words (e.g., ‘I play rug- I train or play rugby’), hesitation markers(e.g., ‘uh’, ‘uhm’), paraverbal elements (e.g., ‘pff’, ‘mhm’) and non-words (e.g., ‘ech’).  9 The total count of syntactic sequences includes the different subtypes of functional and categorical sequences, as proposed by Tanguy et al. (2012). 10 The total count syntactic units includes the following coding categories of Tanguy et al.’s (2012) syntactic coding protocol: dependency clauses, adjuncts and discourse markers.  44  Figure 6 Violin plots of total words (plot A), total syntactic sequences (plot B) and total syntactic units (plot C) per diagnostic group   3.1.2. Types of syntactic units 3.1.2.1. Dependency clauses When looking at the syntactic coding of dependency clauses, post-hoc analyses revealed that there were no significant group differences in total number of dependency clauses, z = 1.224, p = 0.22. However, when dependency clauses were divided into complete and incomplete ones, autistic participants produced less complete dependency clauses, z = -1.965, p = 0.05 but more incomplete dependency clauses, z = 3.125, p = 0.002. Table 18 contains summary statistics of total dependency clauses, complete dependency clauses and incomplete dependency clauses. Figure 7 contains violin plots depicting the data distribution of the  45  variables in the coding category ‘dependency clauses’ per diagnostic group, viz. total dependency clauses, complete dependency clauses and incomplete dependency clauses.  Table 18 Means and standard deviations (in brackets) of counts and percentage of total dependency clauses, complete dependency clauses and incomplete dependency clauses per diagnostic group  ASD NT  Counts Percentage Counts Percentage Total dependency clauses 58.22 (34.66) 58.74 (9.07) 60.11 (20.67) 54.64 (6.67) Complete dependency clauses 53.39 (30.18) 93.14 (5.08) 57.39 (20.25) 95.31 (3.72) Incomplete dependency clauses 4.83 (5.22) 6.86 (5.08) 2.72 (1.96) 4.69 (3.72)  Figure 7 Violin plots of dependency clauses (plot A), complete (plot B) and incomplete dependency clauses (plot C) per diagnostic group    46  3.1.2.2. Dependency clause subtypes When looking at subtypes of complete dependency clauses, autistic participants produced less  complete verbal dependency clauses, z = -3.632, p = 0.0003 and more complete averbal dependency clauses, z = 3.279, p = 0.001. There was no group difference in the number of complete elliptic dependency clauses, z = 1.224, p = 0.22. Table 19 displays summary statistics of dependency clauses subtypes.   Table 19 Means and standard deviations (in brackets) of counts and percentage of dependency clause subtypes per diagnostic group  ASD NT  Counts Percentage Counts Percentage Verbal dependency clauses 43.06 (20.97) 77.97 (11.36) 50.61 (16.94) 85.12 (9.74) Averbal dependency clauses 8.39 (9.98) 12.24 (6.69) 5.39 (6.85) 8.08 (8.42) Elliptic dependency clauses 1.94 (1.73) 2.93 (2.31) 1.39 (1.24) 2.11 (1.51)  Figure 8 contains violin plots depicting the data distribution of the variables in the coding categories ‘dependency clause subtypes’ per diagnostic group, viz. verbal, averbal and elliptic dependency clauses.  47  Figure 8 Violin plots of complete verbal (plot A), averbal (plot B) and elliptic dependency clauses (plot C) per diagnostic group    3.1.2.3. Additional syntactic units : adjuncts, discourse-structuring devices and hesitation markers When examining the production of additional syntactic units, autistic participants produced less discourse-structuring devices, z = -3.017, p = 0.003 but more hesitation markers, z =  2.565, p =  0.01. than neurotypical participants. Group differences in the production of adjuncts failed to reach significance,  z = -1.820, p = 0.07. Table 20 contains summary statistics of the different types of additional syntactic units.    48  Table 20 Means and standard deviations (in brackets) of counts and percentage of additional syntactic units per diagnostic group  ASD NT  Counts Percentage Counts Percentage Discourse-structuring devices 38.67 (26.49) 36.54 (9.70) 44.50 (16.64) 39.96 (8.19) Adjuncts 4.78 (3.96) 4.72 (3.35) 6.11 (4.07) 5.41 (3.45) Hesitation markers 11.17 (9.86) 11.41 (7.17) 8.33 (4.75) 8.43 (6.49)  Figure 9 contains violin plots depicting the data distribution of the variables in the coding category ‘additional syntactic units’ per diagnostic group, viz. discourse-structuring devices, adjuncts and hesitation markers    49  Figure 9 Violin plots of discourse-structuring devices (plot A), adjuncts (plot B) and hesitation markers (plot C) per diagnostic group   3.2. Narrative macrostructure  3.2.1. Story structure There were no group differences in the total number of main story elements, z = -0.601, p = 0.55. Autistic and neurotypical participants produced comparable number of main events, conclusion and codas. There were also no group differences in the production of additional story events, z = 1.290, p = 0.2. However, there was a significant group difference in the production of extraneous comments related to the task and/or the story itself. Autistic participants produced significantly more comments than neurotypical participants, z = 2.612,  50  p = 0.009. The additional story categories Image description and Additional element were not analyzed, as there were not enough instances produced. See Table 21 for summary statistics of the elements of the story structure.  Table 21 Means and standard deviations (in brackets) of total counts of main story elements, additional story events and extraneous comments per diagnostic group  Main story elements Additional story events Extraneous comments ASD 12.61 (1.61) 3.17 (3.29) 2.72 (2.80) NT 13.33 (0.77) 2.44 (1.65) 1.44 (1.50)  Figure 10 contains violin plots depicting the distribution of the data from the variables in the coding category ‘story structure’ per diagnostic group, viz. main story elements, additional story events and extraneous comments.    51  Figure 10 Violin plot for main story elements (plot C), additional story events (plot B) and extraneous comments (plot C) per diagnostic group   3.2.2. Discourse-structuring devices When looking at both types of discourse-structuring devices combined (connectives and discourse markers), autistic participants produced less discourse-structuring devices than neurotypical participants, z = -3.443 , p = 0.0006. When breaking down by connective subtypes, autistic participants produced less additive, z = -3.517, p = 0.0004, causal, z = -5.774, p < .0001 and contrastive connectives, z = -3.264, p = 0.001 than neurotypical participants. There were no significant group differences in the production of temporal connectives, z = 1.548, p = 0.1217. Autistic participants produced more discourse markers than neurotypical  52  participants, z = 2.349, p = 0.02. Table 22 provides summary statistics of discourse structuring devices per group.  Table 22 Means and standard deviations (in brackets) of counts and percentage of discourse-structuring devices per diagnostic group   ASD NT  Counts Percentage Counts Percentage Additive connective 12.50 (10.08) 12.67 (7.53) 16.72 (7.06) 15.35 (5.17) Temporal connective 3.78 (3.64) 3.98 (3.59) 2.78 (1.52) 2.90 (2.11) Causal connective 2.39 (2.28) 2.49 (2.01) 6.56 (5.29) 5.55 (3.63) Contrastive connective 2.50 (1.69) 2.53 (1.13) 4.50 (3.65) 3.69 (2.32) Discourse maker 13.89 (13.25) 12.03 (7.53) 10.89 (5.11) 9.72 (3.61) Total 35.06 (22.96) 33.71 (8.67) 41.44 (17.48) 37.21 (8.88)  In addition to comparisons across groups, frequency statistics were also calculated within group to give a better representation of how different subcategories of linguistic measures were distributed within each group. Table 23 provides frequency statistics of the distribution of each connective subtype and discourse markers within diagnostic group.   Table 23 Distribution of the different discourse-structuring devices within diagnostic group  ASD NT  Counts Percentage Counts Percentage Additive connective 225 36 % 301 40 % Temporal connective 68 11 % 50 7 % Causal connective 43 7 % 118 16 % Contrastive connective 45 7 % 81 11 % Discourse markers 250 40 % 196 26 % Total 631  746   Figure 11 contains violin plots depicting the data distribution of the variables in the coding category ‘discourse-structuring devices’ per diagnostic group, viz. total count, connective subtypes (additive, causal, contrastive, temporal) and discourse markers.     53  Figure 11 Violin plots of total discourse-structuring devices (plot A), additive connectives (plot B), temporal connectives (plot C), causal connectives (plot D), contrastive connectives (plot E) and discourse markers (plot F) per diagnostic group    54     3.2.3. Referential expressions Autistic participants produced significantly less referential expressions overall than neurotypical participants, z = -4.037, p < .0001. Breaking down by type of referential expressions, autistic participants produced less definite and pronominal expressions than neurotypical participants, z = -2.765, p = 0.006 and z = -3.719, p = 0.0002. There were no group differences in the production of indefinite nominal expressions, z = 0.455, p = 0.65. See Table 24 for summary statistics of the different types of referential expressions.     55  Table 24 Means and standard deviations (in brackets) of counts and percent of referential expressions per diagnostic group  ASD  NT   Counts Percentage Counts Percentage Definite 12.28 (10.59) 13.23 (7.63) 15.67 (5.72) 14.44 (4.76) Indefinite 7.50 (3.26) 8.55 (3.71) 7.06 (3.70) 7.54 (6.65) Pronominal 19.06 (11.50) 19.98 (8.87) 24.78 (13.03) 22.31 (7.86) Total reference 38.83 (22.30) 44.28 (11.16) 47.50 (15.95) 41.75 (16.03)  Table 25 provides frequency statistics of the distribution of the referential expression subtypes within diagnostic group.  Table 25 Distribution of the referential expression subtypes within diagnostic group  ASD NT  Counts Percentage Counts Percentage Definite nominal expressions 221 31,62 % 282 32,98 % Indefinite nominal expressions 135 19,31 % 127 14,85 % Pronominal expressions 343 49,07 % 446 52,16% Total 699  855    Figure 12 contains violin plots depicting the data distribution of the variables in the coding category ‘referential expressions’ per diagnostic group, viz. total referential expressions, definite nominal expressions, indefinite nominal expression and pronominal expressions.     56  Figure 12 Violin plots for total referential expressions (plot A), definite nominal expressions (plot B), indefinite nominal expressions (plot C) and pronominal expressions (plot D) per diagnostic group    When looking at the story characters, autistic participants referred less often to the story’s main protagonists, viz. frogs and toads (z = -2.536, p = 0.0112), the man (z = -2.311, p = 0.02), the old lady (z =-2.193  , p = 0.03) and the frog being chased by the dog (z =-3.412, p =0.0006) than neurotypical participants. There were no significant differences for the remaining characters, namely the dog (z =-1.293, p = 0.19), turtle (z =-0.357, p = 0.72), fish (z =-1.088, p = 0.28), birds (z =-1.342, p = 0.18), cat (z = 0.928, p = 0.35), media (z =-0.152, p = 0.89)  and pigs (z =-0.013, p = 0.99). Figure 13 contains violin plots depicting the data distribution of the different story protagonists per diagnostic group.    57  Figure 13 Violin plots for total counts of references to toads and frogs (plot A), man eating a midnight snack (plot B), lady asleep in front of her television (plot C), frog chased by the dog (plot D), dog (plot E), turtle (plot F), fish (plot G), birds (plot H), cat (plot I), media (plot J) and pigs (plot K) per diagnostic group    58    59    3.3. Internal State Language  Autistic participants produced less internal state language overall than neurotypical participants, z = -5.182, p  <.0001.  When looking at specific subtypes of internal state language, autistic participants produced less cognitive mental state terms, emotional mental state and physiological terms than neurotypical participants, z = -3.934, p = 0.0001; z = -2.906 p = 0.0037 and z = -3.495, p = 0.0005, respectively. There were no group differences in the production of modal terms and evaluative markers, z = -1.471, p = 0.1412 and z = -0.158, p = 0.8748, respectively. See Tables 26 and 27 for summary statistics of the different types of internal state terms.   60  Table 26 Means and standard deviations (in brackets) of counts and percentage of internal state language per diagnostic group  ASD  NT   Counts Percentage Counts Percentage Total 16.56 (12.39) 16.59 (7.84) 23.72 (12.04) 20.87 (6.29) Cognition 2.94 (3.39) 2.67 (1.54) 5.56 (5.17) 4.66 (3.07) Emotional 3.39 (4.09) 3.08 (2.92) 5.28 (2.93) 4.99 (2.65) Physiological 2.67 (2.11) 3.38 (2.85) 4.83 (2.48) 4.37 (1.78) Modal 2.39 (2.52) 2.50 (2.57) 3.11 (3.22) 2.63 (2.37) Evaluative 5.06 (4.68) 4.78 (2.99) 5.00 (3.27) 4.25 (2.47) Total 16.56 (12.39) 16.59 (7.84) 23.72 (12.04) 20.87 (6.29)  Table 27 Distribution of the different internal state language subtypes within diagnostic group  ASD NT  Counts Percentage Counts Percentage Emotion 61 20 % 95 22 % Cognition 53 18 % 100 23 % Physiological 48 16 % 87 20 % Modal 43 14 % 56 13 % Evaluative 91 31 % 90 21 % Total 298  427   Figure 14 contains violin plots displaying the data distribution of the variables in the coding category ‘internal state language’ per diagnostic group, viz. total internal state language, cognition terms, emotion terms, physiology terms, modal terms and evaluative terms.     61  Figure 14 Violin plots of  total internal state language (plot A), cognition terms (plot B), emotions terms (plot C), physiology terms (plot D), modal terms (plot E) and evaluative terms (plot F) per diagnostic group   62      63  3.4. Summary of coding results  A summary of all coding categories and their associated group effect are summarized in Table 28. Table 28 Summary of all coding categories and their associated group effect Narrative dimension Feature Group difference Microstructure Syntactic units   Total words NT > ASD  Total syntactic sequences NT > ASD  Total syntactic units NT > ASD  Total dependency clauses NT = ASD  Complete dependency clauses NT > ASD  Incomplete dependency clauses ASD > NT  Verbal dependency clauses NT > ASD  Averbal dependency clauses ASD > NT  Elliptic dependency clauses NT = ASD  Discourse-structuring devices NT > ASD  Adjunct  ASD = NT  Hesitation markers ASD > NT Macrostructure Story Structure   Main story events NT = ASD  Additional story events NT = ASD  Extraneous comments ASD > NT  Discourse-structuring devices   Total connectives NT > ASD  Additive NT > ASD  Causal NT > ASD  Contrastive NT > ASD  Temporal NT = ASD  Discourse markers ASD > NT  Reference   Total reference NT > ASD  Indefinite reference NT = ASD  Definite reference NT > ASD  Pronominal reference NT > ASD  Reference to main characters NT > ASD Internal state language Total internal state language NT > ASD  Cognitive NT > ASD  Emotional NT > ASD  Physiological NT > ASD  Modal NT = ASD  Evaluative NT = ASD  64  IV. Discussion  The current study systematically analyzed the features associated with three central dimensions of narrative production (microstructure, macrostructure and internal state language). The results show that the narratives of autistic adults differed from neurotypical participants on all three dimensions of narrative production, lining up with the conclusions of Baixauli et al. (2016). The presence of group differences in the narrative productions of our participants validates this new corpus as a good source of data to examine spoken discourse in autism. This study also adds to the existing literature on narrative production by yielding new findings, presented in the following paragraphs.  At the level of the microstructure, in addition to being less productive, overall, than their neurotypical peers (fewer words, syntactic sequences and syntactic units), autistic participants also produced narratives composed of more incomplete dependency clauses, viz. more dependency clauses where an obligatory element required by verbs was missing. Example (1) and (2) illustrate the coding of incomplete utterances. In all examples below, dependency clauses are indicated by square brackets, sequences by brackets, non-governed elements by angle brackets and length of silent pauses are indicated in brackets (seconds). Incomplete utterances are highlighted in bold.  In examples (1) and (2), the participants start a clause about the main protagonist (frogs), ‘one of them to-’ and ‘they do not seem to be more’ without finishing it. Instead of completing the clause or restarting it, the participants directly move on to a new clause. While incomplete utterances are a typical phenomenon occurring in spoken speech, the fact that autistic participants produced significantly more of them than their neurotypical peers, and hence more than what would typically be expected, suggests that this difference is likely to be perceived by the listener.  (1) [(une d’elles pr-)] (0.47) <mais> <heureusement> [(ce petit problème) (fut vite résolu)] [(one of them to-)] (0.47) <but> <luckily>[(this small problem) (was rapidly solved)]  Autistic participant (male, 19 years old)    65  (2) <mais> [(elles ont pas l’air d’être plus)] <euh> (6.8) [(c’est un peu) (comme si elles étaient posées) <en fait> (sur leur sur leur feuilles)] <but> [(they do not seem to be more)] <uh> (6.8) [(it’s a bit) (like if they were placed) <actually> (on their on their leaves)] Autistic participant (female, 43 years old)  Furthermore, in comparison to neurotypical participants, autistic participants are less likely to produce complete verbal dependency clauses, i.e., dependency clauses with a verbal head and more likely to produce complete averbal dependency clauses, i.e., dependency clauses without a verbal head. Considering that verbs form the main predicate of sentences and, as such, describe important information such as actions or states, a higher proportion of complete verbal dependency clauses will result in narratives with a strong focus on the verbalization of the characters’ goals and actions (example 3). The verbal head ‘seize the opportunity’ allows the participant to provide a detailed description of the event: that they change the channels and that they watch their favorite program. In contrast, by including more complete averbal dependency clauses, autistic participants were more likely than neurotypical participants to direct the focus on the story’s temporal frame, ‘next day/next Tuesday’ (example 4) or atmosphere, ‘a return to normality’ (example 5). The combination of producing less complete and more incomplete dependency clauses with less verbal and more averbal dependency clauses suggests that already at the microstructure of discourse, autistic participants are communicating less information than their neurotypical peers.  (3) <et> [(elles en profitent) (pour pouvoir changer les chaînes et regarder leur programme favori)] <and> [(they seize the opportunity) (to change the channels and watch their favorite program)] Comparison participant (male, 19 years old)  (4) <et> <euh> [(prochaine journée]] <enfin> [(prochain mardi)] <and> <uh> [(next day]] <well> [(next tuesday)] Autistic participant (male, 32 years old)  66   (5) <après> <voilà> [(une un retour à la normale)] (0.52) <voilà> <then> <there> [(a return to normality)] (0.52) <voilà> Autistic participant (female, 32 years old) Turning to the macrostructure, participants in both groups structured their stories similarly as there were no differences on any of the main story elements. Autistic adults consistently organize their story according to the expected format of narrative discourse: they were as likely than their typical peers to include main story events, to provide a conclusion to their story and to mention pivotal story events. Moreover, the groups did not differ in the production of additional story events, viz. secondary story events. These findings suggest that similar to neurotypicals, autistic adults can systematically organize their discourse according to an implicit structure (Whitworth, Claessen, Leitão, & Webster, 2015). Furthermore, mirroring results found by Diehl et al. (2006) on young autistic children, the present findings suggest that autistic adults were sensitive to the importance of gist story events. However, they were more likely to produce extraneous comments about the story events (e.g., I don’t know what’s happening here) and the narrative task itself (e.g., the sort of tasks that pisses me off). In other words, while autistic participants did not have difficulties in detecting the main events necessary for story comprehension, they did seem to experience difficulties in using these main events as a common thread, resulting in disrupted narratives which were not focused around the gist events.  Furthermore, although the autistic adults in this study have acquired typical knowledge of story organization, they differed from their neurotypical peers in their use of linguistic devices to maintain internal cohesion and coherence, viz. discourse-structuring devices and referential expressions. Recall that in our coding of discourse-structuring devices, connectives were explicit markers of logical relations (causal, additive, contrastive or temporal) and discourse markers were explicit markers with a meta-discursive function such as ‘bah’(well), ‘voilà’ (there you go), ‘enfin’(well). Autistic participants produced less additive, causal and contrastive connectives, but more discourse markers than neurotypical participants. While additive, causal and contrastive connectives provide specific instructions to the listener on how to connect events or ideas together (example 6), discourse markers provide information on the speaker’s attitude towards the production of the story and/or the narration task itself  67  (example 7). Furthermore, while additive, causal or contrastive connectives will introduce the development of the preceding state of affair or statement, discourse markers are more likely to introduce a comment or a revision of what has just been said. In example (6), the causal connective ‘parce que’ (because) instructs the listener to establish a causal relation between 1) the fact that the man is seeing weird things fly near his window and 2) his mental state of considering going to bed. By contrast, in example (6), the discourse marker ‘enfin’ (well) introduces a revision of the first statement, viz. the participant starts saying the frog invasion is arriving in a city and then corrects himself by saying that it is not really a city. There is no logical relation to be established between the two parts of the utterance.   (6) il se demande s’il ne devrait pas aller se coucher parce qu’il commence à voir des choses bizarres voler par le (0.53) par la fenêtre  and he wonders whether he should not go to sleep because he starts to see weird things flying through the (0.53) through the window Comparison participant (male, 19 years old)  (7) l’invasion arrive euh dans euh les coins de la ville (0.73) enfin c’est pas vraiment une ville and the invasion uh arrives uh in parts of the city (0.73) well it’s not really a city Autistic participant (male, 19 years old)  In other words, autistic adults can and do use discourse-structuring devices to piece together their narratives, albeit the relations they establish are less helpful to the listener as they are less informative and more subjective than those established by typical adults. It should be noted that only explicit coherence relations were coded, viz. relations made linguistically explicit. However, relations can also remain implicit and conveyed through the juxtaposition of two sentences or utterances, leaving it up to reader/listener to infer the relation (e.g., Sanders & Noordman, 2000; Taboada, 2014). Leaving a discourse relation implicit will put a higher burden on the listeners who will have to infer the relation themselves (e.g., Zwaan & Radvansky, 1998). For example, it could be the case that autistic participants produce less explicit discourse-structuring devices, but more implicit relations than neurotypical participants, and vice-versa. To gain insight into this hypothesis, future studies should code both for explicit and implicit relations in storybook narratives, in order to explore whether autistic and neurotypical participants differ in their preferences to mark relations explicitly or implicitly.  68  There were also differences in the frequencies of referential expressions used to mention story characters. Autistic participants used less definite nominal and pronominal expressions than neurotypical participants. There were no group differences in the use of indefinite pronominal expressions. When looking at within-group frequencies of referential expressions, the distribution pattern was almost identical for the two groups. This suggests that autistic participants know how to use different types of linguistic expressions to refer to story characters, however they have more difficulties determining in how often they should refer to the different protagonists of the story. In addition to analyzing how story characters were mentioned, which characters were being referred to was also examined. The results show that autistic participants referred less often to the main protagonists of the entire story, viz. the frogs/toads as well as important secondary characters of individual main events, namely the man having a late-night snack, the old lady sleeping in front of her television, the frog manipulating the remote control of the television and the solitary frog being chased by the dog. These latter findings suggest that the two groups differed in the type of information they included in these events. All these characters play an important role in understanding the key events of the story. In other words, although autistic participants mentioned the main events of the story, there were some subtle group differences regarding the actors associated with these events. Simply mentioning the event, but not the protagonists involved in it, is not enough to convey a coherent story. Finally, regarding the dimension of internal state language, autistic participants were less likely to include mental state terms in general than neurotypical participants. When looking at subtypes of internal state language, there were significant group differences in cognitive, emotional and physiological terms, but no group differences in modal and evaluative terms. In other words, the narratives of autistic participant provided less information about the mental reasoning as well as emotional and sensorial experiences of the story characters.   