Research on prosocial behaviour in autism increasingly considers online means of connecting through social networking sites (SNS). Mostly driven by clinical psychology, this research has been focused on psychosocial outcomes and therefore failed to consider how autistic users take up, adapt, and appropriate interactional functions of particular social media platforms. The present article addresses this gap by drawing on advances in interactional linguistics in two otherwise distinct fields ­– computer-mediated discourse analysis and autistic communication – to explore in depth, for the first time, the complexities of applying existing speech act taxonomies to tweets by autistic users. We propose an innovative combination of the CMC act taxonomy with an interactional perspective which, methodologically, advances research on speech acts on SNS while also allowing identification of communicative purposes in autistic users’ SNS posts. Building on our earlier study showing the prevalence of replies in autistic adults’ Twitter use, we combine a quantitative analysis of speech acts across 256 replies with a qualitative examination of interactional functions. Our findings demonstrate the importance of moving away from deficit-based approaches to autistic communication and show how linguistic approaches to computer-mediated discourse can help understand the different ways in which autistic people adapt SNS resources for social purposes.

Autism is a condition which affects “how people communicate and interact with the world” (National Autistic Society, 2021, n.p.). An estimated 1 in 100 people are autistic (Zeidan et al., 2022). Characteristics include repetitive and restrictive behaviours, anxiety, meltdowns and shutdowns, and differences in social interaction from non-autistic (allistic) people (Autistica, 2021). Early research into the condition referred to autism as causing “impairments” and a supposed inability to reflect on mental states known as deficits in “theory of mind” (Baron-Cohen, 2001). This approach, however, analysed autistic traits according to non-autistic, neurotypical norms. In contrast to this deficit model of autism, the neurodiversity movement (Davidson, 2008) led to the conceptualisation of autism as a difference in processing and being social. Recent studies have focused on identifying features of autistic communication (Heasman & Gillespie, 2019; Williams et al, 2021) as opposed to measuring communicative competence by neurotypical standards. This approach highlights the fact that communication requires all interactants (both autistic and allistic) to carry responsibility for the success of the interaction, which in turn leads to the need to investigate full conversations as opposed to focusing on the utterances of only one (autistic) participant, as was the case in earlier research (Bottema-Beutel, 2017). Recent studies have begun to demonstrate how previously considered “impairments” fulfil communicative and interactional functions, or social actions performed through an utterance, for autistic people (Maciejewska, 2019). These studies so far have focused on spoken interactions.

Autistic adults, however, report low engagement with in-person social opportunities, which may stem from stigmatising attitudes to autism and social inequities (DHSC, 2020). In this context, support agencies increasingly draw attention to social networking sites (SNS), as they have been shown to enable “social engagement” (Mazurek, 2013, p. 1711), connections over shared interests (Gillespie-Lynch et al., 2014, p. 13), and a growing “autistic culture” where autistic identity and experiences are at the centre of interactions (Davidson, 2008; Parsloe, 2015¹). Mazurek (2013, p. 1711), for example, found that out of 108 autistic participants, 79.6% reported using SNS, for an average of 3.2 hours per day. These participants suggested that advantages stemming from the computer-mediated mode, like the level of social and physical distance that can be maintained online, compensate for their different social abilities (Mazurek, 2013). Studies generally have found that initiating new relationships online is seen as easier by autistic people than initiating them offline² (Brownlow et al., 2015; Gillespie-Lynch et al., 2014). Moreover, maintaining friendships may be easier online, as digital communities can create a space to “study” friendships and improve social skills (Brownlow et al., 2015), although Gillespie-Lynch et al. (2014) suggest that the lack of emotional cues and response delays online may make the task more difficult. These psychology and communication studies, however, are based on self-reports rather than analysis of user posts and practices. As such, there is considerable scope for drawing on applications of interactional linguistics to computer-mediated discourse (Herring, 2004) to understand how autistic adults interact with other SNS users.

Instead of assuming that technology directly determines what kinds of interaction take place on SNS, approaches under the umbrella of computer-mediated discourse analysis consider how communicative functions of a particular platform can be taken up or creatively adapted by users. In relation to Twitter, for example, several communicative conventions contribute to the use of Twitter as an interactional medium: (1) direct address using the @ symbol, including in replies, (2) retweets, (3) quote tweets, (4) likes, and (5) hashtags. In our previous ethnographic study of autistic adult Twitter users, we combined observation of online activities with quantitative analysis of the frequencies with which each of these communicative conventions were used (Koteyko et al., 2022). The analysis showed a preference for Replies. The prevalencse of Replies appears specific to autistic users, as previous Twitter research found Replies and Broadcast tweets used in near equal proportions (Page, 2012). This article therefore focuses specifically on our participants’ Replies to answer the following questions:

1. What do autistic adults use the Twitter Reply function for, and to what extent do frameworks based on neurotypical use account for these uses?

2. To what extent do autistic Twitter Replies reflect autistic communicative competencies and preferences?

3. Do autistic adults use Replies differently when discussing autism versus other topics?

Unlike the above-mentioned psychological studies of SNS use, which primarily address issues of relational satisfaction and other psychosocial outcomes, our concern is with bottom-up, interactional approaches to autism, and how oral practices may be reproduced as well as reconfigured in computer-mediated settings. We therefore begin by bringing research on interaction in autism into closer dialogue with digital linguistics research on social media interactions.

A number of studies under the rubric of discourse and conversation analysis have demonstrated the importance of situated and contextualised analyses of conversations for identifying and understanding communicative and interactional functions specific to autistic people. By analysing the language of autistic children, Sterponi et al. (2015) for example, show how “features of autistic language (…) often reveal an attempt to mobilize the affordances of language to manage difficulties in relating to and identifying with the other” (p. 524). Autistic conversation can be characterised by various linguistic features including sudden topic shifts (Heasman & Gillespie, 2019; Maciejewska, 2019, 2020; Sterponi & Shankey, 2014), repetition and routine (Dobbinson et al., 1998; Maciejewska, 2019, 2020; Sterponi et al., 2015), a reliance on questions and answers (Dobbinson et al., 1998; Maciejewska, 2019), and a preference for what Belek (2018, p. 164) calls “explicitation.” As autistic people generally find implied meanings unclear, explicitation is the process of clarifying implied social meanings, both of utterances of the self and of others. Topic shifting refers to changes in conversational direction that are unexpected by neurotypical standards, which Sterponi and Shankey (2014) show function to divert attention towards a preferred topic. Similarly, repetition (of words, structures, and topics) has been perceived as a “deficit” without conversational purpose, yet Maciejewska (2019) shows that repetition often contributes to the conversation, for instance as an affirmative or mitigating response. Finally, the reliance on questions and answers as opposed to a “dialogue structure” (Seligman et al. 2003 in Maciejewska, 2019) may depend on the analysed data, with old autism research often reliant on interview-style data between the researcher and the autistic participant (Maciejewska, 2019). However, reliance on questions and answers also functions as conversational structure which aids autistic people who prefer routinised interactions (Lawson, 2020). These findings demonstrate that conversational features have different functions for autistic speakers compared to neurotypical speakers.

Having established that autistic conversational characteristics perform different functions from what they would in neurotypical conversations, it is reasonable to consider that intra-neurotype conversations would be more successful than inter-neurotype conversations. Heasman and Gillespie (2019) investigated this by analysing intra-autistic conversations. They found two characteristics in the construction of intersubjectivity. Firstly, there was a more “generous assumption of common ground” (Heasman & Gillespie, 2019, p. 916) than is generally present in neurotypical conversation, with speakers presuming shared understanding of discourse topics. Secondly, they identified “a low demand for coordination” (Heasman & Gillespie, 2019, p. 916), including “ignored turns, parallel dialogue (…), and misinterpretations.” While these features would generally lead to decreased success in neurotypical conversation, they had little impact on the interpersonal relationship between the autistic participants. This mismatch between neurotypical and autistic conversation was also demonstrated by Crompton et al. (2020), who showed that intra-neurotype conversations are generally more successful than inter-neurotype. The “low demand for co-ordination” echoes conversation analysis-based research on autistic-allistic interactions which showed that stories told by autistic narrators “did not embed opportunities for the co-participant to intervene in the joint realization of meaning” (Fasulo, 2019, p. 627). Instead, autistic people may choose to present narratives as “asides (i.e., concise, fast delivered self-contained anecdotes)” (Fasulo, 2019, p. 621).