V. Conclusion  Taken together, the comprehensive assessment of narrative abilities in autistic adults and their neurotypical peers suggests that autistic individuals do not lack the ability to create a story, however they are less able than neurotypicals to determine accurately how often they should incorporate the different features in their narrative and how to exploit them together. Difficulties to gauge the accurate ‘dosage’ of the different linguistic features led to a higher frequency of features foregrounding the development of the participants’ own evaluation of  69  the story (e.g., discourse markers) or of the narrative task (extraneous comments), as well as to a lower frequency of linguistic features used to develop the story characters (e.g., definite nominal expressions and pronominal expressions, internal state terms) and to establish a relationship between the story events (e.g., connectives). This unequal dosage of linguistic features is likely to make it harder for the listeners to piece together an enlightening and coherent narrative. Let’s take as an example one of the story’s main event, namely the chase in the garden. This event unfolds in two parts, displayed over two story boards. First, there is a frog (detached from the rest of the group) that is wandering in a garden and makes a surprise encounter with a dog. The dog starts chasing the frog who is frightened. Second, this frog is helped by its friends who in turn chase the dog (that is now also frightened). Image 1 illustrates the event.  Image 1 Story boards from the wordless picture book Tuesday (Wiesner, 1991) depicting the chase in the garden.     Example (8) and (9) illustrate how, in spite of creating the structure for the story’s event, a failure to use certain linguistic forms when describing this event — a referential expression to single out the frog, terms to refer to the internal states of the protagonists and connectives to indicate the unexpected reversal of situation — will impact the ‘goodness’ of the story. In example (8), the frog wandering alone in the garden is not singled out, by any linguistic form, from the mass of frogs, described with a plural definite nominal expression (the frogs). Lack of reference to this secondary, important protagonist will limit the use of other linguistic forms, such as connectives and internal state language. To see this, compare with the example (9), whereby referring to the single frog (a frog, it), the participant now has the possibility to  70  mention the internal state of the frog (she starts to be scared). It also creates the possibility to use a contrastive connective (but) and an evaluative term (luckily) to highlight the fortuitous unexpected turn of the initial situation: all the other frogs come to rescue the single frog and in turn scare the dog away. The description of the event in example (8) thus, stands in stark contrast with example (9).   (8) Part1 : puis y’a un chien qui coure après les grenouilles then there’s a dog running after the frogs Part 2 : puis bah y’a les grenouilles qui courent après le chien then well there are frogs running after the dog Autistic participant (female, 33 years old) (9) Part1 : et puis bon c’est quand même un peu un peu l’heure de rentrer donc là elles commencent à sortir mais là y a le chien et le chien du coup il a envie de savoir un peu ce que c’est une grenouille comme ça donc il commence à l'attaquer donc là elle commence à avoir peur parce que elle se sent qu’elle va se faire manger and then well it really is time to go back home so there they start going out by there there’s a dog and so the dog he wants to know what is a frog like that so he starts attacking so it starts to be scared because it feels it will get eaten Part 2 : mais heureusement toute la colonie des petits amis viennent la à la rescousse de la grenouille qui se faisait attaquer et là elles se mettent à la poursuite du chien le chien effrayé commence à à s enfuir but luckily the whole army of friends come to rescue the frog that was being attacked and they start chasing the dog the dog is scared and starts running away  Comparison participant (female, 26 years old)  In other words, it is not sufficient to include individual linguistic features of cohesion, it is also necessary to combine them in such way that they enhance each other, resulting in a strong, connected story. This is the case in example (9). As still another point of comparison, consider example (10). Here, the participant uses a pronominal expression to single out the frog (she), as well as a contrastive marker (but) but does not exploit these forms to further develop the  71  internal states of the frog and dog or convey a convincing effect of unexpectedness. On the basis of this rather weak description of the event, the listener will not be able to build a compelling mental representation of the event and important pieces of information are left unspoken (viz. information about the emotions of the protagonists, the rescue mission of the frogs and the escape of the frightened dog).   (10) Part 1 : bah elle allait atterrir mais y a un chien well she was going to land but there’s a dog Part 2 : toutes les autres grenouilles font partir le chien all of the other frogs make the dog go away  Autistic participant (female, 40 years old)  While it can easily be derived from the present analysis that autistic participants took into account the listener’s needs to a lesser extent than did the neurotypical participants, providing adequate cognitive explanations for this outcome is less straightforward. On the one hand, being less sensitive to the listener’s need and/or knowledge could be related to Theory of Mind (ToM) abilities. Autistic participants could have more problems estimating how much information to communicate to the listener due to difficulties with ToM. However, it should be noted that the storytelling did not require participants to ‘read’ the mind of the listener, viz. picking up on cues in the listener’s face or voice, to create the story. Specifically, the ADOS guidelines instructs the administrator to start narrating the story before handing it over to the participant, who has to take on the storytelling task without any support or help from the administrator. Thus, there is little interaction between the participant and the administrator. The task results in a monologic narrative, where the participant has to attend to and interpret different cues in the pictures. Therefore, lower frequencies of certain features could also reflect reduced attention to these cues.  Finally, it also remains possible that autistic participants could have been simply less motivated to create a good story. The finding that autistic participants produced more extraneous comments provides some evidence in favor of this assumption. Extraneous comments created distance with the story. Autistic participants did not fully immerse themselves in the processing of storytelling, keeping on their role of participants rather than  72  fully endorsing the role of a narrator. Informal evidence gathered during the narrative task provides further evidence for this assumption. Once the narrative task was over, the ADOS administrator explicitly asked all participants whether they enjoyed the task or not.  67% of autistic participants responded either that they didn’t like the story or that they found it very weird. In contrast, 61% of the comparison participants responded that they liked the story or that it was funny. In other words, it could be the case that autistic participants were less motivated to engage in a narrative performance, resulting in narratives that were more difficult to follow by the listener. Another implication of the present analysis is that neurotypical participants exploited the different features more efficiently than autistic participants to create a coherent narrative11. As discussed in the previous discussion, the emergence of coherence does not result from one specific linguistic feature or from the mere presence of these features; rather coherence emerges from the constellation of these features. Discourse analysis of the narratives in this study relied on the underlying assumption that differences in frequency of a certain linguistic form, e.g., a referential expression or discourse-structuring devices, would entail that a certain function, e.g., creating coherence, will not be fulfilled. An outstanding issue is then whether other linguistic forms are used in compensation to fulfill this function. For example, it could be the case that autistic participants compensate for less additive, causal and contrastive connectives by using more discourse markers. Alternatively, they could be compensating for the function of underused connectives with other forms not coded for in this study, for example, with linguistic items of lexical cohesion (e.g. repetitions, synonymy, collocation).  Conversely, in cases where there is no group difference in the frequency of a given linguistic form, it still remains possible that this form is used with different functions in the two groups. For example, in a case study de Villiers (2011) examined the narrative of an autistic boy. She used discourse analysis techniques such as phasal analysis (e.g., Gregory, 2002) to reveal unexpected coherence in a discourse that would be characterized by naïve listeners as incomprehensible. Her detailed analysis highlights that the autistic boy used meaningful, albeit unexpected and unconventional, linguistic patterning (e.g., modality, rhetorical questioning) to create coherence and participate in the on-going interaction. Future studies should complement quantitative analyses with functional analyses to gain a better understanding of the type of features leading to group differences.   11 I would like to thank Prof. Jürgen Jaspers in particular for making me think about the meaning of narrative coherence and its possible implications. His comments have really helped me enrich the present discussion.   73  Chapter 3: How do autistic adults use syntactic and prosodic cues to manage spoken discourse?  Exploring discourse strategies with Basic Discourse Units  I. Introduction  Chapter 2 focused on identifying discourse features related to (in)coherence in narrative productions of autistic and neurotypical adults. My detailed transcript analyses of narrative discourse corroborated previous results on narrative production in autism, as the narratives of autistic adults differed from those by neurotypical adults on the main narrative dimensions (microstructure, macrostructure and internal state language). In this first step, the focus was on gathering information about the different pieces of the discourse of autistic and neurotypical adults, with the aim of singling out specific features that could be associated with a diagnosis of autism. The features that stood out as specific to autistic adults were atypical use of connectives and discourse markers as well as extraneous comments. However, it is not only important to delineate the elements that characterize the spoken discourse of autistic individuals, but one also has to understand how these elements are integrated in the discourse. This will be the aim of the current chapter: to characterize how syntax and prosody come together, in the speech of autistic and neurotypical adults, to convey information. This aim is important because most coding guidelines of narrative production instruct one to code for the presence or absence of a target feature, and do not consider how it is prosodically integrated in the discourse. For instance, the coding category of narrative structure used in Chapter 2 indicated whether a story element was included or not to measure participants’ ability to include conventional story elements. However, this coding will not provide any information about how this element is delivered within the story structure. In other words, such coding will not differentiate between a participant who included the story element ‘conclusion’, but did so while pausing regularly during narration, from another participant who also included this element but at a more fluent pace. As an illustration, consider the following examples of the verbalization of the conclusion of Tuesday. Silent pauses are indicated in brackets (in seconds).   (1) euh elles continuent à voler (8.55) et elles retournent euh (0.46) dans la mare (4.28) euh (1.18)  74  uh they continue to fly (8.55) and they return uh (0.46) in their pond (4.28) uh (1.18) Autistic participant (female, 40 years old)  (2) et ils s’encourent finalement ils n’ont plus leur nénuphar volant donc ils s’encourent à toutes pattes vers l’étang le plus proche (0.59) et là ils retrouvent leur petite vie paisible  and they finally run they don’t have their flying lily pads anymore so they run really fast towards the nearest pond (0.59) and there they are back to their calm little lives Comparison participant (male, 53 years old)  According to the coding scheme of the story structure, the participants in examples (1) and (2) both included the ‘conclusion’ of the narrative as they both mentioned the concluding event of returning to the pond. Furthermore, both speakers were able to create interdependence between the different utterances to create a coherent whole of the event. In these specific examples, the speakers used additive (and) and causal connectives (so). As such, the story structure and syntactic coding will not distinguish between the two participants, but the delivery style of the two speakers does. In example (1), two long pauses of 8 and 4 seconds suggest that the speaker is experiencing difficulties with the interpretation of the story images or is reluctant to verbalize them. In example (2), the speaker verbalizes the event with more fluency, suggesting no such difficulties and/or more enthusiasm to narrate the story. Hence, a more accurate representation of the discourse competence of these two speakers as well as one that would distinguish the two speakers would require a measure integrating both syntactic and prosodic cues. This assumption has been previously argued for by Degand & Simon (2005, 2008, 2009a, 2009b) who claim that neither the morpho-syntactic structure, nor the prosodic arrangement of spoken speech, are individually sufficient as a method to efficiently segment discourse into units. To address this gap, the authors designed an innovative segmentation method whereby the boundaries of syntactic units and prosodic units are mapped onto each other to yield Basic Discourse Units (BDUs).  These two types of boundaries can overlap in different ways, resulting in different types of BDUs, corresponding to different types discourse strategies (Degand & Simon, 2008, 2009a). A one-to-one mapping between a syntactic unit and a prosodic unit will result in a congruent BDU. The authors hypothesize that with this type of BDU, the speaker communicates information active in her mind in ‘one go’, without any topicalization strategy, indicating to the addressee that she can interpret this pack of information as one idea. This  75  type of BDU is assigned an unmarked strategy, with information presented in a straightforward and rather neutral way (Degand & Simon, 2009a). In the following examples, square brackets indicate the boundaries of a syntactic units and the three slashes represent silent boundary.   (3) [c’- c’est pas ma pensée première]1 ///1 [i- it’s not my first thought]1 ///1 Autistic participant (male, 54 years old)  The congruent BDU exemplified in (3) was produced during a conversation about weddings, in which the participant was asked whether he would want to have a wedding or not. He expresses his opinion about the topic in one go, namely that his first thought is not to have a wedding.  Speakers can also communicate more than one idea in ‘one go’ by condensing several syntactic units into one prosodic unit. In this case, there is a many-to-one mapping, yielding   an intonation-bound BDU. This strategy involves information packaging, which indicates to the listener that the different syntactic units are to be understood as one macro-unit of information. Consider examples (4) and (5).  (4) [on va jamais en balade avec son chien]1 [elle elle sort pas ses chiens]2 [c'est bizarre]3 ///1 [we never go walk with her dog]1 [she never walks her dogs]2[ it’s weird]3 ///1 Autistic participant (female, 43 years old)  (5) [il peut t'écouter]1 [tu peux passer du temps normal avec lui]2 [t'es toi-même]3 <quoi> ///1  [he can listen to]1 [you he can casually spend time with him]2 [you’re yourself]3 <yeah> ///1  Comparison participant (male, 20 years old)  In example (4), the topic of the conversation is activities with friends. The participant is explaining that she and her friend never walk their dogs together (syntactic unit 1) and the reason why they don’t, is that her friend never goes out to walk her own dogs (syntactic unit 2) and finally, the speaker provides her own evaluation of this situation, namely that it’s weird  76  (syntactic unit 3). In example (5), the topic of conversation is how to recognize that someone is your friend. The participant provides three different examples (syntactic units 1, 2 and 3), which are all to be understood as supporting the same opinion about friendship. There can also be a one-to-many mapping, viz. one syntactic unit corresponds to several internal prosodic units. This type of mapping yields syntax-bound BDUs, which fulfill different strategies. First, by chunking one idea into several prosodic units, the speaker can create emphasis. In example (6) below, the topic of conversation is about friendships; the participant is asked to describe her friends. The participant isolates ‘une’ (an) and ‘comprehension’ (understanding) with silent boundaries (silent pause > 200 ms) to emphasize a certain quality of her friends, namely that her friends are particularly good at understanding social relationships.  (6) [ils ont ///1 une ///2 compréhension ///3 des liens sociaux avec les gens ///4]1 [They have ///1 an ///2 understanding ///3 of social relations with people ///4]1 Autistic participant (female, 43 years old)  Alternatively, delivering one idea in several pieces can be perceived as disfluent, suggesting processing problems. Consider example (7).  (7) PP : [je vais pas chercher à ///1 euhm oui à ///2 à ///3]1 [i’m not going to ///1 <uh> <yes> to ///1 to ///3]1 Exp: [trouver de la compagnie]1 ///1   [find company]1 ///1  PP : [trouver de la compagnie]1 ///1  [find company]1 ///1  Autistic participant (male, 54 years old)  In example (7), the speaker and experimenter are discussing the feeling of solitude. At this point of the conversation the participant is describing what he does when he feels lonely. However, the participant is struggling to terminate the syntactic clause he initiated, which is  77  visible by this set of features: a hesitation marker (‘uh’), an insert (‘yes’), a repetition (‘to to’) and three silent pauses longer than 200ms (///). His listener, the experimenter, clearly perceives this difficulty and completes his utterance herself.  Whether a syntax-bound BDU will be interpreted as marking emphasis or reflecting dysfluency will depend on the context and neighboring linguistic features. Compare example (6) above, with example (8) below. In example (6), the syntactic unit does not include any dysfluency features such as false starts, hesitation markers, repetitions or repairs. Hence, the silent boundaries segment a ‘fluent syntactic unit’, serving to mark emphasis on specific aspects of the utterance. In contrast, in example (8), the syntactic unit contains several dysfluency markers such as a repetition (c’est pas c’est pas), a false start (qu-) with a reformulation (une chose) and a hesitation marker (euh). This syntax-bound BDU will be perceived as disfluent rather than as the participant’s attempt to emphasize an element of the utterance.   (8) [c'est pas c'est pas qu- une chose à ///1 euh à laquelle j'accorde beaucoup de de valeur ///2]1 [it is not it is not som- a thing to ///1 uh which I give a lot of importance ///2]1         Autistic participant (male, 54 years old)  In regulatory BDUs, a major intonation boundary isolates a non-governed unit such as a discourse marker, connective or an adjunct. Regulatory BDUs reflect a meta-discursive strategy, with a focus on the coherence and/or information structure of the unfolding discourse. Degand & Simon (2009a) describe several ways in which regulatory BDUs contribute to the management of the son-going discourse.  One strategy is to introduce a new topic or end an on-going one. Consider example (9).     78  (9) PP : [ouais ouais]1 [voilà]2 [c'est ça]3 [il avait quatorze et demi]4 <euh> <bon> <bah> ///1 [yeah yeah]1 [exactly]2 [that’s it]3 [he was fourteen and half years old]4 <uh> <well> <you know>///1 Exp: <mhm> <mhm> PP: <voilà voilà > ///1 <so that’s it> ///1 Exp : <et> [est ce que vous avez ///1 un conjoint///2]1 <and> [do you have ///1 a partner ///2]1 PP : <bah> [j'ai un partenaire]1 [oui]2///1 <well> [I have a partner]1 [yes]2 ///1 Autistic participant (female, 43 years old)  At this point of the conversation, the topic is the participant’s dog that died recently. With the regulatory BDU ‘voilà voilà’ (so that’s it) the participant indicates to the experimenter that she agrees with what she said and that she herself has nothing to add. By ending this story line, the participant gives room to the experimenter to start a new topic of conversation. In this example, the experimenter starts a new topic about romantic relationships, viz. whether the participant has a partner or not.   Regulatory BDUs can also reflect the speaker’s evaluation of the validity of the information expressed by the interlocutor, as illustrated in example (10).   (10) <mais> <blindé>///1 [je suis tout à fait d'accord]1 ///1    <but> <really>///1 [I totally agree]1 ///1  Comparison participant (male, 20 years old)  In example (10), the participant indicates to the experimenter that he completely validates what she has said with ‘mais blindé’ (but really) and explicitly confirms with the utterance ‘je suis tout à fait d’accord’ (I totally agree). Regulatory BDUs can also reflect the speaker’s focus  79  on his or her personal opinion with respect to what is being conveyed in the interaction. For example, in (11), the participant emphasizes his point of view that he never had any problems living with another person by first introducing it with ‘bah franchement’ (well frankly).  (11) <bah> <franchement> ///1 [le mec je le connaissais pas d'avant]1 <tu vois> [c'était la première fois que je le rencontrais]2 <et> <euh>///1 <et> [y'a jamais eu aucun s- problème]1 <quoi>///1 <well> <frankly> ///1 [the guy I didn’t know him from before]1 <you know>[it was the first time I was meeting him]2 <and> <euh> ///1 <and> [ there was never any problem]1 <you know>///1 Comparison participant (male, 20 years old)  A final strategy reflected in the use of regulatory BDUs, is the indication of the speaker’s mental processes. For example, in (12), the participant is trying to explain something but has difficulty to do so. To indicate that she is struggling but still wants to maintain her turn, she first says ‘allez’ (oh), before explicitly verbalizing she has difficulties to explain.   (12) <allez>///1 [je sais pas je sais pas comment expliquer]1 ///1 <oh> ///1 [I don’t know how to explain]1 ///1  Autistic participant (female, 43 years old)  Finally, mixed BDUs are a ‘left-over’ category; in these BDUs there is no coincidence between the prosodic and syntactic boundaries, as in example (13). Degand and Simon did not assign any particular strategy to this type of BDU.  (13) [oui]1 [j'imagine]2 <fin> [j'i- j'imagine que c'est plus quelque chose ///1 qu'on fait par rapport à son entourage ou pour]3 <euh> [oui]4 ///2 [yes]1 [I imagine so]2 <well> [I i- I imagine that it’s more something ///1 that you do regarding your relatives or for]3 <euh> [yes]4 ///2 Comparison participant (female, 43 years old)   80  Furthermore, Degand and Simon (2008, 2009a) have demonstrated that the distribution of the different BDUs also varies as a function of discourse genre. Their analysis shows that intonation-bound discourse units are typical of less prepared and informal spoken discourse, such as conversations, while syntax-bound discourse units are typical of more prepared and formal spoken discourse such as radio news or interviews. Congruent BDUs are equally distributed across genres. The discourse strategies of intonation bound units and regulatory BDU are more prevalent in conversations than syntax-bound units and reflect strategies crucial to the successful management of the discourse (e.g., turn-holding, meta-discursive and interactional regulation and information packaging). The different types of BDUs and their strategies are summarized in Table 29.  Table 29 Summary of the different types of Basic Discourse Units and their corresponding strategies BDU Strategy Genre Congruent BDU  One-to-one mapping presenting information in a direct and relatively neutral manner, one conceptual idea communicated in ‘one go’ Formal and informal discourse genres Syntax-bound BDU  One-to-many mapping emphatic, didactic, or resulting from discourse planning (processing difficulties) More typical of formal genres Intonation-bound BDU Many-to-one mapping creation of a macro-unit of information (information packaging), turn-holding device More typical of informal genre Regulatory BDU  Adjunct/discourse marker in a major prosodic unit a non-governed element is autonomized in a major prosodic unit interactional or meta-discursive regulation More typical of informal genres Mixed BDU  Mismatch matching No strategy Formal and informal discourse genres  Managing conversations, as well as tailoring the speech to the discourse context are two known areas of difficulties in ASD. Furthermore, considering that atypical prosody is characteristic in ASD and the role of prosody in BDUs, segmenting autistics’ speech into BDUs seemed particularly suitable to provide novel insights into the communication difficulties experienced by autistic adults.      81  Outlook In light of the preceding discussion, the aim of this study was to explore whether the often-reported perception of autistic people’s discourse as incoherent and atypical could be modeled on the basis of these different types of BDUs and their respective strategies. Specifically, I hypothesized that autistic adults would produce more syntax-bound BDUs, possibly reflecting their pedantic style and/or processing difficulties. In contrast, I hypothesized that neurotypical adults would produce more silence-bound and regulatory BDUs, reflecting better coherence and discourse management skills than the autistic adults.  I modeled the segmentation procedure of Degand and Simon (2008, 2009a) on a speech sample of autistic and neurotypical adults obtained within the context of a semi-structured conversation on relationships. As already mentioned in Chapter 1, I did not replicate the segmentation procedure exactly. In practical terms, the prosodic units in this study are only delimited by silent pauses and not by fine-grained acoustic characteristics as is the case in the original segmentation procedure. Therefore, the BDUs in this corpus emerged from the mapping between syntactic units and silent pauses (unpreceded by a hesitation marker). To reflect more precisely the methodology adopted specifically in this dissertation, I will use the term ‘silence12-bound’ BDUs, rather than intonation-bound BDUs.     II. Methodology  This study received ethical clearance from the Ethics Committee of the Faculty of Psychology and Education at Université libre de Bruxelles and the Behavioural Research Ethics Board of the University of British Columbia. Written consent was obtained from all participants or their parents.  2.1. Participants  Considering the exploratory nature of the study on the one hand and the elaborate segmentation procedure on the other hand, a small portion of the initial participant sample, viz. twelve participants, was processed. Table 30 provides details on participants’  12 As Liesbeth Degand sensibly highlighted, the use of the term ‘silence-bound’ might be misleading because silence implies the absence of sound, whereas a silent pause implies an interruption of speech (or sound). However, it was equally misleading to use the term ‘intonation-bound’ since the BDUs were not mapped based on intonation features. Therefore, for clarity and practical reasons, I opted for the term ‘silence-bound’ to refer to BDUs in which several syntactic units are bound by one silent pause.   82  characteristics. Like for the participant sample in Chapter 2, the ADOS-2 and AQ scores of the autistic participants were significantly higher than those of the neurotypical participants. Neurotypical participants scored significantly higher on the EQ.  Table 30 Descriptive statistics of participants’ characteristics per diagnostic group (ASD is the reference level)  ASD NT t df p N (M:F) 6 (3:3) 6 (3:3)    Age (SD) Age-range 34.71(12.71) 20.00-52.09 36.56 (10.87) 20.10-52.01 -0.2709  9.764 0.7921  ADOS Total (SD) 11.67 (3.83) 1.17 (1.33) 6.3446 6.1873 0.0006 AQ (SD) 38.40 (9.29) 10.40 (4.67) 6.0219 5.8996 0.001 EQ (SD) 25.00 (13.98) 47.40 (7.50) -3.1565 6.1274 0.0191  As can be seen from Table 31, autistic and neurotypical participants did not differ in Full-scale IQ (FIQ), Verbal IQ (IQ) or Perceptual IQ (PIQ).   Table 31 Means and standard deviations (in brackets) of the IQ scores per diagnostic group (ASD is  the reference level)  ASD NT t df p FIQ 116.33 (11.81) 112.17 (8.28) 0.7076 8.9594 0.4972 VIQ 124.00 (10.53) 112.67 (10.41) 1.8756 9.9987 0.0902 PIQ 108.33 (15.31) 108.17 (8.21) 0.0235 7.656 0.982  2.2. Material  The data analyzed in this study comes from the semi-structure tasks Friends, Relationships, and Marriage and Solitude administered during the standard procedure of the ADOS-2. For detailed description of the material, see Chapter 1. These two tasks approximate the interactional contexts of four genres (described below) of the LOuvain Corpus of Annotated Speech-French (LOCAS-F) used by Degand and colleagues to analyze BDUs. LOCAS-F is a multi-genre corpus consisting of 48 samples divided across 14 different genres13. Degand, Martin, & Simon (2014) define discourse genres according to three situational criteria: 1) degree of interactivity between the participants, 2) degree of preparation of the discourse and 3) media nature of the  13 The description of the LOCAS-F composition comes from a talk given by Anne Catherine Simon at the conference ‘Journée d’étude Toulousaine’ in 2015.  83  discourse. Degree of interactivity could be characterized as non-interactive (e.g., political discourse, sermon), semi-interactive where freedom/possibility to interrupt is limited (e.g., radio interview), or interactive where speech is freely distributed (e.g., informal conversation, conversational narrative). Degree of preparation could be characterized as non-prepared/spontaneous (e.g., informal conversation, formal (professional) conversation), semi-prepared whereby the discourse topic is known to participant (e.g., radio interview, political debate) and fully prepared/read discourse (e.g., scientific conference, political discourse). Finally, degree of media nature could be characterized as non-media (e.g., informal and formal conversation, sociolinguistic interview), semi-media nature whereby the situation implies several communicative roles and whereby participants do not address the public directly and media nature whereby the discourse output is produced solely for the purpose of being broadcasted.  According to these criteria, the discourse output used to determine the BDUs in this study, viz. semi-structured interview questions, is characterized as semi-interactive, non-prepared and non-media. Four out of the 14 genres analyzed in LOCAS-F were comparable to the data of this study, viz. informal conversation, formal conversation, conversational narrative and socio-linguistic interview. Table 32 summarizes the characteristics of the four genres.   Table 32 Characteristics of the discourse genres informal conversation, formal conversation, conversational narrative, sociolinguistic interview and semi-structured questions  Genre Interaction Preparation Media LOCAS-F Informal conversation interactive non-prepared non-media  Formal conversation  interactive non-prepared non-media  Conversational narrative interactive non-prepared non-media  Sociolinguistic interview semi-interactive semi-prepared non-media ADOS-2 Semi-structured questions semi-interactive non-prepared non-media  2.3. Procedure  The audios recordings of the two tasks were processed in Praat (Boersma & Weenink, 2017) according to the procedure described in Chapter 1.     84  2.4. Analysis   All statistical analyses were conducted in R (R Core Team, 2016) with generalized logistic  models (Poisson family). Discourse productivity was measured as raw counts of total words, syntactic sequences and syntactic units. To analyze group differences in these measures, models were built with total counts of words, syntactic sequences, syntactic units and BDUs respectively as dependent variable and group diagnosis as fixed effects. The significance of the model was determined by comparing it to a model without the fixed effect of diagnostic group using the anova function from the ‘stats’ package.  To analyze group differences in subtypes of syntactic units and BDUs, models included feature type as dependent variable and the fixed effects of group diagnosis as well as total syntactic units and total BDUs to control for the variability of these measures on the subtypes of syntactic units and BDUs respectively. To examine which specific types differed per group, Tukey post-hoc analyses were conducted using the emmeans function from the ‘emmeans’ package (Lenth, 2016). All significant effects reported in this chapter remained so when controlling for total number of syntactic units and BDUs.  Considering the variability in measurements (count data), percentage scores were also calculated for a more homogeneous representation of the data. To visualize the proportion of a given subtype of syntactic unit, percentage scores were calculated as the total count of a given syntactic feature divided by the total number of syntactic units. Likewise, to visualize the proportion of each BDU subtype, percentage scores were calculated as the total count of given BDU divided by the total number of BDUs. These percentage scores are presented in the tables of summary statistics alongside the raw counts. Models and plots were created using the raw scores of the dependent variables. As in Chapter 2, violin plots were used to illustrate the data.   III. Results  3.1. Syntactic coding  3.1.1. Discourse productivity : words, sequences and syntactic units  Autistic participants produced more words, syntactic sequences and syntactic units overall than neurotypical participants, χ2(1) = 757.59, p < .0001; χ2(1) = 151.29, p < .0001 and χ2(1) = 205.41, p < .0001. See Table 33.  85  Table 33 Regression coefficients of the generalized link model with the additive effect of diagnostic group (ASD diagnosis is the reference level, standard errors is in brackets)  Total words Total syntactic sequences Total syntactic units NT -0.17698 (0.018)***  -0.31859 (0.036)*** -0.41142 (0.034)*** Signif. codes :  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1  Table 34 provides summary statistics of the different variables of discourse productivity.  Table 34 Means and standard deviations (in brackets) of total words, total syntactic sequences and total syntactic units  ASD NT Total words 1188.83 (289.33) 996.00 (274.35) Total syntactic sequences14 297.50 (70.83) 216.33 (51.83) Total syntactic units 364.67 (67) 241.64 (60.84)  In the following figures of violin plots, the median is indicated by a red dot and the mean is indicated by a diamond shape. Figure 15 displays violin plots depicting the data distribution of the following variables: total words, syntactic sequences and syntactic units.      14 The unexpected lower counts of syntactic sequences, subunits of dependency clauses, compared to the total count of syntactic units is due to the phenomenon of overlapping speech, which students were instructed to code as % (not analyzed). In cases where only a small part of the dependency clause was overlapping, the dependency clause was still included in the analysis, but the overlapping sequence or sequences were coded as % (and hence not included in the final analysis).   86  Figure 15 Violin plots of total words (plot A), syntactic sequences (plot B) and syntactic units (plot C) per diagnostic group   3.1.2. Dependency clauses When controlling for total syntactic units, there were no differences in total dependency clauses (z = 0.193, p = 0.84)  complete dependency clauses (z = 0.226, p = 0.82) and incomplete dependency clauses (z = -0.060, p = 0.9524). Table 35 contains summary statistics for the variables total dependency clauses as well as complete and incomplete dependency clauses.       87  Table 35 Means and standard deviations (in brackets) of counts and percentage of total dependency clauses, complete dependency clauses and incomplete dependency clauses per diagnostic group  ASD  NT   Counts Percentage Counts Percentage Total dependency clauses 187.50 (45.90) 51.83 (4.51) 138.17 (35.24) 57.17 (2.43) Total complete dependency clauses 166.83 (39.39) 89.74 (5.35) 122.67 (35.25) 88.25 (3.85) Total incomplete dependency clauses 20.67 (10.73) 10.53 (5.35) 15.50  (4.04) 11.75 (3.85)  Figure 16 contains violin plots displaying the distribution of these variables.  Figure 16 Violin plots for total dependency clauses (plot A), complete dependency clauses (plot B) and incomplete dependency clauses (C) per diagnostic group    88  When analyzing subtypes of complete dependency clauses, there were no group differences in verbal (z = 0.778, p = 0.35). Neurotypical participants produced more elliptic dependency clauses than autistic participants, z = 4.775, p <.0001. Autistic participants produce more averbal dependency clauses than neurotypical participants, z = 1.970, p = 0.04. Table 36 contains summary statistics of verbal, averbal and elliptic dependency clauses.   Table 36 Means and standard deviations (in brackets) of counts and percentage of dependency clause subtypes per diagnostic group  ASD NT  Counts Percentage Counts Percentage Verbal dependency clauses 106.67 (28.14) 56.62 (4.55) 74.83 (26.33) 53.77 (7.75) Averbal dependency clauses 47.67 (13.87) 25.90 (6.03) 29.00 (17.97) 20.32 (9.35) Elliptic dependency clauses 12.50 (6.77) 6.95 (3.97) 18.83 (5.49) 14.16 (5.17)  Figure 17 contains violin plots depicting the data distribution of these three types of dependency clauses.    89  Figure 17 Violin plots for complete verbal (plot A), averbal (plot B) and elliptic (plot C) dependency clauses per diagnostic group   3.1.3. Additional syntactic categories: adjuncts, discourse-structuring devices and hesitation markers Autistic participants produced less discourse-structuring devices, z = -3.157, p = 0.002 and less adjuncts, z = -2.981, p = 0.003. There were no group differences in hesitation markers, z = 0.028, p = 0.98. Table 37 contains summary statistics of the additional coding categories discourse-structuring devices, adjuncts and hesitation markers. Figure 18 contains violin plots depicting the distribution of the different additional syntactic categories.    90  Table 37 Means and standard deviations (in brackets) for counts and proportions (in italics) of discourse-structuring devices, adjuncts and hesitation markers per diagnostic group  ASD NT  Counts Percentage Counts Percentage Discourse-structuring devices 103.33 (36.44) 28.37 (6.39) 91.67 (22.76) 38.02 (3.10) Adjuncts 9.50 (5.09) 2.66 (1.55) 11.83 (5.38) 4.81 (1.56) Hesitation markers 43.67 (23.02) 13.43 (8.76) 31.83 (14.68) 14.09 (7.82)   Figure 18 Violin plots for discourse-structuring devices (plot A), adjuncts (plot B) and hesitation markers (plot C) per diagnostic group      91  3.2. Basic discourse units  There was a significant group difference in total number of BDU produced, χ²(1) = 7.2084, p = 0.007. Overall, autistic participants produced significantly more BDUs than neurotypical participants. When looking at types of BDUs separately, and controlling for total number of BDU, autistic participants produced more silent-bound BDUs, z = 2.625, p = 0.008 but less regulatory BDUs and mixed BDUs than neurotypical participants, z = -1.957, p = 0.05 and z = -2.569, p = 0.01, respectively. There were no significant group differences for congruent and syntax-bound BDUs, z = 0.333,  p = 0.74 and z = -1.727,  p = 0.08, respectively. Table 38 contains summary statistics of total BDUs as well as the five BDU subtypes. Figure 19 contains violin plots depicting the data distribution of the different BDU subtypes.   Table 38 Means and standard deviations (in brackets) for counts and percent (in italics) of total BDUs, congruent BDUs (bdu-c), silence-bound BDUs (bdu-sil), regulatory BDUs (bdu-r), syntax-bound BDUs (bdu-s) and mixed BDUs (bdu-x) per diagnostic group  ASD NT  Counts Percentage Counts Percentage bdu-c 47.50 (12.79) 47.08 (6.21) 38.50 (19.32) 44.20 (9.52) bdu-sil 38.17 (9.02) 38.67 (8.94) 24.00 (8.49) 29.15 (6.65) bdu-r 4.33 (3.72) 4.08 (2.70) 6.00 (4.05) 6.57 (4.13) bdu-s 4.00 (3.10) 3.70 (2.24) 5.33 (2.66) 6.91 (4.07) bdu-x 6.83 (4.88) 6.47 (4.21) 9.67 (2.50) 13.16 (6.61) Total BDU 102.67 (23.39)  65.67 (31.37)      92  Figure 19 Violin plots for total BDUs (plot A), congruent BDUs (plot B), silence-bound BDUs (plot C), regulatory BDUs (plot D), syntax-bound BDUs (plot E) and mixed BDUs (plot F) per diagnostic group   93    Table 39 summarizes the distribution of BDU subtypes within each group for a better visualization of the tendencies of BDU distribution within group.    Table 39 BDU distribution within diagnostic group BDU Types ASD NT  Total counts Percentage Total counts Percentage bdu-c 285 47.11 % 231 46.11 % bdu-sil 229 37.85 % 144 28.74 % bdu-r 26 4.30 % 36 7.19 % bdu-s 24 3.97 % 32 6.39 % bdu-x 41 6.78 % 58 11.58 % Total BDU 605 100 % 501 100 %  94  As a benchmark, the BDU distribution15 in the corpus LOCAS-F is compared to the BDU distribution in the present study. Table 40 displays the BDU distribution in this study, summarized over diagnostic group. Table 41 displays the BDU distribution of four discourse genres of the LOCAS-F corpus and Table 42 displays BDU distribution across the entire LOCAS-F corpus.  Table 40 Distribution of different BDUs summarized over diagnostic group BDU Types Total counts Percentage bdu-c 516 46.65 % bdu-i 373 34.00 % bdu-r 62 5.61 % bdu-s 56 5.06 % bdu-x 99 8.95 % Total BDU 1106 100%   Table 41 BDU distribution across informal conversation (conv-i), formal conversation(conv-f), conversational narrative(conv-narr) and sociolinguistic interview(int-soc) BDU Types conv-i conv-f conv-narr int-soc  Count Percentage Count Percentage Count Percentage Count Percentage bdu-c 184 54,76 % 92 47,92 % 33 28,45 % 103 46,60 % bdu-i 104 30,95 % 38 19,79 % 57 49,14 % 43 19,45 % bdu-r 16 4,76 % 20 10,41 % 10 8.62 % 14 6.33 % bdu-s 14 4,17 % 30 15,63 % 7 6,03 % 39 17.65 % bdu-x 18 5,37 % 12 6,25 % 9 7,76 % 22 9.95 % Total count 336 100 % 192 100 % 116 100 % 221 100 %     15 The frequencies reported in tables 13 and 14 come from a talk given by Anne Catherine Simon at the conference ‘Journée d’étude Toulousaine’ in 2015.    95  Table 42 BDU distribution in the multi-genre corpus LOCAS-F BDU Types Total counts Percentage bdu-c 1225 43 % bdu-i 503 17 % bdu-r 258 9 % bdu-s 648 23 % bdu-x 241 8 % Total count 2875 100 %  As can be seen from the different tables, the distribution patterns of the different BDUs in the present study are quite similar to those of the corpus LOCAS-F. First, congruent BDU (bdu-c) is the most common type of BDU. As can be seen from Tables 12 and 13, the distribution pattern of bdu-c in this study is almost identical to that of formal conversation and sociolinguistic interview. This pattern makes sense as the ADOS interview is a blend of these two genres, viz. semi-interactive like the sociolinguistic interview but non-prepared like the formal conversation. Furthermore, analogous to the four genres of the LOCAS-F, the second most frequent BDU type in this study is silence-bound BDUs (bdu-sil). With 34 %, the frequency in this study is closest to the frequency of intonation-bound BDUs in informal conversations (30.95%). In the present study as is the case in the LOCAS-F corpus, the frequency of syntax-bound BDUs (bdu-s, 5.06%), regulatory BDUs (bdu-r, 5.61%) and mixed BDUs (bdu-x, 8.95%) were quite low.   3.3. Summary of results  The variables analyzed in this study and their associated effects are summarize in Table 43.    96  Table 43 Summary of all coding categories and their associated group effect Variable Feature Group difference Syntactic coding Total words ASD > NT  Total syntactic sequences ASD > NT  Total syntactic units ASD > NT  Total dependency clauses ASD = NT  Complete dependency clauses ASD = NT  Incomplete depedency clauses ASD = NT  Verbal dependency clauses ASD = NT  Averbal dependency clauses ASD > NT  Elliptic dependency clauses NT > ASD  Adjuncts NT > ASD  Discourse-structuring devices NT > ASD  Hesitation markers ASD = NT BDU segmentation Total BDUs ASD > NT  Congruent BDUs ASD = NT  Silence-bound BDUs ASD > NT  Regulatory BDUs NT > ASD  Syntax-bound BDUs ASD = NT  Mixed BDUS NT > ASD  IV. Discussion  In this Chapter, I asked whether autistic and neurotypical adults differed in their strategies to combine prosodic and syntactic information when delivering speech. Taken together, the results provide mixed evidence for group differences in discourse strategies. Before delving deeper into the results of the BDUs, it is important to note an interesting difference in the results of the syntactic segmentation of this study and that of the narrative study. Recall that in the latter study, autistic participants were less productive then their neurotypical peers: they produced fewer words, syntactic sequences and syntactic units overall. In the present study, the reversed pattern is observed, as the autistic participants produced more words, syntactic sequences and syntactic units overall. This pattern of results in speech productivity favors the assumption that narrative production is a difficult and demanding task for autistic individuals, even more so than conversation (Botting, 2002). In the tasks used in this study, the different questions of the experimenter provided participants with a structure to develop subsequent turns in the conversation. Furthermore, the finding that autistic participants were more verbose on topics such as solitude and relationships suggest that they were able to  97  discuss these topics, precluding any assumptions that group differences can be attributed to difficulties with conversation topic rather than to an autism diagnosis. Likewise, regarding subtypes of dependency clauses, a different pattern emerged. Specifically, in the narrative study, autistic participants produced less verbal but more averbal dependency clauses than their typical peers while elliptic dependency clauses did not differ across groups. In the present study, there were no group differences in verbal dependency clause but one in elliptic dependency clauses, with neurotypical participants producing more than autistic participants. This latter finding makes sense considering the semi-structured nature of the tasks on relationships and solitude. In these tasks the questions asked by the experimenter can provide an adequate context to omit an obligatory element of a clause when responding to the question, without impacting the meaning of the answer. For example, in (14), the experimenter asks the participant one of the interview questions on solitude, viz. what can be done in moments of solitude, and the participant responds with an elliptic clause (in bold below) omitting the verbal sequence ‘je ne fais’ (I don’t)  (14) Exp : <et> [tu fais quoi dans ces moments-là]1  ///1 <and> [what do you do at times like this]1 ///1 PP : <euh> [rien de très intéressant]1 ///1 <genre> [je patoufle]1 ///1  <uh> [nothing particularly interesting]1 ///1 <like> [I’m being lazy]1 ///1  Comparison participant (male, 21 years old)  Ellipsis is one way to mark cohesion (Halliday & Hasan, 1976) and is exploited to a greater extent by the neurotypical participants in this study than autistic participants. Furthermore, it is worth emphasizing that group differences in averbal dependency clauses and the additional syntactic category of discourse-structuring devices replicate the results of the narrative study, highlighting these features as a stable difference between the discourse of autistic and neurotypical adults. In other words, although the semi-structured tasks of Relationships and Solitude might have been less demanding than the narrative task for autistic participants, establishing and maintaining coherence during discourse production remains challenging for them. Now turning to the results of the BDUs, contrary to our initial hypothesis that autistic adults would produce more syntax-bound BDUs, reflecting processing difficulties and/or  98  pedantic style, autistic adults did not produce more syntax-bound BDUs than neurotypical adults. Similarly, the expectation that neurotypical adults would produce more silence-bound BDUs was not fulfilled and, in fact, the exact opposite trend emerged, with autistic adults producing more silence-bound BDUs than neurotypical adults. In other words, autistic adults produced more macro-units of information than neurotypical adults. However, our hypothesis that neurotypical adults would produce more regulatory BDUs than autistic adults was corroborated. An unexpected result was a group difference for mixed BDUs, with neurotypical participants producing more mixed BDUs than autistic participants. There were no group differences in the production of congruent BDUs, the baseline BDUs. Although at first glance, the direction of the study’s initial hypotheses was not fully corroborated, the results of the syntactic segmentation and the distribution pattern of the BDU subtypes in the present speech sample converge together to highlight meaningful differences between diagnostic groups, viz. differences in discourse coherence and management. First, considering that autistic participants produced less discourse-structuring devices overall than neurotypical participants, the finding that they also produced less regulatory BDUs, viz. the isolation of a discourse-structuring device with a silent pause, seems logical. In turn, the lower frequency of discourse-structuring devices leaves room to compensate with alternative means to relate different syntactic units, resulting in more silence-bound BDUs, viz. several syntactic units grouped by one silent pause. Indeed, as already mentioned in Chapter 2, one way to create discourse coherence is by establishing coherence relations across utterances or text segments (Knott & Sanders, 1998; Sanders & Noordman, 2000; Sanders, Spooren, & Noordman, 1992). Coherence relations can be marked explicitly, by means of connectives, such as ‘because’, ‘however’ or ‘and’, but they can also be left implicit, as for example, when a coherence link is conveyed through the juxtaposition of two clauses or sentences. Implicit relations require that the readers or listeners themselves infer the relation between the discourse segments using the information from the linguistic context and/or their world knowledge (e.g. Kintsch, 1998; Zwaan & Radvansky, 1998). Discourse studies on annotated corpora of discourse relations suggest that some types of relations can be more easily left implicit than others. For example, while causal and additive relations are often expressed implicitly, conditional and concession relations tend to be explicitly marked (Asr & Demberg, 2012; Taboada, 2009). According to Degand and colleagues, intonation-bound BDUs reflect an alternative way of marking coherence relations between utterances. Rather than using explicit discourse-structuring devices, the speaker can use prosody to group  99  otherwise unrelated syntactic clauses, indicating that these clauses should be interpreted as one information unit. In this sense, intonation-bound BDUs resemble an implicit coherence relation. Example (15) is a silent-bound BDU, with an implicit causal relationship between syntactic unit 1 and 2. Specifically, this causal relationship could be made explicit by adding the connective ‘parce que’ (because) in between the two syntactic units as illustrated in example (16).  (15) <et> <puis> <euh> <finalement> <bah> [je crois que les gens me supportent pas]1 [je suis toxique pour les gens]2 ///1 <and> <then> <uh> <ultimately> <well> [I think that people can’t stand me]1 [i’m toxic for people]2///1  (16) <et> <puis> <euh> <finalement> <bah> [je crois que les gens me supportent pas]1 <parce que>  [je suis toxique pour les gens]2 ///1 <and> <then> <uh> <ultimately> <well> [I think that people can’t stand me]1 <because> [i’m toxic for people]2 ///1    Autistic participant (male, 21 years old) At this point in the discussion, it is important to consider a methodological difference with the prosodic segmentation applied in this dissertation and that applied by Degand and Simon. In this study, major prosodic boundaries were identified according to one criterion only, viz. silent pauses longer than 200 milliseconds (ms). Due to the exploratory nature of the study and practical reasons, acoustic criteria such as vowel lengthening and intra-syllabic rise were not taken into account. In other words, the BDUs in this analysis reflect a rather simple prosodic signal to chunk up the on-going flow of the discourse. Neurotypical participants probably deployed a wider-range of and more subtle prosodic cues to segment their speech, which is not reflected in the current procedure. One piece of evidence for this assumption is the finding that neurotypical participants produced more mixed-BDUs than autistic participants. Recall that a mixed BDU occurs when the boundaries of the syntactic and prosodic units overlapped but did not coincide to yield any of the other types of BDUs. Consider (17) and (18), two examples of mixed BDUs.   100  (17) [c’est la première fois qu’il se voyaient]1 <tu vois> [alors que c’était trois ///1 trois tu vois  groupes de de bons potes quoi]2 ///2 [it was the first time they saw each other]1 <you know> [even though it was three ///1 three you know groups of good friends]2 ///2  Comparison participant, male, 21 years old (18) <euh> [je vais aussi une fois par semaine aux ///1 aux réunions de section pour les réanimateurs]1 <donc> [là j’ai aussi ///2 des amis]2 ///3 <uh> [also go once a week to ///1 to department meetings for the reanimator]1 <so> [there I also have ///2 friends]2 ///3 Comparison participant, male, 52 years old  Degand and Simon (2009a) do not consider mixed-BDUs as actual basic discourse units but as a left-over category, with no specific discourse strategy. Mixed BDUs could emerge as a result of production difficulties or as a result of insufficient criteria to fully capture the speech being produced. In light of the prosodic segmentation implemented in this study, the distribution pattern of less silence-bound BDUs and more mixed BDUs in the speech of neurotypicals could reflect the latter possibility. For example, in (17), it is possible that the discourse marker you know bears a specific prosodic contour that would lead to the detection a major prosodic boundary. This would lead to a change from one mixed-BDU to two BDUs, viz. one congruent BDU (19) and one syntax-bound BDU (20).  (19) [c’est la première fois qu’il se voyaient]1 <tu vois> ///1 [it was the first time they saw each other]1 <you know> ///1 (20) [alors que c’était trois ///1 trois tu vois groupes de de bons potes quoi ///2]1  [even though it was three ///1 three you know groups of good friends ///2]1  Comparison participant (male, 21 years old)  101  Future studies should examine whether adding the acoustic criteria proposed by Degand and Simon would reduce the frequency of mixed BDUs and increase the frequency of other types of BDU. Future studies could also examine how the proportion of the different BDU subtypes vary as a function of the types of prosodic criteria, viz. simple and gross prosodic features (e.g., silent pauses) and more complex and fine-grained ones (e.g., intra-syllabic rise). Methodologically, such studies would provide further insight into which prosodic features are most reliable and stable to determine prosodic boundaries. Clinically, such studies would also inform on which features might distinguish best between the discourse style of autistic and neurotypical individuals. While refining the methodology might provide further insight into the pattern of silence-bound and mixed BDUs, it is also worth exploring the research avenue that mixed BDUs reflect production difficulties (Degand & Simon, 2009a). Consider the following example of mixed BDUs. Several cues of processing difficulties, viz. hesitation markers (uh, uhm) and repetitions (without any problem) and reformulation (well without /// without delay) suggest that the participant is having real troubles delivering her message.  (21) [y’a rien à dire]1 <euh> [c’est euh c’est quelqu’un sur qui on peut compter]2 <euhm> [c’est quelqu’un qui va te ///1 pardonner ///2 sans sans problème fin sans ///3 sans delai]3 ///4 [there's nothing to say]1 <uh> [it's uh it's someone you can count on]2 <uhm> [it's someone who's going to ///1 forgive you ///2 without any problem well without ///3 without delay]3 ///4 Comparison participant (female, 39 years old)  While (re)defining reliable segmentation criteria might help reduce some instances of mixed BDUs to actual basic discourse units, there might always be a ‘left-over’ category, reflecting deeper production difficulties.   V. Conclusion  Taken together, the present study provides evidence that Basic Discourse Units can distinguish the speech delivery strategies of autistic adults from those of their neurotypical peers, even based on simple prosodic cues like silent pauses. Specifically, autistic participants used silent pauses to create macro-units of information, grouping several syntactic units within a silent -pause boundary. One outstanding issue is to examine in more details the composition of such macro-units of information. How many dependency clauses, sequences and words do they  102  contain? For example, Degand & Simon (2009a) have observed that conversational narration has shorter dependency clauses (6.2 words per dependency clause) than radio news (16.7 words per dependency clause), political address (15.6 words per dependency clause) and conference talk (15.1 words per dependency clause). Likewise, the duration of a major prosodic units differs per discourse genre. For example, in conference talks, major prosodic units last on average 4.1 seconds while they only last 2.6 seconds in political address. Analyzing the composition of the different BDUs would provide insights into the information density and structure of the different BDUs. For example, the frequency of congruent BDUs was similar for autistic and neurotypical participants. However, considering that autistic participants produced more words and dependency clauses, their congruent BDUs might be ‘denser’ than those of neurotypical participants. Finally, only one discourse genre was investigated in this study. Hence, an outstanding issue is to examine whether silence-BDUs’ reflect a general preference of autistic participants to deliver their discourse or whether this strategy is genre-bound. This hypothesis can be tested by segmenting into BDUs the data of the storybook narratives analyzed in Chapter 2.      103  Chapter 4: Relating corpus findings to their impressions by naïve listeners  A rating experiment on judgments of spoken discourse and impression formation of autistic adults  I. Introduction  Chapter 2 and 3 focused on identifying linguistic features that distinguish the speech of autistic and neurotypical adults. To this aim, transcript analyses were performed, highlighting both features related to the organization of the discourse content (Chapter 2) and features related to strategies of delivering this content (Chapter 3). These analyses have allowed for a singling out of a set of speech characteristics that distinguishes the two groups, such as connectives, discourse markers as well as regulative BDUs. However, it is not possible to determine whether and how meaningful these features are beyond a transcript as they have been decontextualized from their production environment, that is, from the social interaction in which it was produced. For example, transcript analyses of the storybook narrative in Chapter 2 showed that autistic adults produced significantly fewer additive, causal and contrastive connectives in their narratives in comparison to their neurotypical peers. What these results do not tell us is whether naïve listeners would actually notice this difference and whether it would influence their perception of discourse coherence. This is the aim of the current chapter.  In recent years, several studies sought to investigate which verbal and non-verbal characteristics are linked to atypical social presentations in autism. Most of these studies have focused on features related to atypical prosody. For example, Bone et al. (2015) examined which objective acoustic features could be related to subjective perceptions of prosodic awkwardness. To this aim, speech samples from story retellings of autistic and neurotypical adolescents were analyzed for prosodic atypicalities often associated with autism (intonation, volume, rate, and voice quality). In addition to these analyses, a group of naïve raters (blind to stimulus participants’ diagnoses) were recruited on Mechanical Turk to rate the way the stories were told along the dimensions of awkwardness and expressivity. Results of the perceptual analysis of awkward prosody suggest that subjectively, more awkward speech can be characterized as less expressive (more monotone) and more often involves perceived  104  awkward rate/rhythm, volume, and intonation. The results of the acoustic analysis suggest that cues related to timing such as speaking rate and rhythm are highly predictive of perceived awkwardness. Using a classification task, the authors also show that acoustic-prosodic features can significantly distinguish autistic participants from neurotypical participants.   The study by Bone et al. (2015) suggests that naïve listeners are able to perceive and interpret specific acoustic features as atypical. Their study is limited in that it focused exclusively on auditory modality. Furthermore, participants were asked to rate the prosodic qualities, but not the speaker per se. As such, this study does not allow for a determination of whether these perceptions would also result in perceiving the speakers themselves as atypical. The study by Grossman (2015) addresses these limitations, as she examined, across different modalities, the first impressions typical adults formed of high-functioning autistic children and neurotypical children. More specifically, the authors examined whether high functioning autistic children would be perceived as more socially awkward than their neurotypical peers based on very brief exposure to static (still images) and dynamic information (audio-visual information). During a story-telling task, audio and video recordings were made of children with and without a diagnosis of autism. These recordings were turned into 1 s and 3 s clips of different modalities: audio-only, visual-only, audio-visual and still images. A group of typical adults was recruited and presented with these clips and static images. The adults, who were blind to the diagnostic status of the children, were instructed to judge whether the person in the clip was socially awkward. The author found that typical adults evaluated high-functioning autistic children as socially awkward significantly more frequently than their typical peers. These results remain consistent for brief exposures (1 second) to dynamic information such as visual and/or auditory cues, as well as for still images. It seems that typical adults make use of various subtle cues produced by high functioning autistic children to form judgments of social awkwardness, potentially having significant repercussions on the on-going social interaction. Sasson et al. (2017) expanded the findings of these two studies by exploring the association between impressions of social awkwardness of autistic individuals and subsequent behavioral dispositions. Across different experimental settings, the authors found that the first impressions neurotypical adolescents and adults formed of autistic adolescents were more negative than those they formed of their matched typical peers. Furthermore, these first impressions remained less favorable across different modalities (audio, visual and static images) and across short (2–4 s) as well as longer (10 s) examples of social behavior. These impressions  105  also remained consistent along repeated exposure to the same stimuli. Most importantly, negative first impressions of autistic adolescents were associated with a reduced disposition to initiate or pursue social interaction with these individuals. This latter finding suggests that the social difficulties that autistic individuals experience might not only emerge from their own communication difficulties but also from negative biases and attitudes adopted by neurotypical individuals. Echoing assumptions made by Perkins (2007; 2010) and Sterponi et al. (2014), Sasson et al. (2017) postulate that the social interaction difficulties experienced by autistic individuals are not only an individual impairment but also one arising in the relation between the autistic individual and his/her conversational partner.  In addition to overlooking the relational dimension of social impairments in autism, previous studies have also overlooked the perspective of autistic individuals. Both aforementioned studies have examined atypical social presentation from a one-sided perspective, viz. how neurotypical individuals perceive autistic individuals in comparison to other neurotypicals, but the reverse scenario has not been investigated yet. What about the perception of autistic individuals of other autistic and neurotypical peers? For example, it could be the case that when listening to other people, autistic people do not attend to the same discourse features and hence register these features and their function in discourse differently than neurotypical individuals. If this is the case, perceived discourse (in)coherence and atypicality would be driven by different features for autistic and neurotypical people. Thus, communication difficulties might arise from different expectations about discourse quality from autistic and neurotypical individuals. This latter conclusion lines up with the ‘double empathy problem’, a term coined by self-advocate Damian Milton (2011a). Milton (2012) describes the ‘double empathy problem’ as “a breach in the ‘natural attitude’ that occurs between people of different dispositional outlooks and persona conceptual understandings when attempts are made to communicate meaning.” According to this definition, it is a ‘double problem’, as both individuals involved in the communication experience this breach: the problem is not specific to any of the two interlocutors, but rather it is in the social interaction between two social actors with different dispositions. However, while there seems to be a bias against autistic people, this bias is modulated by the medium by which the person is portrayed. Sasson et al. (2017) found that autistic individuals were rated as socially more awkward than their typical peers when these impressions were based on audio, visual and static stimuli. However, this difference did not hold when impressions of social awkwardness were based on transcripts of the spoken  106  discourse content. The authors suggest that style more than content influences negative impressions of ASD. This latter conclusion seems to contradict a finding by Canfield, Eigsti, De Marchena, & Fein (2016). These authors included subjective ratings of narrative quality, in addition to the traditional transcript analyses of narrative production, in their analyses. They collected picture-based narratives from three experimental groups of adolescents; adolescents with optimal outcome16, high-functioning autistic adolescents and neurotypical adolescents. The rating scale included four questions targeting different narrative qualities, i.e., story goodness (“How good a story is this?”), cohesiveness (“How well were you able to follow this story?”), accuracy (“How well does this story reflect the actual story in the cards?”) and oddness (“How odd/unusual did you find this story?”). The results of this study are interesting in two ways. First, while transcript analyses did not show any significant differences in overall narrative quality between the adolescents in the optimal outcome group and the high-functioning autistic participants, participants in the optimal outcome group received significantly lower ratings of story goodness and story cohesiveness than neurotypical participants. Therefore, even in individuals with an optimal outcome of autism, subtle pragmatic difficulties continue to be present. As for the high-functioning autistic participants, they differed from the neurotypical group in all four rating domains: goodness, cohesiveness, accuracy, and oddness. Second, these rating differences were obtained on the basis of reading narrative transcripts which did not include any prosodic cues or other personal characteristics, restricting their potential effect on the perceptions of narrative quality. Canfield et al. (2016) and Sasson et al. (2017) reach a similar conclusion, namely that there are robust cues that drive negative impressions of autistic adults, although these conclusions are based on opposite features (content vs. prosody and other personal characteristics). Hence, awkwardness and incoherence are both attributes often used to qualify language in autism but do not seem to be correlated. A speaker can be awkward without necessarily being incoherent, or vice versa. Taken together, these two studies suggest that perceived social awkwardness might be driven more strongly by style/manner of speaking, while (in)coherence seems to be driven more strongly by content.    16 Autism Spectrum Disorder is described as a lifelong condition (American Psychiatric Association, 2013). However, some studies have suggested that there is a subset of autistic people, who improv so significantly that they no longer meet diagnostic criteria for autism, achieving ‘optimal outcome’ (e.g., Fein et al. 2013; Fein et al. 2005; Kelley et al. 2006, 2010). Individuals with Optimal Outcome (OO) usually score in the average range in a range of measures such as cognitive, linguistic, adaptive, behavioral and social measures (Fein et al. 2013; Helt et al. 2008). However, they may still show subtle residual difficulties in higher-order language functions.  107  Outlook The few existing studies on the perception of autistic individuals suggest that they are perceived less positively, both with respect to the content and the style of their discourse. However, to date, no study has sought to directly relate the linguistic features identified in transcripts with the impression the speakers trigger in their audience. Furthermore, studies on impression formation have been limited to the impressions of neurotypicals. However, if we want to accurately define the socio-communicative difficulties emerging from a social interaction between two individuals, it is important to account for the social dynamic of this interaction, viz. that it is two-way. In that respect, as much as neurotypicals form impressions about others, so do autistic individuals - and their perspective on the interaction also needs to be represented.  The aim of the study presented in this chapter was to address these two gaps by investigating the perception of different discourse features by naïve listeners, and their contribution to impression formation of the speaker. Furthermore, this study sought to evaluate the relation between discourse abilities and speaker impression from the perspective of both autistic and neurotypical adults. To do so, a scale was developed to assess the audio recordings of conversational dyads between a participant with or without a diagnosis of autism and an experimenter. Two new groups of autistic adults and pairwise matched neurotypical adults were recruited to rate these audio recordings. I hypothesized that audios recordings with autistic speakers would receive lower ratings on all scale items than those with neurotypical speakers by both autistic and neurotypical participants. Considering the novelty of the scale it was first pretested on a neurotypical sample before using it with the two new groups of participants.  II. Pretest  2.1. Methodology  2.1.1. Material  Experimental stimuli consisted in audio extracts from Module 4 of the Autism Diagnostic Observation Schedule-2 (ADOS-2, Lord et al., 2012). The audio extracts of the present study cover the topics of friendship, relationships and marriage described in Chapter 1. To respect confidentiality, any identifying information was removed from the audio extracts. Audio extracts from twelve participants were selected for the pretest (3 male autistic participants, 3  108  neurotypical participants, 3 female autistic participants, 3 neurotypical participants.) The length of the extracts was on average 407.17 seconds (SD = 98.38 seconds; range= 225-606 seconds).  The scale designed for the pretest study involved 13 items, 9 of which targeted the content of the answers of the participants (C01-C09) and 4 of which targeted the participant’s impression of the speaker (S10-S13). The content items targeted both qualities relating to the structure (C01, C02, C03, C04, C05) and style of the answers (C06, C07, C08, C09). See Table 44 for a description of the scale items.  Table 44 Scale Items  Item Question Item Response C01 Par rapport aux questions posées, le contenu des réponses est With respect to the questions asked, the content of the responses are Hors de propos 1   2   3   4   5   6   7 A propos Irrelevant 1   2   3   4   5   6   7 Relevant C02 Est-il facile de suivre le fil de la conversation ? Is it easy to follow the conversation Difficile 1   2   3   4   5   6   7 Facile Difficult 1   2   3   4   5   6   7 Easy C03 La structure des réponses vous paraît The structure of the answers seems  Illogique 1   2   3   4   5   6   7 Logique Illogical 1   2   3   4   5   6   7 Logical C04 Par rapport au style des questions posées, le type d’information fourni dans les réponses est-il approprié ? With respect to the style of the questions asked, is the type of information provided in the answers is  Inapproprié 1   2   3   4   5   6   7 Approprié Inappopriate 1   2   3   4   5   6   7 Appropriate C05 Le niveau de détails fourni par les réponses est The level of detail provided in the response is  Insuffisant 1   2   3   4   5   6   7 Excessif Insufficient 1  2   3   4   5   6   7 Excessive C06 Les réponses paraissent  The answers seem Artificielles 1   2   3   4   5   6   7 Naturelles Artifical 1   2   3   4   5   6   7 Natural C07 Les réponses sont énoncées de manière :  The answers are produced in a way that is Saccadée 1    2    3    4    5    6   7 Fluide Choppy 1    2    3    4    5    6   7 Fluent C08 Par rapport au style des questions posées, le style des réponses vous paraît Formel 1    2    3    4    5    6   7 Informel  109  With respect to the style of the questions asked, the style of the responses are Formal 1    2    3    4    5    6   7 Informal C09 Par rapport aux différentes questions posées, les réponses sont : With respect to the different types of questions asked, the answers are Répétitives 1   2   3   4   5   6   7 Variées Repetitive 1   2   3   4   5   6   7 Varied S10 Si vous deviez faire connaissance avec cette personne, vous faire comprendre par celle-ci serait If you had to get to know this person, making yourself understood by this person would be Difficile 1   2   3   4   5   6   7 Facile Difficult 1   2   3   4   5   6   7 Easy S11 Si vous deviez faire connaissance avec cette personne, celle-ci vous comprendrait : If you had to get to know this person, s/he would understand you Difficilement 1   2   3   4   5   6   7 Facilement With difficulty 1   2   3   4   5   6   7 Easily S12 Dans la vie de tous les jours, cette personne deviendrait mon ami(e) In everyday life, this person would become my friend En aucun cas 1   2   3   4   5   6   7 Certainement Absolutely not 1   2   3   4   5   6   7 Absolutely yes S13 Est-ce que la personne répondant aux questions ressemble à la plupart des gens ? Is the person answering the questions similar to most people? Pas du tout 1   2   3   4   5   6   7 Beaucoup Not at all 1   2   3   4   5   6   7 A lot  2.1.2. Participants Participants were recruited on the online crowdsourcing platform Prolific. To access the questionnaire, participants had to meet the following criteria: 1) be native French speakers and 2) be of French, Belgian or Swiss nationality. Thirty-two participants (19 male, 13 female) completed the questionnaire online, mean age = 26.34 (SD = 6.26; age range = 20-44 years).   2.1.3. Procedure The audio recordings and scale items were implemented as a questionnaire on LimeSurvey (Schmitz, 2012). Participants accessed the LimeSurvey questionnaire via the crowdsourcing platform Prolific. The instructions told participants that they would be presented with several audio extracts involving two speakers, and that they would be asked to evaluate the speaker  110  answering the questions (not the one asking them). They were instructed to listen to each audio in full before progressing to the rating questions. The questionnaire started with a practice item, allowing participants to familiarize themselves with the scale. Test audios were then presented one by one, each followed by the thirteen rating items. In four recordings, an attention check question was added to the rating items. The twelve recordings were divided into two parts and two versions (with the order of two parts counter-balanced). The second part of the questionnaire included three additional questions asking what participants thought were the most important 1) quality and 2) goal of a conversation, as well as, 3) their guess about the purpose of the study.   2.2. Analysis  The effects of speaker diagnosis (speaker in the audio) of the audio on participants’ ratings were analyzed with cumulative link mixed models in R (R Core Team, 2016) using the clmm function from the ‘ordinal’ package (Christensen, 2015). Scale reliability, viz. the scale’s internal consistency was analyzed using the alpha function from the ‘psych’ package (Revelle, 2017).  2.3. Results  2.3.1. Internal consistency of the scale  Cronbach’s alpha (α) was used to measure the internal consistency of the scale. The total scale consisted of 13 items (α = .87), of which 9 items made up the content subscale (α = .80) and 4 items the speaker subscale (α = .77). Dropping two items of the content subscale (C05 & C08) increased inter-item correlation. Removing these items from the scale increased Cronbach’s alpha for the total scale (α = .90) and the content subscale (α = .84). After the removal of these two items, the final questionnaire included 11 items and was used in the auditory rating experiment examining raters’ impressions of discourse abilities and speaker (described in Section III).      111  2.3.2. Ratings  A cumulative link mixed model with the fixed effects of speaker diagnosis and questionnaire version and participant as random factor revealed a significant effect of diagnosis on mean rating scores, χ²(1) = 98.876, p <.0001 but no significant effect of questionnaire version on scale items, χ²(1) = .657, p=0.198; see Table 45.   Table 45 Cumulative link mixed model with additive effects of speaker diagnosis  and questionnaire version. Includes participant as random factor (ASD diagnosis  and Version 1 are the reference levels, standard errors are between brackets) Rating Scale NT 1.8664 (0.197) *** Version 2 -0.6363 (0.508) Log Likelihood -1381.43 Number of observations 384 Signif. codes :  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1  On average, audios with neurotypical speakers received higher ratings than audios with autistic speakers, suggesting better perceived discourse abilities of neurotypical speakers as well as a better impression of that speaker, see Table 46.  Table 46 Mean and standard deviations (in brackets)  of total rating scores per speaker diagnosis Mean Rating Score (SD) ASD 4.48(1.06) NT 5.38(1.04)  2.4. Summary of results  The pretest validated the rating scale. Naïve neurotypical listeners, blind to the diagnostic status of the speakers in the audios, rated neurotypical speakers significantly higher than autistic speakers. This suggests that the discourse characteristics included in the scale successfully discriminated the speech of autistic and neurotypical speakers     112  III. Auditory Rating Experiment   3.1. Methodology  This study received ethical clearance from the Ethics Committee of the Faculty of Psychology and Education at Université libre de Bruxelles. Written consent was obtained from all participants or their parents.  3.1.1. Participants  Participants included 18 French-speaking autistic adults and 18 neurotypical adults. Autistic participants were recruited via the Autism in Context: Theory and Experiment (ACTE) register of volunteers or by word of mouth. Participants in the comparison group were recruited via advertisements on the internet or by word of mouth. Inclusion criteria for both groups included: 1) age between 16 and 60 years, 2) a Full-Scale IQ (FIQ) score above 70, 3) Verbal IQ (VIQ) score above 70 and 4) normal or corrected-to-normal vision and audition. For the control group, there was the extra inclusion criterion of no known psychiatric, developmental or neurological disorder. Participants were matched individually on age and gender and were group matched on FIQ and VIQ.  Autistic participants had all been diagnosed with autism by professional clinicians. If ADOS scores were not available for participants at the time of the study, their diagnosis was confirmed by administering Module 4 of the ADOS-2 (Lord et al., 2012). Furthermore, as advised by Baron-Cohen et al., the Empathy Quotient (EQ; Baron-Cohen & Wheelwright, 2004) was administered conjointly with the Autism Quotient (AQ; Baron-Cohen, Wheelwright, Skinner, Martin, & Clubley, 2001), which provides an estimate of autistic-like traits presented by an individual, and allows for situating them on the continuum from autism to neuro-typicality. As can be seen from Table 47, the ADOS score indicate that autistic participants all met the autism cut-off (> 7). Furthermore, the AQ scores of participants in the autism group were significantly higher than those of the participants in the comparison group. Participants in the comparison group scored significantly higher on the EQ.      113  Table 47 Descriptive statistics of participants’ characteristics (ASD is the reference level)  ASD NT t df p N (M : F) 18 (10 :8) 18 (10 :8)    Age 31.39 (8.40) 31.33 (8.46) 0.01977 33.999 0.9843 Total ADOS-2 Score 11.67 (5.01)  NA    AQ 38.65 (4.14) 13.36 (5.44) 14.31 23.914 < 0.0001 EQ 19.18 (7.58) 38.29 (10.19) -5.813 23.575 < 0.0001  Participants’ Intellectual Quotient (IQ) was assessed using the full version of the Wechsler Adult Intelligence Scale (WAIS-IV; Wechsler, 2008). As can be seen from Table 48, autistic and neurotypical participants did not differ in Full-scale IQ (FIQ), Verbal IQ (IQ) or Perceptual IQ (PIQ).   Table 48 Means and standard deviations (in brackets) of IQ scores (ASD is the reference level)  ASD NT t df p FIQ (SD) 116.33 (14.32) 118.81 (12.54) -0.538 31.996 0.594 VIQ (SD) 121.71 (19.90) 124.12 (14.03) -0.405 28.797 0.688 PIQ (SD) 113.41(16.29)s 108.12 (12.64) 1.045 29.944 0.304  3.1.2. Material The same audio recordings used in the pretest and the revised rating scale (11 items) were used in the present rating experiment.  3.1.3. Procedure Participants were either tested at Université libre de Bruxelles, at their institution or at their home, according to their convenience. During the first session, participants’ IQ was evaluated, and at the end of the session they were given a questionnaire to fill in at home. During the second session, participants performed two experimental tasks on a computer which are not part of this dissertation. Finally, during the third session, participants performed the rating task. The rating task was implemented in LimeSurvey and was self-paced. Before starting the rating task, participants were given time to familiarize themselves with the scale items. Participants were instructed that they would listen to two sequences of 6 audios each. After each audio, they will be asked to rate the speaker answering the questions of the experimenter. At the  114  end of the entire task, participants answered three additional questions asking what participants thought were the most important 1) quality and 2) goal of a conversation, as well as, 3) their guess about the purpose of the study.  3.2. Analyses  The effects of the diagnosis of the speakers in the audios, henceforth speaker diagnosis, on participants’ rating, henceforth rater diagnosis, of the audio recordings were analyzed with cumulative mixed models in R (R Core Team, 2016) using the clm function from the ‘ordinal’ package (Christensen, 2015). For each dependent variable (total rating score and scores of individual scale items) a model was created with speaker diagnosis and rater diagnosis entered as fixed effects. Significance of fixed effects was assessed by performing likelihood ratio tests relative to a model without the fixed effects. When appropriate, interaction effects between speaker diagnosis and rater diagnosis were also tested. Post-hoc analyses were evaluated with the emmeans function from the ‘emmeans’ package (Lenth, 2016).  IV. Results  4.1. Full Scale  There was a main effect of speaker diagnosis on the overall ratings of the scale, χ2(1) = 200.52, p < .0001 and rater diagnosis, χ2(1) = 8.57, p = 0.003. See Table 49.  Table 49 Cumulative link model with additive effects of speaker diagnosis and rater diagnosis  (ASD diagnosis is the reference level for both effects, standard errors are between brackets) Total Rating Score NT Speaker 0.718 (0.051)*** NT Raters 0.155 (0.050)** Log Likelihood -8940.10 Number of observations 4917 Signif. codes :  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1  As can be seen from Table 50 below, autistic speakers received lower scores than neurotypical speakers. Autistic raters gave lower scores than neurotypical raters.    115  Table 50 Means and standard deviations (in brackets) of rating scores per speaker diagnosis  and rater diagnosis Total Rating Score (SD) ASD Speaker NT Speaker ASD Rater NT Rater 5.40 (1.69) 6.06 (1.56) 5.66 (1.66) 5.80 (1.66)  There was also a significant interaction between speaker diagnosis and Rater diagnosis, χ2(1) = 43.985, p < .0001, see Table 51.  Table 51 Cumulative link model with additive effects of speaker diagnosis and rater  diagnosis and interaction effect between speaker diagnosis and rater diagnosis (ASD  diagnosis is the reference level for all effects, standard errors are between brackets) Total Rating Score NT Speaker 0.39063 (0.07113)*** NT Raters -0.18309 (0.07161)* NT Speaker * NT Rater 0.66850 (0.10095)*** Log Likelihood -8918.11 Number of observations 4917 Signif. codes :  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1  Post-hoc analyses revealed that audio recordings with autistic speakers received similar rating scores by the autistic and neurotypical raters (z = 2.557, p = 0.052). Audio recordings with neurotypical speakers received higher scores than those with autistic speakers, with the neurotypical raters giving higher scores than the autistic raters (z = 6.843, p <.0001), see Table 52.   Table 52 Means and standard deviations (in brackets) rating scores of  all scale items per speaker diagnosis & rater diagnosis  Total Rating Score (SD) ASD rater  ASD 5.48 (1.70) NT 5.84 (1.60) NT rater  ASD 5.32 (1.67) NT 6.29 (1.49)  116  4.2. Subscale on discourse content  Speaker diagnosis had a significant effect on the ratings scores of all 7 discourse items, viz. relevance (C01), referential cohesion (C02), coherence (C03), pedantic style (C04), rehearsed (C06), fluency(C07) and perseverance (C09), (χ2(1) = 26.634, p < .0001; χ2(1) = 15.435, p < .0001; χ2(1) = 26.453, p < .0001; χ2(1) = 30.577, p < .0001; χ2(1) = 4.0192, p = 0.04;  χ2(1) = 13.136, p = 0.0003 and χ2(1) =6.8421, p = 0.0089, respectively).  There was no effect of rater diagnosis on all 7 discourse items (χ2(1) = 0.97557, p = 0.3233; χ2(1) = 0.031272, p = 0.8596; χ2(1) = 0.42699, p = 0.5135; χ2(1) = 1.5969, p = 0.2063; χ2(1) = 0.15714, p = 0.6918; χ2(1) = 0.00589; p = 0.9388 and χ2(1) = 1.6868 , p = 0.194).   Table 53 Cumulative link model with additive effects of speaker diagnosis and rater diagnosis  (ASD diagnosis is the reference level for all effects, standard errors are between brackets) Rating Score Speaker Diagnosis Rater Diagnosis Relevance 0.8770 (0.1717)*** -0.1632 (0.1686) Referential cohesion 0.6617 (0.1695)*** -0.0133 (0.1675) Coherence 0.8712 (0.1714)*** -0.0978 (0.1681) Pedantic style 0.9396 (0.1720)*** -0.2103 (0.1678) Rehearsed 0.3365 (0.1693)* 0.0532 (0.1688) Fluency 0.6083 (0.1687)*** 0.0009 (0.1668) Perseverance 0.4443 (0.1690)** 0.2250 (0.1680) Number of observations 447  Signif. codes :  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1  As can be seen from Table 54, audio recordings with autistic speakers received lower ratings on all scale items those with neurotypical speakers.  Table 54 Means and standard deviations (in brackets) of the rating scores for the seven items of the discourse subscale per speaker diagnosis  Relevance Referential Cohesion Coherence Pedantic style Rehearsed Fluency Perseverance ASD Speakers 4.86  (1.62) 4.65 (1.69) 4.56 (1.60) 4.55 (1.67) 5.30 (1.48) 4.30 (1.57) 5.83 (1.52) NT Speakers 5.64  (1.31) 5.26 (1.53) 5.33 (1.31) 5.40 (1.35) 5.53  (1.52) 4.83 (1.57) 6.20 (1.36)  117   Diverging stacked bar charts were used to represent the ratings given by autistic and neurotypical participants (i.e., the raters) on the audios of autistic and neurotypical speakers (i.e., the “rated”). In diverging stacked bar charts, the responses are positioned horizontally so that positive responses are stacked on the right and negative responses are stacked left from this baseline. Figure 20 represents the ratings for the seven discourse items: relevance (plot A), referential cohesion (plot B), coherence (plot C), rehearsed (plot D), fluency (E), perseverance. (F) and pedantic style (G).  Figure 20 Diverging stacked bar charts representing the ratings of the speakers for the seven items of the discourse subscale    118       4.3. Subscale on speaker impressions  For all four items of the speaker scale, viz. ease of speaker (audio) understanding rater (participant, S10), ease of rater (participant) understanding speaker (audio, S11), likelihood to become friends (with speaker in the audio, S12) and typicality of speaker in the audio (S13), there was an effect of Speaker Diagnosis, χ2(1)= 14.067, p = 0.0002; χ2(1)= 12.772, p = 0.0004; χ2(1)= 20.223, p < .0001 and  χ2(1) = 91.367, p < .0001, respectively. There was also a significant effect of rater diagnosis, χ2(1) = 30.694, p < .0001; χ2(1) = 31.213, p < .0001 and χ2(1) = 3.933,  119  p = 0.047 for items ease of making oneself (rater) understood by the speaker, ease of speaker understanding rater and likelihood to become friends, respectively. For the item speaker typicality, there was no effect of rater diagnosis, χ2(1) = 0.39562, p = 0.53. See Table 55.   Table 55 Cumulative link model with additive effects of speaker diagnosis and rater diagnosis and the interaction effect between speaker diagnosis and rater diagnosis (ASD diagnosis is the reference level for all effects, standard errors are between brackets) Total Rating Score Speaker Diagnosis Rater Diagnosis Making oneself understood by Speaker 0.6318 (0.1694)*** 0.9425 (0.1722)*** Speaker understanding Rater 0.5992 (0.1685)*** 0.9482 (0.1718)*** Becoming friends 0.7596 (0.1702)*** 0.3313 (0.1673)* Speaker Typicality 1.6770 (0.1819)*** -0.1187 (0.1663) Number of observations 447  Signif. codes :  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1  Table 56 represent the mean ratings of the speaker subscale, audio recordings with autistic speakers received lower ratings those with neurotypical speakers on all four items of the subscale.  Table 56 Means and standard deviations (in brackets) of the ratings on the four items of the speaker subscale per speaker diagnosis  Making oneself understood by Speaker Speaker understanding Rater Becoming friends Speaker Typicality ASD (SD) 5.45 (1.60) 5.29 (1.62) 4.36 (1.52) 4.21 (1.52) NT (SD) 5.97 (1.67) 5.82 (1.67) 5.00 (1.57) 5.69 (1.59)  Table 57 represents the mean ratings according the diagnostic group of the rater. Neurotypical raters gave higher rating scores than autistic raters for the subscale items, ease of making oneself (rater) understood by the rated speaker, ease of the rated speaker understanding rater and likelihood to become friends. There was no significant difference in the ratings given by autistic and neurotypical raters on the item speaker typicality.    120  Table 57 Means and standard deviations (in brackets) of the rating scores on the four items of the speaker subscale per rater diagnosis  Making oneself understood by Speaker Speaker understanding Rater Becoming friends Speaker Typicality ASD (SD) 4.33 (1.60) 4.16 (1.52) 3.56 (1.61) 4.02 (1.74) NT (SD) 5.12 (1.59) 4.97 (1.68) 3.83 (1.52) 3.90 (1.68)  There was also a significant interaction between speaker diagnosis and rater diagnosis, χ2(1) = 17.577, p < .0001; χ2(1) = 17.412 p < .0001 and χ2(1) = 22.817, p < .0001. Post-hoc analyses revealed that for the items 1) ease of making oneself (rater) understood by the speaker in the audio, 2) ease of speaker in the audio to understand the rater and 3) likelihood to become friends, there was no significant differences in the ratings given by the autistic raters (z = 0.082, p = 0.9346; z = 0.041, p = 0.9669 and z =  0.098, p = 0.9222, respectively). In contrast, on these three items, neurotypical raters provided higher ratings for the audio recordings with neurotypical speakers than with autistic speakers (z = 5.732, p <.0001; z = 5.541, p <.0001 and z = 6.523, p <.0001, respectively)  Table 58 Means and standard deviations (in brackets) of the rating scores for the items ease of making oneself be understood, ease of speaker understanding rater and likelihood of becoming friends per speaker diagnosis and rater diagnosis  Making oneself understood by Speaker Speaker understanding Rater Becoming friends ASD    ASD 5.32 (1.63) 5.17 (1.54) 4.58 (1.68) NT 5.29 (1.62) 5.12 (1.56) 4.50 (1.58) NT    ASD 5.58 (1.57) 5.41 (1.69) 4.14 (1.31) NT 6.67 (1.41) 6.53 (1.46) 5.51 (1.41)  Figure 21 represents the ratings for the four discourse items: ease to make oneself (rater) understood by the speaker in the audio (plot A), ease for the speaker to understand the rater (plot B), likelihood to become friends (plot C) and typicality of the speaker in the audios (plot D).    121  Figure 21 Diverging stacked bar charts representing the ratings on the four items of the speaker subscale per rater diagnosis and speaker diagnosis    4.4. Summary of results  Taken together, the results suggest that just like neurotypical participants, autistic participants rated the speech of autistic individuals less positively than that of neurotypical individuals. This was the case for all items of the discourse subscale, suggesting that they recognized and interpreted these discourse characteristics like neurotypical participants. This was also the case for all items of the speaker subscale, suggesting that not only the discourse quality of  122  autistic speakers was less positive than that of neurotypical speakers, but the impression of these speakers themselves was less positive. Furthermore, for three items of the speaker subscale, 1) making oneself understood by speaker, 2) speaker understanding rater and 3) likelihood to become friends), there was an interaction effect between speaker diagnosis (rated) and participant (rater) diagnosis. Neurotypical participants judged it more difficult to communicate (1 and 2) and less likely to become friends (3) than autistic speakers, whereas autistic participants did not judge these features differently in neurotypical and autistic speakers. The following section discusses these results in more details.  V. Discussion  Recent studies have suggested that, on the one hand, the discourse content of autistic individuals is less coherent than that of their neurotypical peers, and on the other hand, that autistic individuals are perceived more negatively than their neurotypical peer by other neurotypical individuals across different modalities (Bone et al., 2015; Grossman, 2015; Sasson et al., 2017). To date, no study had tried to investigate the relationship between discourse features and subjective impressions of the speaker nor to consider the impressions of autistic individuals themselves. The present results suggest that based on audio recordings, naïve listeners with and without a diagnosis of autism rated autistic speakers overall less positively than neurotypical speakers.  When looking at individual items, autistic speakers received significantly lower ratings on all items of the discourse subscale than the neurotypical speakers. Furthermore, there was no effect of rater diagnosis for these items. In other words, when asked to rate the quality of spoken discourse, autistic adults seem to rely on the same cues as neurotypical adults to evaluate discourse characteristics. It should also be noted that the quality of the discourse of autistic speakers was not perceived as extremely bad, as ratings did not go below 4.30 (fluency) and went as high as 5.83 (perseverance), but they were still lower than neurotypical ratings, which were not lower than 4.83 (fluency) and as high as 6.20 (perseverance). This suggests that differences in spoken discourse between autistic and neurotypical speakers are subtle but nevertheless sufficient to be perceptible, irrespective of diagnostic group. In other words, the autistic speakers represented in the audios had acquired the necessary tools to communicate but use them more awkwardly and/or inconsistently, as has already been suggested by Larkin, Hobson, Hobson, & Tolmie (2017). This conclusion allows us to refine assumptions made in Chapter 3. Specifically, I had proposed that autistic speakers might use prosody as an  123  alternative to explicit discourse-structuring devices such as connectives to signal coherence relations between utterances, albeit doing so in an atypical way, and consequently this cue is not registered by the listener. Further supporting this assumption, is the finding from the speaker subscale that autistic speakers were perceived as less typical than neurotypical speakers, both by autistic and neurotypical raters. At a first glance, these results suggest that autistic and neurotypical adults attend to the same discourse features and do not have different expectations with respect to what constitutes a coherent and typical discourse. In contrast, for the subscale targeting the impressions of the speakers in the audios, results suggest that the perceived discourse features might have different impacts for autistic and neurotypical adults. Indeed, while perceived quality and (a)typicality of discourse did not influence the judgments of autistic raters on subsequent ease of communication (making oneself easily understandable and being easily understood), as well as future attitudes (likelihood to become friends), this was not the case for neurotypical raters. Neurotypical raters judged it easier to mutually understand each other with neurotypical speakers and more likely to become friends with them than with autistic speakers. In other words, there was a one-sided (neurotypical) bias against autistic speakers. Again, the impressions of autistic speakers were not specifically poor, ranging from 4.14 to 5.58, but they were different enough to influence future behaviors such as likelihood to become friends.   VI. Conclusion  The results of this study provide a glimpse into the real-life implications of differences in spoken discourse. For individuals who have autism symptoms but cognitive and verbal abilities in the average or above average range, autism can be a hidden disability (e.g., Portway & Johnson, 2005): although they do not have obvious communication difficulties, subtle linguistic differences can nevertheless hinder their social integration. Reduced disposition of neurotypical individuals to initiate a friendship can have high implications on the well-being of autistic individuals. For example, in a study on mental health problems in adults with autism, autism acceptance from external sources such as society, family and friends predicted depression and stress, personal acceptance also predicted depression (Cage, di Monaco & Newell, 2018). Differences in communication abilities also have high implications at the professional level. For example, compared to other disability groups (mental retardation, learning disabilities and speech impairments), autistic individuals are less likely to find employment (Shattuck et al., 2012). Communication difficulties are one of the barriers they  124  face when searching for a job, even when they have adequate competence for the job itself (e.g. Hendricks, 2010; Hurlbutt & Chalmers, 2004). Considering the potential personal and professional challenges that can arise from differences in communication abilities, these differences cannot be simply considered as variations in speech style, they are actual difficulties. Therefore, it is important to both 1) raise awareness that differences in spoken discourse can lead to neurotypical biases and 2) provide support to autistic individuals to help them improve their communication efficiency.  The present study has several limitations which invite for additional research undertaking. While the results suggest that even for cognitively able and verbal autistic individuals, differences in spoken discourse have negative implications, future studies should examine first impressions of a wider range of profiles on the autism spectrum (e.g., autistic speakers with more severe cognitive, social and linguistic difficulties). Second, raters evaluated a small sample of dyadic conversations (N=12) and future studies should investigate a wider range of speakers to examine whether the present findings can be replicated. Finally, effects of conversation topic were not investigated. Furthermore, the findings of this dissertation are based exclusively on verbal behavior. However, Sasson et al. (2017) and Grossman (2015) have shown that more negative impressions of neurotypical adolescents and adults are also present in visual modality and static images. Future studies should also look at non-verbal behavior like gaze or facial expressions.     125  Chapter 5: Serendipitous findings  Gender differences in discourse characteristics and subjective impressions  I. Introduction  The main objective of this dissertation was to identify autism-specific features in spoken discourse and relate them to the perception of naïve listeners to better understand why autistic individuals are so often perceived more negatively than their neurotypical peers. Accordingly, to address this aim, my hypotheses and research designs were built to test the effect of one factor: the presence of an autism diagnosis. However, during my PhD, more and more studies started suggesting that autism presents differently in boys and girls (e.g., Duvekot et al., 2017; Hiller et al., 2016; Little, Wallisch, Salley, & Jamison, 2017). This different presentation is thought to partially explain why autistic girls are diagnosed less often and later than autistic boys (Begeer et al., 2013). Recently, the late diagnosis of autistic women has also been linked to camouflaging: hiding or masking autism characteristic from others (Bargiela, Steward, & Mandy, 2016; Cage & Troxell-Whitman, 2019; Hull et al., 2017). An important aspect of camouflaging that distinguishes it from typical efforts to fit, is its ‘cost’. Research suggests that the experiences of camouflaging could be costly in terms of mental health (e.g., stress and anxiety) and physical exhaustion (Hull et al., 2017). Autistic men also camouflage but to a lesser extent than autistic women (Lai et al., 2017). Camouflaging strategies include copying the behaviors of popular peers or studying psychology books. Recently, studies have suggested that autistic women also disguise their socio-communicative difficulties in linguistic features such as internal state language (e.g., think, know) or hesitation markers (uhs and uhms). For example, Kauschke, van der Beek, & Kamp-Becker (2016) found that during narrative production, girls (autistic and neurotypical) included more terms of internal states related to physiological sensations and modality than autistic boys. Moreover, girls provided causal explanations for the internal states of the story characters more often than boys. However, while autistic girls resemble neurotypical girls for some internal state terms, both autistic girls and boys produced less emotion terms than neurotypical girls. Similarly, Boorse et al. (2019) found that during a storytelling task, both autistic girls and boys produced more concrete, object-focused language (e.g., they produced more nouns) than neurotypical boys and girls. However, when looking at cognitive process words (e.g., think), autistic boys were the only  126  group producing less of these terms. Parish-Morris et al. (2017) showed that girls suppressed their uhs, resulting in a higher ratio of uhms, helping them ‘normalize’ the way they speak to sound more like typical girls and less like autistic boys. Taken together, these three recent studies show that autistic girls neither fully resemble autistic boys nor neurotypical girls. This ‘blended-phenotype’ can make it more difficult to recognize autism in girls (Boorse et al., 2019), delaying diagnosis. However, timely identification and access to treatment is crucial for optimizing outcomes in a neurodevelopmental disorder like autism (MacDonald, Parry-Cruwys, Dupere, & Ahearn, 2014; Reichow, 2012). Differentiating sex-neutral from sex-specific linguistic markers of ASD is crucial to pin-point more accurately core differences representing the essence of autism (Boorse et al., 2019). Although my studies were not designed to detect any effects of participants’ gender, the studies in Chapter 3 and 4 included participant samples composed of equal or almost equal numbers of male and female participants, allowing for exploratory analyses of potential gender effects in discourse features and/or subjective impressions.  II. Results  2.1. Participants  Participants’ characteristics are fully described in Section 2.1 of Chapter 3 and Section 3.1 of Chapter 4.  2.2. Syntactic coding and BDUs  The significance of gender effects on type of syntactic units and BDUs, were examined using the glm function (Poisson family) in R (R Core Team, 2016), which included the dependent variables (e.g., total BDUs, total dependency clause, discourse-structuring devices) and participant’s gender as fixed effects. These models were compared to a model without the effect of gender. There was no effect of participants’ gender on any of the BDUs. There was only one significant effect of gender on the syntactic coding category of discourse structuring devices, χ2(1) = 14.035, p = 0.0002.       127  Table 59 Regression coefficients of the generalized link model with additive  effects of speaker sex, total syntactic units and discourse-structuring devices.  ASD and female are the reference levels.  Discourse-structuring devices Sex Male -0.254 (0.068)*** Signif. codes :  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1  As previous analyses (Chapter 3) also showed an effect of participants’ diagnosis on the production of discourse-structuring devices, the interaction effect between participants’ diagnosis and gender was also modeled. The interaction was not significant, χ2(1) = 0.666, p = 0.41. Overall, women produced more discourse markers than men, see Table 60.  Table 60 Means and standard deviations (in brackets) of total  counts and percentage of discourse-structuring devices per sex  Discourse-structuring devices  Counts Percentage Female (SD) 88.33 (26.41) 39.92 (2.80) Male (SD) 63.50 (30.64) 32.54 (5.98)  Figure 22 contains two boxplots displaying the data distribution of discourse-structuring devices per diagnosis and sex of the participant.     128  Figure 22 Boxplots of discourse-structuring devices per participants’ diagnosis and sex   2.3. Auditory rating experiment  2.3.1. Full Rating Scale There was an effect of speaker sex, χ2(1) = 36.24, p < .0001 and rater sex, χ2(1) = 273.21, p < .0001 on the overall ratings of the scale. There was also a significant interaction between speaker sex and rater sex, χ2(1) = 6.2013, p = 0.013. See Table 61.    129  Table 61 Regression coefficients of the generalized link model with  additive effects of speaker Sex, rater sex and the interaction effect  between speaker sex and rater sex (Female is the reference level   for all effects, standard error is in brackets)  Total Rating Score Speaker male -0.45410 (0.0754)*** Rater male -0.97408 (0.0738)*** Speaker*Rater 0.25224 (0.10132) * Signif. codes :  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1  As can be seen from Table 62, audios with female speakers received higher rating than audios with male speakers. Female raters gave higher rating than male raters.  Table 62 Means and standard deviations (in brackets) of the rating scores  per speaker sex and rater sex Mean Ratings Total Scale Female Speaker Male Speaker Female Rater Male Rater 4.87 (1.69) 4.60 (1.61) 5.10 (1.69) 4.39 (1.54)  Post-hoc analyses revealed that the ratings of male and female participants were significantly different from each other, and that both sexes gave significantly higher ratings for the audios with female speakers than with male speakers. See Table 63.  Table 63 Means and standard deviations (in brackets) of total rating  scores per speaker and rater Sex  Total Score  Female Speaker Male Speaker Female Rater 5.28 (1.69) 4.93 (1.68) Male Rater 4.49 (1.60) 4.30 (1.47)  2.3.2. Subscale discourse abilities Recall from Chapter 4 that there was a significant main effect of speaker diagnosis on all items of the discourse subscale. There were no was no effect of speaker sex on the items referential cohesion, rehearsed, fluency and perseverance, χ2(1) = 2.6199, p=0.1055; χ2(1) = 0.4187, p=0.5176 and χ2(1) = 1.7626, p=0.1843, respectively, although there was a main effect of  130  speaker sex on the items relevance, coherence, and pedantic style, χ2(1)= 7.7487, p=0.005375; χ2(1)= 6.7707, p=0.009 and χ2(1)= 9.29, p= 0.002, respectively. See Table 64.  Table 64 Cumulative link model with additive effects of speaker sex and the interaction  effect between speaker sex and diagnosis (ASD diagnosis and female sex are the reference  levels, standard errors are between brackets) Discourse Item Speaker Sex Diagnosis*Sex Relevance -0.4944 (0.1698)** -0.6659 (0.3397)* Coherence -0.4463 (0.1691)** -0.92470 (0.3394)** Pedantic style -0.5268  (0.1691)** -1.06106 (0.3395)** Number of observations 447  Signif. codes :  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1  As can be seen from Table 65 below, audios with female speakers received higher ratings than audios with male speakers.  Table 65 Means and standard deviations (in brackets) for the rating scores  of the items relevance, coherence and pedantic speech per speaker sex  Relevance Coherence Pedantic Female 5.45 (1.47) 5.11 (1.55) 5.19 (1.58) Male 5.05 (1.56) 4.78 (1.46) 4.77 (1.54)  There was also a significant interaction between speaker diagnosis and speaker sex, χ2(1) = 3.8542, p = 0.049; χ2(1) = 7.4658, p = 0.006 and χ2(1)= 9.8424, p = 0.002 for the items relevance, coherence and pedantic style respectively. Post-hoc analyses revealed for the items relevance, coherence and pedantic, neurotypical female speakers received higher ratings than neurotypical male speakers (z = 3.443, p = 0.003; z = 3.780, p =0.0009 and z = 4.393, p = 0.0001, respectively), autistic female speakers (z = 5.042, p < .0001; z= 5.490, p < .0001 and z = 6.049, p < .0001, respectively) and autistic male speakers (z = 5.576, p < .0001; z = 5.462, p < .0001 and z = 6.021, p < .0001, respectively). The other speaker groups did not differ significantly from one another. See Table 66.     131  Table 66 Means and standard deviations (in brackets) for the rating scores of the  items relevance, coherence and pedantic speech per speaker diagnosis and sex  Relevance Coherence Pedantic  NT Female 5.96 (1.18) 5.65 (1.26) 5.81 (1.22) NT Male 5.32 (1.36) 5.00 (1.29) 4.99 (1.35) ASD Female 4.94 (1.55) 4.55 (1.63) 4.55 (1.66) ASD Male 4.79 (1.69) 4.57 (1.58) 4.55 (1.68)  Figure 23 represents the ratings of the discourse subscale items relevance, coherence and pedantic style. As can be seen from the charts representing neurotypical women, the majority of the ratings they received were 6 or 7, distinguishing themselves from the other speakers.  Figure 23 Diverging stacked bar charts of the ratings on the items relevance (plot A), coherence (plot B) and pedantic style (plot C) per speaker diagnosis and sex   132    2.3.3. Speaker subscale For the items, making oneself (rater) understood by speaker (in audios), speaker understanding rater, becoming friends and typicality, there was no effect of speaker sex, χ2(1) =2.8821, p = 0.089; χ2(1) = 3.2332, p = 0.072; χ2(1) =1.3441, p = 0.2463 and χ2 (1) = 2.6709, p = 0.10, respectively. See Table 67.   Table 67 Cumulative link model with additive effect of speaker sex.  Female sex is the reference level, standard errors in brackets.  Discourse item Speaker Sex Making oneself understood by Speaker -0.2900 (0.1676) Speaker understanding Rater -0.3099 (0.1673) Becoming friends -0.2085 (0.1673) Speaker Typicality -0.2873 (0.1668) Number of observations 447 Signif. codes :  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1  Women and men received similar ratings on the four items of the speaker subscale: 1) ease of making oneself (rater) understood by the speaker in the audio, ease of speaker understanding rater, likelihood to become friends and speaker typicality.     133  Table 68 Means and standard deviations (in brackets) of the rating scores on the items ease of making oneself be understood, ease of speaker understanding rater and likelihood of becoming friends per speaker sex  Making oneself understood by speaker Speaker understanding rater Becoming friends Speaker typicality Female 5.83 (1.70) 5.68 (1.72) 4.79 (1.69) 5.09 (1.78) Male  5.60 (1.60) 5.42 (1.60) 4.58 (1.40) 4.81 (1.65)  III. Discussion  These additional analyses suggest that sex-specific differences emerged both in transcript analyses and subjective ratings of spoken discourse quality. Transcript analyses suggest that autistic and neurotypical women produced more discourse-structuring devices than autistic and neurotypical men. This latter finding brings new evidence in favor of linguistic camouflage. When looking at the proportion of discourse-structuring devices produced by the two genders and two diagnostic groups, neurotypical and autistic women were virtually indistinguishable from each other (59% and 58 %, respectively). Furthermore, the proportion of discourse-structuring devices they produced was higher than that of neurotypical and autistic men (41% and 42%, respectively). In other words, autistic women may normalize the way they sound in the production of discourse-structuring devices. The present distribution pattern of discourse-structuring devices reflects previous findings that women use more discourse-structuring devices than men (e.g. Escalera, 2009; Matei, 2011) Interestingly, while autistic women may resemble neurotypical women in their use of discourse-structuring devices in transcript analyses, subjective ratings did distinguish between neurotypical and autistic women. Specifically, the ratings of the audios with neurotypical women superseded audios with neurotypical men, autistic women and autistic men on three discourse items, viz. relevance, coherence relations and pedantic style. In other words, while autistic women might resemble neurotypical women in the quantity of discourse-structuring devices, the integration of these features in the on-going discourse might remain atypical. Initial evidence in favor of this assumption is the observation that there were no gender differences in regulatory BDUs. Recall that this latter type of BDU reflects how discourse-structuring devices such as connectives, discourse markers and adjuncts are integrated prosodically into the discourse. In other words, linguistic camouflage of autistic women might be limited to surface-level features such as discourse-structuring devices, with more difficulties to camouflage lower-level linguistic features such as prosodic and acoustic characteristics. The  134  data from this corpus can be used to test this hypothesis in future studies by analyzing a bigger sample of regulatory BDUs. Furthermore, analyzing BDUs using the original procedure described by Degand & Simon (2009a, 2009b), viz. a prosodic segmentation incorporating finer prosodic characteristics would also provide more insight into the camouflaging strategies of autistic women. In addition to evidence in favor of linguistic camouflage in autistic women, the effect of participants’ gender on the subjective ratings of certain discourse features suggest that there is a gender bias in addition to a diagnostic bias. However, only three out of the seven scale items on discourse features yielded a gender differences, suggesting that some discourse features are more ‘vulnerable’ to gender biases than others. Relevance, relational coherence and pedantic style seemed to be more strongly associated with men (autistic and neurotypical) as well as autistic women. This effect could be due to the rating task, viz. the evaluation of communication skills, which could have activated gender-related expectations of good communication skills in women. Neurotypical women corresponded more precisely to these expectations, distinguishing them more starkly from the rest of the speaker groups. If the speaking style of autistic women is perceived as more similar to men’s speaking style (autistic and neurotypical) and men are perceived more negatively, autistic women will still be perceived more negatively than neurotypical women even if they show better skills than autistic men.  Furthermore, there was not only a general gender bias in the ‘targets’ of the ratings (speakers in the rated audios) but also in the ‘source’ of these ratings (participants rating the audios). Specifically, both men and women rated audios with women speakers higher than audios with men, with men raters giving lower ratings than female raters and male speakers received lower ratings than female speakers. Further studies should try to examine more specifically the role of gender stereotypes and/or autism stigmas on perceptions of discourse features, as well as the interaction between the characteristics of the individual being rated and the rater. In a very recent study, Morrison, DeBrabander, Faso, & Sasson (2019), initiated such a research direction by showing that variability in rater stigma toward autism and autism knowledge influenced first impression ratings of autistic individuals, sometimes more than the diagnosis of autism. Considering that autistic women fit neither the standards of typical women nor typical autism, they might be evaluated with more variability, depending on the reference point used to make the evaluation, viz. gender or autism (Biernat & Manis, 1994; Biernat, Manis, & Nelson, 1991).     135  Chapter 6: General Discussion  In this dissertation, thorough discourse analyses and subjective ratings exposed a range of linguistic measures differentiating the speech of neurotypical and that of autistic adults. Taken together, its results indicate that incoherence and atypicality are subtle but robust qualities of the spoken discourse of autistic adults whose verbal and non-verbal IQs are, nevertheless, in the typical range. Crucially, this dissertation shows that perceived differences in discourse (in)coherence and (a)typicality by others have significant implications, as they led to a ‘neurotypical’ bias towards autistic speakers.   From a methodological point of view, this dissertation provides one of the most comprehensive and systematic analyses of spoken discourse in autistic adults. More than six hours of speech were transcribed, segmented and annotated, yielding a set of more than 30 discourse features that could be analysed. For the narrative study, this tight set of linguistic features was designed to reflect all three narrative dimensions of microstructure, macrostructure and internal state language at once as this has not been done by previous studies. The meta-analysis by Baixauli et al. (2016) had provided initial evidence that autistic individuals (essentially children and young adolescents) differed from their typical peers in all three narrative dimensions. However, their results reflected a tendency across participants from different samples. It remained to be examined whether the participants of one sample would show difficulties in all three dimensions as well. The comprehensive analysis of narrative discourse confirmed the results of Baixauli et al. (2016), viz. autistic adults differed from neurotypical adults at the level of the microstructure, macrostructure and internal state language. Furthermore, the complex segmentation procedure developed by Degand & Simon (2008, 2009a) allowed us to represent both syntactic and prosodic surface-level features of spoken discourse, viz. syntactic units and silent pauses as well as how these features work together to convey information. However, to every methodological advantage there is a drawback. Specifically, such thorough and fine-grained discourse analysis was very time-consuming. At every stage of the process (transcription, segmentation and annotation), students had to be hired and trained. The full corpus could not be exploited due to time-constraints.  From a theoretical point of view, this dissertation exposes consistent difficulties at the level of discourse coherence, especially in the use of explicit discourse-structuring devices such as connectives (e.g., but, and, because, then) and discourse markers (e.g., well, you know,  136  like). Both in study 1 (Chapter 2) and 2 (Chapter 3), autistic speakers differed from neurotypical speakers by using less discourse-structuring devices. They further distinguished themselves from their neurotypical peers in the way these markers were integrated prosodically within the on-going discourse. Specifically, in comparison to autistic adults, neurotypical adults used silent pauses more frequently to isolate connectives, discourse markers and adjuncts (regulatory BDUs). Within the context of a semi-structured interview, as was the case for the present participants, such regulatory BDUs were very useful cues to the experimenter. Specifically, neurotypical participants used regulatory BDUs such as <donc> <voilà> (<so> <there>), <et> <voilà> (<and> <there>) and <voilà> (<there>) to indicate to the interviewer that they had nothing more to add to the question under discussion. The experimenter interpreted these cues by taking up the floor again by either pursuing with questions (1) or contributing herself to what the participant had just said (2). Regulatory BDUs are highlighted in bold and silences are indicated into brackets in seconds.   (1) PP: [je pense que j’ai ///1 trois quatre bon amis]1 (2.05) <donc> <voilà> [I think that I have ///1 three four good friends]1 (2.05) <so> <there>  Exp : [ok]1 [c’est des gens qui sont ici aussi]1 ///1 [ok]1 [it’s people that are also here]1 ///1 Comparison participant (male, 20 years old) (2) PP: [ils ont une manière plus égoïste d’entreprendre la chose]1 ///1 [they have a more selfish way of dealing with things]1 ///1 (0.27) Exp: <mhm> <mhm> (1.66) PP:  <voilà> < there>  (0.98)  137  Exp: <donc> <ouais> [des gens sur qui tu peux ]1 <euh> ///1 <so> <yes> [people who you can]1 <uh> ///1 Comparison participant (male, 32 years old) In example (1), the experimenter asked the participant whether he had any friends, to which he replies that he has three-four good friends. He has nothing to add so after a silent pause of two seconds, he uses a regulatory BDU to indicate to the experimenter she can go on with the questions. She attends to this cue and asks another question about the participant’s friends, viz. whether they come from the same area as the participant. In example (2), the experimenter asked the participant how he would distinguish a friend from an acquaintance. After responding to this question, the experimenter signals through backchannelling that she is paying attention to what the participant is saying. However, the participant does not seem to have anything more to add and signals it with <voilà> (there). The experimenter takes up the floor again and checks whether she has understood the participant’s answer. These two examples highlight how regulatory BDUs can be helpful to the experimenter by restricting possible ambiguities about topic continuation or shift. The possibility that regulatory BDUs play a role in turn-taking lines up with findings that utterance-final discourse markers can serve as turn-taking devices in speech and instant messaging (Degand & van Bergen, 2016). Autistic adults produced less regulatory BDUs which suggests that they more often left it up to the experimenter to resolve ambiguity about potential topic continuation or shift. Without being provided an explicit cue, the experimenter will have to rely on other cues such as long silences which might be more difficult to interpret and result in interruptions. Consider the following example. Overlapping speech is indicated in between slashes in bold. The speech of the participant is represented in blue, the speech of the experimenter in green.  (3) et y a des fois quand la personne est vraiment (0.25) alors je lui dis (0.94) je (2.03) alors on /trouve/ /c’est un/ échange /aussi/ /euh ouais c’est / c’est un compromis    138  and there are times when the person is really (0.25) so then I tell her (0.94) I (2.03) so we /find/ /it’s an/ exchange /also/ /uh yeah it’s/ it’s a compromise  Autistic participant (female, 32 years old)  In the example above, the participant seems to have difficulties communicating her idea, interrupting herself and starting a new utterance after each silent pause. The sequence of three interrupted syntactic units with a silent pause of 2 seconds would suggest that the participant will not finish what she is saying. However, the participant does start a new utterance which is interrupted by the experimenter, who is again interrupted by the participant. Examples (1) and (2) illustrate how successful communication can unfold with the help of explicit cues. In contrast, example (3) illustrates how communication difficulties can arise when cues are not explicated. While neurotypical adults highlighted the presence of connectives, discourse markers and adjuncts by isolating it with a silent pause, autistic adults tend to integrate them in full-blown syntactic units (example 4 and 5) or not use explicit discourse-structuring devices/adjunct at all, using a silent pause instead to group several syntactic units together (viz. silence-bound BDUs, example 6).  (4) [le mariage de ma mère et mon père a duré que trois ans]1 <hein> <puis> <après> [c'était bardaf]2 ///1  [the marriage of my mom and dad lasted only three years]1 <eh> <then> <after> [it crashed]2 ///1  Autistic participant (male, 22 years old)  (5) [c'est c'est pour ça que je dis ils arrivent à me les faire comprendre ]1 <et et> [j'adore]2 <et et et> [moi]3 <en fait> [j'ai j'ai j'admire ces gens]4 <parce que> <justement> [ils ont ce décodeur canal plus que je n'ai pas]5 ///1     139  [that’s why I say that they succeed in making me understand them]1 <et et> [I love it]2 <et et> [me]3 <actually> [I have I admire these people]4 <because> <exactly this > [they have the signal decoder that I don’t have]5 ///1  Autistic participant (female, 43 years old)  (6) [on peut pas mettre tout le monde d'accord]1 [y'aura toujours des problèmes]2 ///1  [we cannot get everyone to agree]1 [there will always be problems]2 ///1   (6bis) [on peut pas mettre tout le monde d’accord]1 <parce que> [y’aura toujours des problèmes]2 ///1     [we cannot get everyone to agree]1 <because> [there will always be problems]2 ///1  Autistic participant (male, 21 years old)  (7) [je vais être très nerveuse]1 [ça va se répercuter sur le gamin]2 ///1  [I will be very nervous]1 [it will have an impact on the kid]2 ///1  (7bis) [je vais être très nerveuse]1 <du coup> [ça va se répercuter sur le gamin]2 ///1  [I will be very nervous]1 <so > [it will have an impact on the kid]2 ///1  Autistic participant (female, 32 years old)  Example (4) and (5) illustrate how the discourse markers (e.g., hein), adjuncts (e.g., justement) and connectives (e.g., et, parce que) are integrated in the preceding or following syntactic unit rather than being isolated by silent pauses as was the case in examples (1) and (2). Examples (6) and (7) illustrate how the relations between the utterances are left implicit. The causal relationship between the two syntactic units can be explicated with the causal connectives ‘parce que’ and ‘du coup’ as shown in the examples (6bis) and (7bis). Taken together, the coded data suggest that the discourse output produced by autistic participants both in the narrative task and semi-structured tasks include less explicit linguistic cues to create coherence, both at the local level (e.g., discourse-structuring devices) and at the global level  140  (less regulatory BDUs). This will make it more difficult for the listener to derive a coherent mental representation of the discourse being produced. And indeed, Chapter 4 provides initial evidence for this assumption. The ratings of naïve listeners – both with and without a diagnosis of autism – of discursive features systematically distinguished the speech of autistic adults from that of neurotypical adults. The latter group of speakers received higher ratings, indicating better discourse abilities. Furthermore, neurotypicals judged it more difficult to communicate effectively with autistic speakers than with neurotypical speakers. The implication of such a ‘neurotypical-specific’ bias on the social interactions involving autistic individuals are far from trivial. A successful social interaction requires a two-sided dynamic, whereby both interlocutors have to learn about the other. If neurotypical individuals are biased and reluctant to interact with autistic individuals, they make it difficult for autistic individuals to form accurate impressions of their interlocutor. Furthermore, initial impressions can influence a wide range of subsequent choices, behaviors and attitudes such as initiating a conversation or a friendship. In this respect, autistic individuals face a ‘double-whammy’. Not only are they limited by their own individual specificities, but also by biases of their neurotypical interlocutor. Evidence for this conclusion was found in the ratings of the neurotypical participants who showed more reluctance to become friends with autistic speakers than neurotypical speakers. Therefore, it is important to raise awareness to differences in spoken discourse so that they can be recognized by neurotypical individuals and addressed by autistic individuals. On the one hand, recognition of linguistic differences in communication can help mitigate the adverse effects of neurotypical biases (e.g., reluctance to initiate friendships or hire an autistic person). On the other hand, recognition of these differences can help autistic individuals remediate specific aspects of their discourse (e.g., discourse-structuring devices) to improve communication efficiency as well as the first impressions they make on their interlocutors.   Future research directions The annotated corpus of spoken discourse created for this dissertation provides data for future studies to complement and extend the current findings. One important objective for future research would be to analyze larger quantities of data as well as larger sample sizes. For instance, due to time-constraint, the findings on BDUs are based on a small sample size (N=12). However, future studies could apply this procedure to the remaining participants to confirm the findings of Chapter 3. A second objective for future research concerns the context in  141  which the data was collected. In this dissertation, conversational data was obtained during a clinical interview, the ADOS-2. Consequently, the interlocutor of the participants was an accredited ADOS administrator, trained both to follow the guidelines of the semi-structured interview and to ensure the interview would unfold smoothly17. Therefore, the full potential of the interlocutor’s role was limited and was not considered in the analytical approach to the data. Accordingly, the discourse output was essentially analyzed as a finished and self-contained product of the speaker’s mental processes.  Hence, future studies should design more naturalistic situations to examine speech patterns outside of a clinical setting with ‘ordinary’ interaction partners. In turn, such experimental design would make it possible to apply interactionally-driven approaches to the spoken output. For example, some authors have already taken a step in this direction, analyzing narrative production in ASD in more naturalistic contexts, viz. with family members (e.g., Solomon, 2004) or peers (e.g., Bottema-Beutel & White, 2016) Likewise, some authors have highlighted that a more accurate understanding of pragmatic difficulties in autism should recognize discourse production as the result of an interactional process. Specifically, discourse analytic techniques such as conversation analyses (e.g., Sterponi & de Kirby, 2016; Sterponi, de Kirby, & Shankey, 2014) or phasal analysis (e.g., de Villiers, 2011) have demonstrated that instances of language use that would traditionally be interpreted as incoherent or demonstrating pragmatic deficit can be reinterpreted as instances of interactional participation. While this may be the case, findings of this dissertation, as well as previous studies (e.g., Grossman, 2015; Sasson et al., 2017), robustly converge to the conclusion that subtle differences in verbal and non-verbal behaviors of autistic and neurotypical individuals are (rapidly) perceived by naïve listeners, and that, perhaps more importantly, these differences lead to more negative impressions of autistic individuals. Hence, on the one hand, the former research avenue on rethinking language in autism reminds us that detailed discourse analyses can help us uncover hidden conversational strengths of autistic individuals. On the other hand, the latter research avenue on first impressions reminds us of the ‘reality’ viz. of the real-life implications of subtle differences in communicative abilities. In future steps, studies should pursue research on the link between specific linguistic features and their impressions of others to determine which linguistic features improve most (or conversely deteriorate) first impressions. The results of such studies would provide a basis to design concrete intervention targets.   17 I would like to thank in particular Prof. Inge-Marie Eigsti for highlighting this point.   142  Another objective for future research is to use the corpus data to remodel the segmentation procedure more accurately by incorporating acoustic characteristics of the prosodic segmentation. As already discussed in Chapter 3, the frequency of certain types of BDUs, especially the left-over category of mixed BDUs, might change when more subtle acoustic cues are considered. Another factor that might influence the distribution of BDUs is the discourse genre. A first step could be to also segment the discourse output of the narrative task in BDUs. This would allow for examination of distribution patterns both across clinical groups and across genres, providing further evidence for BDUs as discourse strategies. Finally, the serendipitous findings on gender also provide important future research directions. Specifically, gender effects indicated that both autistic and neurotypical women produced more explicit discourse-structuring devices than autistic and neurotypical men. This could suggest better ability to infer when a coherence relation should be marked linguistically. By ‘normalizing’ their use of discourse-structuring devices, autistic women could camouflage their communication difficulties. However, this strategy was not successful enough for them to be perceived as typical by naïve listeners. Indeed, both autistic and neurotypical raters judged the discourse quality of autistic women and men similarly. Data from more women in the present corpus could be analyzed to validate these findings and explore other potential features employed as camouflage strategies by autistic women.  143  References   American Psychiatric Association. (2013). Diagnostic and statistical manual of mental disorders : DSM-5. American Psychiatric Association. DSM. https://doi.org/10.1176/appi.books.9780890425596.744053 Baixauli, I., Colomer, C., Rosello, B., & Miranda, A. (2016). Narratives of children with high-functioning autism spectrum disorder: A meta-analysis. Res Dev Disabil, 59, 234–254. https://doi.org/10.1016/j.ridd.2016.09.007 Baltaxe, C. A. M., & Simmons, J. Q. (1985). Prosodic Development in Normal and Autistic Children. In Communication Problems in Autism (pp. 95–125). https://doi.org/10.1007/978-1-4757-4806-2_7 Bamberg, M. G. (Ed.). (1997). Narrative development: Six approaches. Psychology Press. Bang, J., Burns, J., & Nadig, A. (2013). Brief report: Conveying subjective experience in conversation: Production of mental state terms and personal narratives in individuals with high functioning autism. Journal of Autism and Developmental Disorders, 43(7), 1732–1740. https://doi.org/10.1007/s10803-012-1716-4 Bargiela, S., Steward, R., & Mandy, W. (2016). The Experiences of Late-diagnosed Women with Autism Spectrum Conditions: An Investigation of the Female Autism Phenotype. Journal of Autism and Developmental Disorders, 46(10), 3281-3294. https://doi.org/10.1007/s10803-016-2872-8 Baron-Cohen, S., Leslie, A. M., & Frith, U. (1985). Does the autistic child have a “theory of mind” ? Cognition, 21(1), 37–46. https://doi.org/10.1016/0010-0277(85)90022-8 Baron-Cohen, S., Wheelwright, S., Skinner, R., Martin, J., & Clubley, E. (2001). The Autism-Spectrum Quotient (AQ): Evidence from Asperger Syndrome/High-Functioning Autism, Males and Females, Scientists and Mathematicians. Journal of Autism and Developmental Disorders, 31(1), 5–17. https://doi.org/10.1023/A:1005653411471 Beaumont, R., & Newcombe, P. (2006). Theory of mind and central coherence in adults with high-functioning autism or Asperger syndrome. Autism, 10(4), 365–382. https://doi.org/10.1177/1362361306064416 Begeer, S., Mandell, D., Wijnker-Holmes, B., Venderbosch, S., Rem, D., Stekelenburg, F., & Koot, H. M. (2013). Sex differences in the timing of identification among children and adults with autism spectrum disorders. Journal of Autism and Developmental Disorders, 43(5), 1151–1156. https://doi.org/10.1007/s10803-012-1656-z  144  Biernat, M., & Manis, M. (1994). Shifting standards and stereotype-based judgments. Journal of Personality and Social Psychology, 66(1), 5–20. https://doi.org/10.1037/0022-3514.66.1.5 Biernat, M., Manis, M., & Nelson, T. E. (1991). Stereotypes and standards of judgment. Journal of Personality and Social Psychology, 60(4), 485–499. https://doi.org/10.1037/0022-3514.60.4.485 Bilger, M., & Campione, E. (2002). Propositions pour un étiquetage en «séquences fonctionnelles». Recherches sur le français parlé, (17), 117-136. Boersma, P., & Weenink, D. (2017). Praat: doing phonetics by computer (Version 4.5.)[Computer program]. Retrieved from Http://Www.Praat.Org/. Bone, D., Black, M. P., Ramakrishna, A., Grossman, R., & Narayanan, S. (2015). Acoustic-prosodic correlates of “awkward” prosody in story retellings from adolescents with autism. In Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH (Vol. 2015-Janua, pp. 1616–1620). Boorse, J., Cola, M., Plate, S., Yankowitz, L., Pandey, J., Schultz, R. T., & Parish-Morris, J. (2019). Linguistic markers of autism in girls: Evidence of a “blended phenotype” during storytelling. Molecular Autism, 10(1), 1–12. https://doi.org/10.1186/s13229-019-0268-2 Bottema-Beutel, K., & White, R. (2016). By the Book: An Analysis of Adolescents with Autism Spectrum Condition Co-constructing Fictional Narratives with Peers. Journal of Autism and Developmental Disorders, 46(2), 361–377. https://doi.org/10.1007/s10803-015-2524-4 Botting, N. (2002). Narrative as a tool for the assessment of linguistic and pragmatic impairments. Child Language Teaching and Therapy, 18(1), 1–21. https://doi.org/10.1191/0265659002ct224oa Bretherton, I., & Beeghly, M. (1982). Talking about internal states: The acquisition of an explicit theory of mind. Developmental Psychology, 18(6), 906–921. https://doi.org/10.1037/0012-1649.18.6.906 Bruner, J. (1991). The Narrative Construction of Reality. Critical Inquiry, 18(1), 1–21. https://doi.org/10.1086/448619 Bruner, J. (2004). Jerome Bruner Life as Narrative. Social Research, 71(1), 691–711. https://doi.org/10.1007/s10780-008-9039-2 Cage, E., & Troxell-Whitman, Z. (2019). Understanding the Reasons, Contexts and Costs of Camouflaging for Autistic Adults. Journal of Autism and Developmental Disorders, 49(5), 1899-1911. https://doi.org/10.1007/s10803-018-03878-x  145  Canfield, A., Eigsti, I. M., De Marchena, A. B., & Fein, D. (2016). Story Goodness in Adolescents With Autism Spectrum Disorder (ASD) and in Optimal Outcomes From ASD. Journal of Speech, Language, and Hearing Research, 59, 533–545. https://doi.org/10.1044/2015 Capps, L., Losh, M., & Thurber, C. (2000). “The frog ate the bug and made his mouth sad”: Narrative competence in children with autism. Journal of Abnormal Child Psychology, 28(2), 193–204. https://doi.org/10.1023/A:1005126915631 Chafe, W. (1987). Cognitive constraints on information Flow. Coherence and Grounding in Discourse. https://doi.org/10.1075/tsl.11.03cha Christensen, R. (2015). Package ‘ ordinal.’ Cran, 1–61. Coelho, C. A., Liles, B. Z., & Duffy, R. J. (1991). The use of discourse analyses for the evaluation of higher level traumatically brain-injured adults. Brain Injury, 5(4), 381–392. https://doi.org/10.3109/02699059109008111 Colle, L., Baron-Cohen, S., Wheelwright, S., & Van Der Lely, H. K. J. (2008). Narrative discourse in adults with high-functioning autism or Asperger syndrome. Journal of Autism and Developmental Disorders, 38(1), 28–40. https://doi.org/10.1007/s10803-007-0357-5 Crible, L. (2017). Chapter 3. Towards an operational category of discourse markers. https://doi.org/10.1075/slcs.186.04cri De Marchena, A., & Eigsti, I. M. (2016). The art of common ground: Emergence of a complex pragmatic language skill in adolescents with autism spectrum disorders. Journal of Child Language, 43(1), 43-80. https://doi.org/10.1017/S0305000915000070 de Marchena, A., & Miller, J. (2017). “Frank” presentations as a novel research construct and element of diagnostic decision-making in autism spectrum disorder. Autism Research, 10(4), 653-662. https://doi.org/10.1002/aur.1706 de Villiers, J. (2011). “I saw the yellowish going south”: Narrative discourse in autism spectrum disorder. Belgian Journal of Linguistics, 25(1), 3–29. https://doi.org/10.1075/bjl.25.02vil Degand, L., Martin, L. J., & Simon, A.-C. (2014). Unités discursives de base et leur périphérie gauche dans LOCAS-F, un corpus oral multigenres annoté. SHS Web of Conferences. https://doi.org/10.1051/shsconf/20140801211 Degand, L., & Simon, A. C. (2005). Minimal Discourse Units: Can we define them, and why should we. Proceedings of SEM-05. Connectors, Discourse Framing and Discourse Structure: From Corpus-Based and Experimental Analyses to Discourse Theories, (477), 65–74.  146  Degand, L., & Simon, A. C. (2008). Minimal discourse units in spoken French : Uncovering genre-bound segmentation strategies. Linearisation and Segmentation in Discourse. Multidisciplinary Approaches to Discourse 2008 (MAD 08), Feb 20-23 2008, 2008(Mad 08), 1–11. Retrieved from http://www.hf.uio.no/ilos/forskning/konferanser/mad08/proceedings/Degand&Simon-MAD08.pdf Degand, L., & Simon, A. C. (2009a). Mapping prosody and syntax as Discourse Strategies: How basic discourse units vary across genres, 8, 79. In Where Prosody Meets Pragmatics. https://doi.org/10.1163/9789004253223_005 Degand, L., & Simon, A. C. (2009b). On identifying basic discourse units in speech: theoretical and empirical issues. Discours. Revue de linguistique, psycholinguistique et informatique. A journal of linguistics, psycholinguistics and computational linguistics, (4). Degand, L., & van Bergen, G. (2016). Discourse Markers as Turn-Transition Devices: Evidence From Speech and Instant Messaging. Discourse Processes, 6950(September), 1–25. https://doi.org/10.1080/0163853X.2016.1198136 Deliens, G., Papastamou, F., Ruytenbeek, N., Geelhand, P., & Kissine, M. (2018). Selective Pragmatic Impairment in Autism Spectrum Disorder: Indirect Requests Versus Irony. Journal of Autism and Developmental Disorders, 48(9). https://doi.org/10.1007/s10803-018-3561-6 DePape, A.M.R., Hall, G. B., Tillmann, B., & Trainor, L. J. (2012). Auditory processing in high-functioning adolescents with Autism Spectrum Disorder. PloS One, 7(9), e44084. https://doi.org/10.1371/journal.pone.0044084 DePape, A. M. R., Chen, A., Hall, G. B. C., & Trainor, L. J. (2012). Use of prosody and information structure in high functioning adults with Autism in relation to language ability. Frontiers in Psychology, 3(MAR), 1–13. https://doi.org/10.3389/fpsyg.2012.00072 Diehl, J. J., Bennetto, L., & Young, E. C. (2006). Story recall and narrative coherence of high-functioning children with autism spectrum disorders. Journal of Abnormal Child Psychology, 34(1), 87–102. https://doi.org/10.1007/s10802-005-9003-x Diehl, J. J., Watson, D., Bennetto, L., McDonough, J., & Gunlogson, C. (2009). An acoustic analysis of prosody in high-functioning autism. Applied Psycholinguistics, 30(3), 385–404. https://doi.org/10.1017/S0142716409090201    147  Dijkstra, K., Bourgeois, M. S., Allen, R. S., & Burgio, L. D. (2004). Conversational coherence: Discourse analysis of older adults with and without dementia, 17(4), 263-283. Journal of Neurolinguistics. https://doi.org/10.1016/S0911-6044(03)00048-4 Dunn, D. S., & Andrews, E. E. (2015). Person-first and identity-first language: Developing psychologists’ cultural competence using disability language, 70(3), 255. American Psychologist. https://doi.org/10.1037/a0038636 Duvekot, J., van der Ende, J., Verhulst, F. C., Slappendel, G., van Daalen, E., Maras, A., & Greaves-Lord, K. (2017). Factors influencing the probability of a diagnosis of autism spectrum disorder in girls versus boys. Autism, 21(6), 646–658. https://doi.org/10.1177/1362361316672178 Dworzynski, K., Ronald, A., Bolton, P., & Happé, F. (2012). How different are girls and boys above and below the diagnostic threshold for autism spectrum disorders? Journal of the American Academy of Child and Adolescent Psychiatry, 51(8), 788–797. https://doi.org/10.1016/j.jaac.2012.05.018 Eigsti, I.-M., de Marchena, A. B., Schuh, J. M., & Kelley, E. (2011). Language acquisition in autism spectrum disorders: A developmental review. Research in Autism Spectrum Disorders, 5(2), 681-691. https://doi.org/10.1016/j.rasd.2010.09.001 Escalera, E. A. (2009). Gender differences in children’s use of discourse markers: Separate worlds or different contexts? Journal of Pragmatics, 41(12), 2479-2495. https://doi.org/10.1016/j.pragma.2006.08.013 Frith, U., & Happé, F. (1994). Autism: beyond “theory of mind.” Cognition, 50(1–3), 115–132. https://doi.org/10.1016/0010-0277(94)90024-8 Fusaroli, R., Lambrechts, A., Bang, D., Bowler, D. M., & Gaigg, S. B. (2017). “Is voice a marker for Autism spectrum disorder? A systematic review and meta-analysis.” Autism Research, 10(3). https://doi.org/10.1002/aur.1678 Gillam, S. L., Hartzheim, D., Studenka, B., Simonsmeier, V., & Gillam, R. (2015). Narrative Intervention for Children With Autism Spectrum Disorder (ASD). Journal of Speech, Language, and Hearing Research, 58(3), 920-933. https://doi.org/10.1044/2015_jslhr-l-14-0295 Goldman, J. P. (2011). EasyAlign: An automatic phonetic alignment tool under Praat. In Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH (pp. 3233–3236).   148  Gregory, M. (2002). Phasal analysis within communication linguistics: two contrastive discourses. Relations and Functions within and around Language, 316-45. Grossman, R. B. (2015). Judgments of social awkwardness from brief exposure to children with and without high-functioning autism. Autism, 19(5), 580-587. https://doi.org/10.1177/1362361314536937 Grossman, R. B., Bemis, R. H., Plesa Skwerer, D., & Tager-Flusberg, H. (2010). Lexical and Affective Prosody in Children With High-Functioning Autism. Journal of Speech Language and Hearing Research, 53(3), 778. https://doi.org/10.1044/1092-4388(2009/08-0127) Happé, F., & Frith, U. (2006). The weak coherence account: Detail-focused cognitive style in autism spectrum disorders. Journal of Autism and Developmental Disorders, 36(1), 5–25. https://doi.org/10.1007/s10803-005-0039-0 Heier, F., Simmel, M., & College, S. (1944). An experimental study of apparent behavior. The American Journal of Psychology. Hendricks, D. (2010). Employment and adults with autism spectrum disorders: Challenges and strategies for success. Journal of Vocational Rehabilitation, 32(2), 125-134. https://doi.org/10.3233/JVR-2010-0502 Hiller, R. M., Young, R. L., & Weber, N. (2016). Sex differences in pre-diagnosis concerns for children later diagnosed with autism spectrum disorder. Autism, 20(1), 75–84. https://doi.org/10.1177/1362361314568899 Hughes, D. L., McGillivray, L., & Schmidek, M. (1997). Guide to narrative language: Procedures for assessment. Eau Claire, WI: Thinking Publications. Hull, L., Petrides, K. V., Allison, C., Smith, P., Baron-Cohen, S., Lai, M. C., & Mandy, W. (2017). “Putting on My Best Normal”: Social Camouflaging in Adults with Autism Spectrum Conditions. Journal of Autism and Developmental Disorders, 47(8), 2519–2534. https://doi.org/10.1007/s10803-017-3166-5 Hurlbutt, K., & Chalmers, L. (2004). Employment and Adults With Asperger Syndrome. Focus on Autism and Other Developmental Disabilities, 19(4), 215-222. https://doi.org/10.1177/10883576040190040301 Kauschke, C., & Klann-Delius, G. (1997). The acquisition of verbal expressions for internal states in German. The language of emotions, 173-194.     149  Kauschke, C., van der Beek, B., & Kamp-Becker, I. (2016). Narratives of Girls and Boys with Autism Spectrum Disorders: Gender Differences in Narrative Competence and Internal State Language. Journal of Autism and Developmental Disorders, 46(3), 840–852. https://doi.org/10.1007/s10803-015-2620-5 Kehler, A. (2002). Coherence, reference, and the theory of grammar. CSLI Publishers. Kibble, R., and R. Power. Kenny, L., Hattersley, C., Molins, B., Buckley, C., Povey, C., & Pellicano, E. (2016). Which terms should be used to describe autism? Perspectives from the UK autism community. Autism, 20(4), 442-462. https://doi.org/10.1177/1362361315588200 Kissine, M., & Geelhand, P. (2019). Brief Report: Acoustic Evidence for Increased Articulatory Stability in the Speech of Adults with Autism Spectrum Disorder. Journal of Autism and Developmental Disorders, 49(6), 2572-2580. https://doi.org/10.1007/s10803-019-03905-5 Klin, A. (2000). Attributing social meaning to ambiguous visual stimuli in higher-functioning Autism and Asperger syndrome: The social attribution task. Journal of Child Psychology and Psychiatry and Allied Disciplines, 41(7), 831–846. https://doi.org/10.1017/S0021963099006101 Knott, A., & Sanders, T. (1998). The classification of coherence relations and their linguistic markers: An exploration of two languages. Journal of pragmatics, 30(2), 135-175. Lai, M. C., Lombardo, M. V., Ruigrok, A. N. V., Chakrabarti, B., Auyeung, B., Szatmari, P., … Baron-Cohen, S. (2017). Quantifying and exploring camouflaging in men and women with autism. Autism, 21(6), 690-702. https://doi.org/10.1177/1362361316671012 Larkin, F., Hobson, J. A., Hobson, R. P., & Tolmie, A. (2017). Collaborative competence in dialogue: Pragmatic language impairment as a window onto the psychopathology of autism. Research in Autism Spectrum Disorders, 43–44, 27–39. https://doi.org/10.1016/j.rasd.2017.09.004 Lê, K., Coelho, C., Mozeiko, J., & Grafman, J. (2011). Measuring goodness of story narratives. Journal of Speech, Language, and Hearing Research, 54, 118–126. Lenth, R. V. (2016). Least-Squares Means: The R Package lsmeans. Journal of Statistical Software, 69(1), 1–33. https://doi.org/10.18637/jss.v069.i01 Little, L. M., Wallisch, A., Salley, B., & Jamison, R. (2017). Do early caregiver concerns differ for girls with autism spectrum disorders? Autism, 21(6), 728–732. https://doi.org/10.1177/1362361316664188  150  Lord, C., Rutter, M., DiLavore, P., Risi, S., Gotham, K., & Bishop, S. (2012). Autism Diagnostic Observation Schedule, Second Edition. Torrance, CA: Western Psychological Services. Lord, C, Risi, S., Lambrecht, L., Cook, E. H., Leventhal, B. L., DiLavore, P. C., … Rutter, M. (2000). Autism Diagnostic Observation Schedule (ADOS). Journal of Autism and Developmental Disorders, 30(3), 205-223. https://doi.org/10.1007/BF02211841. Lord, Catherine, Rutter, M., & Le Couteur, A. (1994). Autism Diagnostic Interview-Revised: A revised version of a diagnostic interview for caregivers of individuals with possible pervasive developmental disorders. Journal of Autism and Developmental Disorders, 24(5), 659–685. https://doi.org/10.1007/BF02172145 Losh, M., & Capps, L. (2003). Narrative ability in high-functioning children with autism or Asperger’s Syndrome. Journal of Autism and Developmental Disorders, 33(3), 239–251. https://doi.org/10.1023/A:1024446215446 MacDonald, R., Parry-Cruwys, D., Dupere, S., & Ahearn, W. (2014). Assessing progress and outcome of early intensive behavioral intervention for toddlers with autism. Research in Developmental Disabilities, 35(12), 3632–3644. https://doi.org/10.1016/j.ridd.2014.08.036 Manolitsi, M., & Botting, N. (2011). Language abilities in children with autism and language impairment: Using narrative as a additional source of clinical information. Child Language Teaching and Therapy, 27(1), 39-55. https://doi.org/10.1177/0265659010369991 Matei, M. G. (2011). The Influence of Age and Gender on the Selection of Discourse Markers in Casual Conversations. Bulletin of the Transilvania University of Braşov, Series IV: Philology & Cultural Studies, 4(1), 213–220. McCann, J., & Peppé, S. (2003). Prosody in autism spectrum disorders: A critical review. International Journal of Language and Communication Disorders, 38(4), 325-350. https://doi.org/10.1080/1368282031000154204 Mertens, P. (2004). The Prosogram: Semi-Automatic Transcription of Prosody Based on a Tonal Perception Model. In Proceedings of the 2nd International Conference on Speech Prosody. Milton, D. E. M. (2012). On the ontological status of autism: The “double empathy problem.” Disability and Society, 27(6), 883-887. https://doi.org/10.1080/09687599.2012.710008    151  Morrison, K. E., DeBrabander, K. M., Faso, D. J., & Sasson, N. J. (2019). Variability in first impressions of autistic adults made by neurotypical raters is driven more by characteristics of the rater than by characteristics of autistic adults. Autism. https://doi.org/10.1177/1362361318824104  Müller, N., Guendouzi, J. A., & Wilson, B. (2008). Discourse analysis and communication impairment. The handbook of clinical linguistics, 3-31. Nippold, M. A., Frantz-Kaspar, M. W., & Vigeland, L. M. (2017). Spoken language production in young adults: Examining syntactic complexity. Journal of Speech, Language, and Hearing Research,  60(5), 1339-1347. https://doi.org/10.1044/2016_JSLHR-L-16-0124 Norbury, C. F., & Bishop, D. V. M. (2003). Narrative skills of children with communication impairments. International Journal of Language & Communication Disorders, 38(3), 287–313. https://doi.org/10.1080/136820310000108133 Ozonoff, S., Pennington, B. F., & Rogers, S. J. (1991). Executive function deficits in high functioning autistic individuals: relationship to theory of mind. Journal of Child Psychology and Psychiatry, 32(7), 1081–1106. https://doi.org/10.1111/j.1469-7610.1991.tb00351.x Parish-Morris, J., Liberman, M. Y., Cieri, C., Herrington, J. D., Yerys, B. E., Bateman, L., … Schultz, R. T. (2017). Linguistic camouflage in girls with autism spectrum disorder. Molecular Autism, 8(1). https://doi.org/10.1186/s13229-017-0164-6 Park, C. J., Yelland, G. W., Taffe, J. R., & Gray, K. M. (2012). Morphological and syntactic skills in language samples of pre school aged children with autism: Atypical development? International Journal of Speech-Language Pathology, 14(2), 95–108. https://doi.org/10.3109/17549507.2011.645555 Paul, R., Augustyn, A., Klin, A., & Volkmar, F. R. (2005). Perception and production of prosody by speakers with autism spectrum disorders. Journal of Autism and Developmental Disorders, 35(2), 205–220. https://doi.org/10.1007/s10803-004-1999-1 Peppé, S., McCann, J., Gibbon, F., O’Hare, A., & Rutherford, M. (2006). Assessing prosodic and pragmatic ability in children with high-functioning autism. Journal of Pragmatics, 38(10), 1776–1791. https://doi.org/10.1016/j.pragma.2005.07.004 Perkins, M. R. (2010). Pragmatic Impairment. The Handbook of Language and Speech Disorders, (March 2010), 227–246. https://doi.org/10.1002/9781444318975.ch10    152  Petersen, D. B., Brown, C. L., Ukrainetz, T. A., Wise, C., Spencer, T. D., & Zebre, J. (2014). Systematic Individualized Narrative Language Intervention on the Personal Narratives of Children With Autism. Language, Speech, and Hearing Services in Schools, 45(1), 67-86. https://doi.org/10.1044/2013_lshss-12-0099 Portway, S. M., & Johnson, B. (2005). Do you know I have Asperger’s syndrome? Risks of a non-obvious disability. Health, Risk and Society, 7(1), 73-83. https://doi.org/10.1080/09500830500042086 R Core Team. (2016). R Core Team (2016). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL Http://Www.R-Project.Org/., R Foundation for Statistical Computing. Reichow, B. (2012). Overview of meta-analyses on early intensive behavioral intervention for young children with autism spectrum disorders. Journal of Autism and Developmental Disorders, 42(4), 512–520. https://doi.org/10.1007/s10803-011-1218-9 Revelle, W. (2017). How To: Use the psych package for Factor Analysis and data reduction, 86. Retrieved from http://personality-project.org/r/psych/HowTo/factor.pdf Sanders, T. J. M., & Noordman, L. G. M. (2000). The role of coherence relations and their linguistic markers in text processing. Discourse Processes, 29(1), 37-60. https://doi.org/10.1207/S15326950dp2901_3 Sanders, T., & Spooren, W. (2009). The cognition of discourse coherence. Discourse, of course, 197-212. Sasson, N. J., Faso, D. J., Nugent, J., Lovell, S., Kennedy, D. P., & Grossman, R. B. (2017). Neurotypical Peers are Less Willing to Interact with Those with Autism based on Thin Slice Judgments. Scientific Reports, 7, 40700. https://doi.org/10.1038/srep40700 Schegloff, E. A. (2000). Overlapping talk and the organization of turn-taking for conversation. Language in Society, 29(1), 1-63. https://doi.org/10.1017/s0047404500001019 Schmitz, C. (2012). Limesurvey: an open source survey tool. Limesurvey Project, Hamburg, Germany. Retrieved from http://www.limesurvey.org Selting, M. (2000). The construction of units in conversational talk. Language in Society, 29(4), 477-517. https://doi.org/10.1017/S0047404500004012 Shattuck, P. T., Narendorf, S. C., Cooper, B., Sterzing, P. R., Wagner, M., & Taylor, J. L. (2012). Postsecondary education and employment among youth with an autism spectrum disorder. Pediatrics, 129(6), 1042-1049. https://doi.org/10.1542/peds.2011-2864  153  Shriberg, L. D. (2001). Speech and Prosody Characteristics of Adolescents and Adults With High-Functioning Autism and Asperger Syndrome. Journal of Speech, Language, and Hearing Research, 44(5), 1097–1115. https://doi.org/10.1044/1092-4388(2001/087) Siller, M., Swanson, M. R., Serlin, G., & Teachworth, A. G. (2014). Internal state language in the storybook narratives of children with and without autism spectrum disorder: Investigating relations to theory of mind abilities. Research in Autism Spectrum Disorders, 8(5), 589–596. https://doi.org/10.1016/j.rasd.2014.02.002 Solomon, O. (2004). Narrative introductions: discourse competence of children with autistic spectrum disorders. Discourse Studies, 6(2), 253-276. Sterponi, L., & de Kirby, K. (2016). A Multidimensional Reappraisal of Language in Autism: Insights from a Discourse Analytic Study. Journal of Autism and Developmental Disorders, 46(2), 394–405. https://doi.org/10.1007/s10803-015-2679-z Sterponi, L., de Kirby, K., & Shankey, J. (2014). Rethinking language in autism. Autism, 19(5), 517–526. https://doi.org/10.1177/1362361314537125 Stirling, L., Douglas, S., Leekam, S., & Carey, L. (2014). The use of narrative in studying communication in Autism Spectrum Disorders: A review of methodologies and findings. Communication in Autism, (July 2016), 171–215. Taboada, M. (2014). Implicit and explicit coherence relations. In Discourse, of course. Amsterdam: John Benjamins, 127-140. https://doi.org/10.1075/z.148.13tab Tager-Flusberg, H. (1995). “Once upon a ribbit”: Stories narrated by autistic children. British Journal of Developmental Psychology, 13, 45–59. https://doi.org/10.1111/j.2044-835X.1995.tb00663.x Tanguy, N., Van Damme, T., Degand, L., & Simon, A.-C. (2012). Projet FRFC " Périphérie gauche des unités de discours " - Protocole de codage syntaxique. Volden, J., Coolican, J., Garon, N., White, J., & Bryson, S. (2009). Brief report: Pragmatic language in autism spectrum disorder: Relationships to measures of ability and disability. Journal of Autism and Developmental Disorders, 39(2), 388–393. https://doi.org/10.1007/s10803-008-0618-y Volden, J., Dodd, E., Engel, K., Smith, I. M., Szatmari, P., Fombonne, E., … Duku, E. (2017). Beyond Sentences: Using the Expression, Reception, and Recall of Narratives Instrument to Assess Communication in School-Aged Children With Autism Spectrum Disorder. Journal of Speech Language and Hearing Research, 60(8), 2228. https://doi.org/10.1044/2017_JSLHR-L-16-0168  154  Whitworth, A., Claessen, M., Leitão, S., & Webster, J. (2015). Beyond narrative: Is there an implicit structure to the way in which adults organise their discourse? Clinical Linguistics & Phonetics, 29(6), 455–481. https://doi.org/10.3109/02699206.2015.1020450  Zwaan, R. A., & Radvansky, G. A. (1998). Situation models in language comprehension and memory. Psychological bulletin, 123(2), 162.    155  Appendix 1 : Guidelines for orthographic transcription (French Version)  I. Règles d’Or de la transcription orthographique  1. Le langage doit est transcrit tel qu’il est produit (transcription littérale) et non comme il est prévu d’être produit ou comme il est typiquement attendu d’être produit. 2. Ne jamais corriger ce que l’on entend pour le rendre plus acceptable. 3. A l’oral, la notion de phrase n’a aucun sens et n’est donc pas représentée dans la transcription. 4. Seulement les énoncés qui représentent une prise de parole ininterrompue du locuteur sont représentés dans la transcription. Une transcription ne comporte donc pas de points, virgules ou points-virgules. De même, les énoncés ne commencent pas par une majuscule (réservée aux noms propres).   II. Procédure de la transcription orthographique  La transcription orthographique a pour but principal de mettre par écrit le contenu des conversations et se fait dans le niveau ‘ortho’ du TextGrid. A ce stade de la transcription, les variations phonétiques et les éléments interactionnels (ex. chevauchement de paroles, silence) ne sont pas transcrit (seulement délimités par une frontière). Ces éléments seront inclus lors de la phase de conversion en transcription phonétique (niveau ‘phono’) ainsi que la procédure de segmentation syntaxique et phonétique (niveau ‘words’ et ‘syll’).     156   2.1. Etapes de la transcription orthographique 1. Ouvrir les fichiers .wav et .TextGrid (portant le même nom) dans Praat. Sélectionner les deux fichiers et cliquer sur ‘View & Edit’.      157  2. Écouter la conversation plusieurs fois afin d’identifier et se familiariser avec les participants et thème(s) de la conversation. 3. Rendre une transcription orthographique de la conversation dans le niveau ‘ortho’. Dans un premier temps, transcrire des séquences de paroles délimités par des pauses silencieuses et changement de tour de parole. Pour insérer une frontière dans le TextGrid, cliquer sur l’onglet ‘Boundary’. Ensuite sélectionner ‘Add Boundary’ et ajouter la frontière sur le(s) niveau(x) désiré(s). Par après, repasser sur la transcription et créer des segments de paroles de max 2 secondes (dans la mesure du possible). 4. Ne pas inventer de nouvelles formes écrites afin de simuler la prononciation orale (il existe des symboles spécifiques et standardisés pour représenter certains aspects de la prononciation, ceux-ci seront représentés dans la transcription phonétique).  5. Les segments de parole incompréhensibles ou inaudibles seront représentés par 3 Xs : xxx 6. Les pauses silencieuses sont marquées par le symbole underscore : _  7. Les sons qui n’appartiennent pas à la parole (ex. toux, soupir, rire) sont marqués par le symbole dièse : # 8. Lors de la transcription de conversations, une difficulté surgit lors de la transcription d’énoncés produits en même temps par les deux interlocuteurs. A ce stade de la transcription orthographique, isolée par des frontières les paroles simultanées/qui se chevauchent le plus précisément possible, c'est-à-dire transcrire à l’intérieur des frontières, seulement les mots qui se chevauchent.  9. Lorsque la transcription nécessite une apostrophe, insérer un espace après l’apostrophe mais pas avant : c’ est 10. Vérifier l’orthographe de la transcription (MAIS PAS CORRIGER LES FAUTES PRODUITES À L’ORAL !). Voir Règles d’Or 2 et 3.  11. Vérifier que ce qui est dit oralement correspond bien à ce qui est transcrit (ex. pas de mots insérés ou oubliés).   2.2. Quelques conseils en cas de difficultés pour transcrire  1. Dans le cas où la parole à transcrire est difficilement compréhensible, alterner une écoute globale (le segment de parole incompréhensible dans son contexte, c-à-d, +/- 5 secondes) et locale (seulement le bout de parole incompréhensible). 2. Parfois il est difficile de différencier une hésitation, mot de remplissage et syllabes prolongées.       158  Appendix 2 : Annotation guidelines for narratives (French Version)  I. Description du matériel  Matériel: le matériel d’annotation provient de la tâche ADOS Raconter une histoire à partir d’un livre sans parole. Le livre utilisé pour obtenir l’histoire est Tuesday (David Wiesner, 1991).  Résumé de l’histoire : Un mardi soir, des grenouilles et crapauds acquièrent soudain la capacité de voler sur leurs nénuphars, et commencent à explorer les environs, surprenant ceux qui sont encore debout. Au lever du soleil, ils perdent leur capacité de voler. Des policiers sont confus lorsqu’ils trouvent les nénuphars par terre dans les rues. Le mardi suivant, c’est les cochons qui se mettent à voler.  Protagonistes principaux : grenouilles/crapauds Protagonistes secondaires : corbeaux, monsieur, vieille dame regardant la télévision, chat, chien, policiers, cochons Milieux : étang, ciel, village, cuisine, jardins, maisons, télévision, rues, ferme  II. Catégories de codage  2.1. Structure de l’histoire  2.1.1. Eléments principaux La structure narrative sera analysée en comptant le nombre d’éléments de la grammaire d’histoire (c’est-à-dire, la verbalisation de cet élément).  2.1.1.1. Setting (S) Le setting initial est introduit par l’expérimentatrice (selon les instructions de l’ADOS).  2.1.1.2. Evènement Déclencheur (ED1) Evénement principal 1 (selon Kauschke et al., 2016; Rumpf et al., 2012): Les grenouilles/crapauds acquièrent soudainement la capacité de voler sur leur nénuphar. Cet élément est également introduit par l’expérimentatrice (selon les instructions de l’ADOS).    159  2.1.1.3. Séquences d’évènements (SE)  - Evénement A (SE_A) La tortue et les poissons regardent les crapauds/grenouilles s’envoler  PP A13: donc soudainement la tortue voit des grenouilles en lévitation en l air et donc ça l a surpris ainsi que les poissons  - Evénement B (SE_B) Les grenouilles/crapauds survolent/découvrent les environs.  PP A35: ils s‘éloignent et là on voit qui se trouvent au-dessus des champs et d’habitations PP C27: ils se dirigent tous vers les secteurs résidentiels  - Evènement C (SE_C) Elles volent au milieu des corbeaux.  PP A32: proche d’une line éléctrique où y‘ a des oiseaux et ils s’amusent à leur faire peur PP A35: euh ils passent ensuite près de fils éléctriques et coursent euhm un corbeau  - Evénement D (SE_D) Elles passent devant un monsieur qui est en train de prendre une collation de minuit.  PP A32: euh y‘ a une personne habitant dans la maison est surprise de voir ce qui se passe derrière avec les crapauds qui sont en train de voler alors qu’elle est en train de manger un sandwich et boire un per de lait  - Evénement E (SE_E) Les grenouilles/crapauds poursuivent leur aventure dans un jardin et se prennent le linge qui pend.  PP C27: ils se mettent à aller partout et n’importe où à se heurter dans des draps PP A26: et puis les grenouilles ils se prennent euh les les essuies les serviettes comme on dit  160  - Evénement F (SE_F) Visite d’une maison avec une vieille dame qui dort devant sa télévision  SE_F: Les grenouilles/crapauds rentrent par les fenêtres et la cheminée  PP C13: elles continuent leur chemin alors et là elles arrivent près d une maison et elles entrent par toutes les ouvertures ici on voit les grenouilles qui rentrent par la fenêtre ouverte des grenouilles qui rentrent par la cheminée  SE_F1 : elles visitent la maison d’une vieille dame qui s’est endormie devant la télévision  PP A32: elles sont en train de regarder la télévision en face d’un euh bah en face d’une télévision proche d’une vieille dame qui s’est assoupie  SE_F2 : celle-ci ne se rend pas compte de ce que font les grenouilles  PP A26: ils prennent euh le contrôle de la télévision chez une vieille dame qui ne l‘ a même pas remarqué  SE_F3: les grenouilles ont pris sa télécommande et regarde la télévision.    PP A17: ils savent util- manipuler une télécommande  - Evénement G (SE_G) Rencontre avec un chien  SE_G1 : une des grenouilles rencontrent un chien qui la poursuit.  PP A32: un chien sort et euh vient perturber le cra- qui est en train de voler  SE_G2: Les autres grenouilles viennent en renfort et pourchassent le chien.  PP A32: et les crapauds reviennent en nombre et arrivent à effrayer le chien  161  2.1.1.4. Evènement Décisif (ED2) Evénement principal 2 (selon Kauschke et al., 2016; Rumpf et al., 2012): Le soleil commence à se lever et les grenouilles/crapauds perdent leur capacité de voler.   PP A17 : ah comme cendrillon leur feuilles magiques perd leur pouvoir pendant la journée PP C34 : et le soleil se lève à ce moment là evidemment la magie s’arrête les les feuilles volantes ne volent plus  2.1.1.5. Conclusion (Conc) Elles tombent du ciel et rejoignent leur étang.  PP C34 : et donc les crapauds se retrouvent et doivent rentrer à pied dans leur lac  2.1.1.6. Codas (Coda1 ou Coda2) Coda 1 : Le matin, dans la ville, des policiers sont confus de découvrir des nénuphars dans les rues.   PP A13 : et toute la police se mobilise en essayant de trouver une explication à ce phénomène hors du commun  Coda 2 : Le mardi suivant, il y a l’ombre d’un cochon sur le toit d’une ferme.   PP C34 : la nuit arrive la prochaine nuit arrive de nouveau et c’est les cochons qui s’envolent  2.1.2. Eléments additionnels 2.1.2.1. Evénements secondaires (ES) Lorsque le participant verbalise un événement qui n’est pas un des éléments principaux mais qui et qui est présent dans l’image et n’est pas une simple description, cet évènement est codé comme un événement secondaire.  PP A23 : et en fait ces grenouilles ont l’air peut être d’espèces ou de races différentes à cause de de leurs couleurs PP A31 : y’en a d’autres qui se désintéressent de la télé en regardant les peintures    162  2.1.2.2. End (End) Le participant verbalise explicitement la fin de l’histoire.  PP A22 : fin de l’histoire  2.1.2.3. Description de l’image (DI)  Le participant décrit seulement ce qui est sur l’image, sans mentionner d’actions ou d’états psychologiques des personnages. Les descriptions d’images prennent souvent la forme d’expressions nominales simples.  PP A6: la télévision PP A29: paysage aquatique  2.1.3. Eléments perturbateurs 2.1.3.1. Eléments additionnels (EA) Lorsque le participant mentionne des éléments qui ne se passent pas dans l’histoire.  PP A27: là ici y a des poissons qui qui font un spectacle ou quoi y a y a une révolution entre les grenouilles et les tortues PP A27: et les aliens c‘est des connards  2.1.3.2. Commentaires (Comm) Ce codage s’applique aux commentaires produits par le participant concernant l’histoire ou la tâche narrative en elle-même.  A31: le genre de truc qui me fait chier  2.2. Référence  Référence sera mesurée en comptant le nombre de références faites aux protagonistes de l’histoire et les décors. Afin de mesurer comment les participants font références aux personnages et décors, leur forme syntaxique sera codée (et ensuite assigné un score).    163  2.2.1. Expressions référentielles définies - Expression nominale définie (END)  PP A3: les grenouilles et les nénuphars s‘envolent PP C31: du coup le chien va dans le sens inverse il s’échappe  - Expression pronominale (EPP)  PP A10: ils continuent leur chemin      PP C31: et elles sont pas contentes elles ont l‘air de râler un bon coup de pas pouvoir encore bénéficier de ce pouvoir magique PP C26: mais ils rencontrèrent pas le père noël ils rencontrèrent une dame assise sur un grand fauteuil devant sa télévision et elle dormait  - Expression adjectivale démonstrative (EAD)  PP A13: mais cette grenouille reçoit des renforts de ces amis qui terrorisent le chien  - Expression pronominale démonstrative(EPD) :  PP C32: mais pourtant ils en trouvent une ouverte ceux qui ne trouvent pas la seule fenêtre d‘ouverte rentrent par la cheminée  2.2.2. Expressions référentielles indéfinies - Expression nominale indéfinie (ENI)  PP C26: malheureusement y a un crapaud qui s est pris des draps PP A35: une grenouille est complètement entortillée dans le chiffon PP A13: donc soudainement la tortue voit des grenouilles en lévitation en l‘air  - Expression adjectivale indéfinie (EAI)  PP A34: et voilà certaines en profitent pour se prendre dans les dans le linge qui qui sèche  164   - Expression pronominale indéfinie (EPI)  PP C29: et puis tout à coup y a quelqu'un dans sa cuisine qui déjeune tardivement qui aperçoit par la fenêtre les crapauds volants  2.3. Connecteurs  Les connecteurs marquent une relation entre deux unités syntaxiques et seront codés selon le type de relation qu’ils expriment. L’utilisation des connecteurs dans les narratives sera mesurée en comptant le nombre de connecteurs totaux et par sous-type.   2.3.1. Relation temporelle (T) Exemples: après, ensuite, le matin, à 15 heures, puis, avant  PP C11: ils vont chez une mamy qui est en train de dormir puis pour regarder la télé tous ensemble  2.3.2. Relation additive (A) Exemples: et, de plus  PP C25: une grenouille traverse un jardin et euh rencontre un chien tout excité  2.3.3. Relation causale (Ca) Exemples: parce que, car, puisque, donc, du coup  PP A23 : mais là c’est assez étrange parce que là c’est comme si sur ces trois dernières images elles étaient revenues à leur état stable PP C6 : là y’a beaucoup de vent du coup les tortues qui sont amusées font des tours fin des tourbillons PP A21 : malheureusement ça ne fait pas des heureux car la police intriguée par ce qui se passait arrivèrent    165  2.3.4. Relation de contraste (Co) Exemples: mais, cependant, néanmoins, bien que, par contre  PP C29: donc les poissons volants les poissons les crapauds volants eux par contre ils ont l'air d‘être de bien s’amuser PP A34 : et le lendemain en soirée cette fois c’est plus les grenouilles mais c’est c’est les cochons qui qui s’envolent  2.3.5. Marqueur de discours (MD) Exemples: bah, ah, alors,  PP A4: bah ils passent devant une maison  2.4. Langage d’état interne  Le langage d’état interne fait référence aux états intérieurs des protagonistes de l’histoire. Les catégories de codage ont été définis selon celles proposées par Kauschke & Klann-Delius (1997) et Kauschke, van der Beek & Becker (2016). L’utilisation du langage d’état interne dans les narratives sera mesurée en comptant le nombre de mot pour chaque catégorie de langage d’état interne.  2.4.1. Emotion (E) Termes faisant référence à des émotions distinctes ou des comportements expressifs d’émotions.   PP A32: et donc la tortue a peur PP C4: les grenouilles elles sont contentes de voler  2.4.2. Physiologie (P)  Termes faisant référence à des perceptions et sensations physiques et biologiques subjectives  PP A30: les le monsieur les voit euh par la fenêtre PP A34 : il entend il entend un bruit bizarre  166   2.4.3. Cognition (C)  Termes faisant référence à des états mentaux/cognitifs, expression de connaissance/croyance/souvenir.   PP A4: ils se demande probablement pourquoi y a autant de nénuphars sur la route PP C22: elles se sont dit que là que c’était quand même un peu risqué ce qu’elles avaient fait  2.4.4. Modalité (M) Termes de volition, obligations et intentions  PP A34: parce qu‘en fait elles veulent regarder la télé PP C22: elles décident de prendre un petit bain  2.4.5. Evaluation (Ev) Termes qui expriment des jugements moraux ou évalue des personnes/évènements  PP A31: elles s’abrutissent devant la télé PP C19: donc là y‘a un truc qui vole c’est bizarre    167  Appendix 3 : Examples of narratives  Autistic participant, female, 40 years old Story Structure Participant’s production SE_A ben la tortue voit toutes les grenouilles qui qui s’envolent ES ben y’en a partout dans le dans le ciel SE_B là elles passent au-dessus des maisons SE_C Not verbalized SE_D les le monsieur les voit par la fenêtre SE_E  elles passent dans le jardin de la maison en faisant tomber tout le linge qui était dehors SE_F elles rentrent par les fenêtres et par la cheminée SE_F1 elles sont rentrées dans la maison d’une vieille dame qui s’est endormie dans le fauteuil SE_F2 Not verbalized SE_F3 elles regardent la télé SE_G1 bah elle allait atterrir mais y a un chien SE_G2 toutes les autres grenouilles font partir le chien ED2 Not verbalized Conc elles continuent à voler et elles retournent dans la marre Coda1  le lendemain le monsieur qui les a vues par la fenêtre raconte aux journalistes et la police trouve les nénuphars sur lesquels elles volaient Comm je sais pas ce que c’est en fait Coda2 ah oui c’est un cochon qui vole et là le mardi suivant c’est le cochon     168  Comparison participant, female, 39 years old Story Structure Participant’s production SE_A donc la tortue je pense que ce qu’elle a vu c’est les crapauds volants et du coup elle semble avoir quand même un peu peur et j’ai l’impression qu’elle est en train d’essayer de rentrer dans sa dans sa carapace et même les poissons les ont vus parce qu’ ils ont la tête en l’air ils regardent ES donc les poissons volants les poissons les crapauds volants eux par contre ils ont l'air d’être de bien s’amuser ils font des petites des petites pirouettes sur leurs feuilles de nénuphar  SE_B et puis là ils ont l'air de se diriger vers un vers un village parce que y’a le clocher de l’église qui est allumé et ils arrivent jusqu’au jusque dans un petit lotissement on va dire avec plein de grandes maisons donc ils survolent ça SE_C et quand ils arrivent ben justement dans le cœur du village les oiseaux sont effrayés et donc s’en vont à toute vitesse par contre ça a l'air de bien faire rire les crapauds SE_D et donc les voilà qui passent devant la fenêtre d’un d’un monsieur qui visiblement prend son petit dej et ouais ils passent devant sa fenêtre lui il est les a visible fin je pense qu’il les a entendus par contre il les a pas encore vus SE_E  donc les crapauds continuent leur leur passage entre les maisons et se prennent le linge qui sèche du coup avec les serviettes les torchons ils se sont faits des petites capes donc ça va encore plus vite SE_F ils arrivent encore plus vite à la maison suivante qui elle avait laissé les fenêtres fin les résidents avaient laissé les fenêtres ouvertes et donc ils s’introduisent dans la maison SE_F1 donc c’est la maison d’une mamy qui s’est endormie devant sa télé SE_F2 Not verbalized SE_F3 et donc tous les crapauds volants sont obnubilés par la télé et y a le chef des crapauds qui change les chaînes SE_G1 et donc y a un petit crapaud lui qui est toujours resté dans le jardin et qui tombe nez à nez avec un chien qui n’a pas l’air d’apprécier sa visite  169  et qui se met à le poursuivre donc il fait demi-tour et le crapaud fait demi-tour et essaye d’éviter le chien SE_G2 mais c’est sans compter l’arrivée de tous ses copains crapauds qui à leur tour poursuivent le chien qui est qui s’enfuit Conc et donc là les crapauds ben quittent le quittent l’endroit où ils étaient parce que ils ont l'air d’être de se précipiter parce que le jour se lève et donc ils regagnent finalement à pieds fin si je peux à pattes la leur mare ED2 parce que ben les feuilles n’ont plus le pouvoir qu’elles avaient tout à l’heure à savoir elles sont toutes tombées par terre Coda1  et puis maintenant y’a un enquêteur qui vient sur place parce qu’il est interloqué d’avoir trouvé autant de nénufars et y’a le monsieur tout à l' heure-là qui prenait son petit dej qui est en train de témoigner que qu’il a entendu un bruit voilà venu du ciel Coda2 et voilà là c’est la fin de la journée on voit que le soleil est en train de re disparaitre donc je pense que les crapauds vont peut-être venir mais non ce sont des cochons ok des cochons volants  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            data-media="{[{embed.selectedMedia}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0384568/manifest

Comment

Related Items