With most studies having focused on face-to-face interactions, the question arises to what extent autistic adults mobilize “autistic resources” in computer-mediated communication and whether existing frameworks developed based on neurotypical data can be applied to autistic SNS use. In other words, in addition to “the affordances of language” (Sterponi et al., 2015), we need to consider the interaction between social practices and platform affordances³ (Costa, 2018). In this regard, Herring (2007) distinguishes between situation factors, which include the participants, the norms of communication, and the linguistic code, and medium factors (e.g., the (a)synchronous mode, opportunity to post anonymously, persistence of transcript, and size of the message). On a general level, medium factors such as asynchronicity may reflect communicative preferences of autistic people by removing the need for timely turn coordination and eye contact, whereas persistence of transcript (Costa, 2018; Herring, 2007) ensures availability of recorded interactions which can be accessed at one’s own pace (Davidson, 2008; Koteyko et al., 2022). The opportunities to design and edit messages as well as multimodal resources like GIFs, emojis, and hashtags, which can aid in clarifying and understanding intended meaning, are further examples of medium factors relevant for autistic communication. At the more specific level of individual platforms, Twitter contrasts with other online social networks such as Facebook in that it affords brief text-based communication and allows pseudonymous accounts and interaction with diverse, not necessarily pre-existing contacts. While concise character limit is a constraining element for all users, some practices emerging as a result of user adaptation to Twitter affordances (such as sharing of short stories that deviate from conventional narrative structure) can be an enabling factor from the perspective of autistic communication. However, existing analysis of these functions has not displayed any specific focus on autistic users and has generally focused on neurotypical Twitter use. This bias is particularly crucial when we consider the interactional purpose of Twitter, as autistic people have different preferences for interaction from neurotypical people (Idriss, 2021). Considering the recent shift in autistic conversational research to investigating functions of autistic conversational characteristics and the use of SNS by autistic people, there is a need to investigate how and to what extent online communicative and interactional functions are enacted by autistic users.

In terms of situation factors, one common content feature of digital autistic communication is the focus on autism itself. SNS provide a space for autistic people to converge, share experiences, and redefine their autistic identity (Davidson, 2008; Parsloe, 2015). SNS provide an interactional space for autistic people, who may disprefer face-to-face interaction for various reasons including sensory overload or social anxiety, to participate in an autistic community (McGhee Hassrick et al., 2021) and advocate for autistic preferences (Davidson, 2008; Parsloe, 2015). The development of an online autistic community has allowed autistic users to “embrace their autistic identity” (Parsloe, 2015, p. 337; see also Bagatell, 2010), and online inter-autistic interactions consequently often discuss autism. These usages of online autistic spaces therefore suggest a link between autism as a topic of conversation and autistic-to-autistic interaction, and they allow for the study of inter-autistic communication by examining conversations about autism. While not a direct replacement, studying conversations about autism can approximate inter-autistic online discussion without the need for requesting personal information (i.e., whether someone is autistic).

We argue that examining SNS interactions by autistic users necessitates a functional approach that focuses on how meanings operate within the particular computer-mediated context in which they are created. While the argument has been made that “the main function of social media discourse is enacting relationships online” (Zappavigna, 2012, p. 192), the approaches to building relationships have been shown to be diverse. Examples of such approaches include Zappavigna’s (2012) research on how tweets function to invite digital “followers” to respond to and share evaluative bonds, research on the function of “small stories” in sharing personal experiences and creating rapport on Twitter (Bamberg & Georgakopoulou, 2008; Dayter, 2015), and the analysis of speech acts in digital discourse (Dayter, 2016). Such studies frequently draw on Goffman’s legacy to account for how SNS users do self-presentation and ensure “that events do not occur which might effectively carry an improper expression” (Goffman, 1971, p. 40). In light of the studies reviewed above, we aim to show how the application of such frameworks to the communication of autistic SNS users has to take into account autistic users’ unique impression management concerns, stemming from experiences of being misunderstood and the history of epistemic injustice, specifically the assumptions regarding autistic people’s lack of sociability (Catala, 2021). The resulting “camouflaging” strategies involve modification of autistic behaviours and copying non-autistic social skills (Hull et al., 2017) and have been observed in both offline and online environments (Cook et al, 2021).

In what follows, we first discuss the need for a function-based analysis to complement speech act analysis. We then use the CMC act taxonomy (Herring et al., 2005) to explore RQ1, identifying usage of Twitter Replies, and consider autistic communicative preferences and competencies in an online context (RQ2). We also compare the results across Replies discussing autism and those discussing other topics (addressing RQ3).

Speech act taxonomies have generally focused on spoken language (Austin, 1962; Francis & Hunston, 1992; Searle, 1979). The CMC act taxonomy (Herring et al., 2005) is the most elaborate adaptation of speech act theory for online discourse. It follows common speech act approaches in not considering the interaction between discourse function and form, as for instance Francis and Hunston (1992) do in their discussion of spoken speech acts. While Francis and Hunston’s (1992) approach may be unusual in this regard, considering the function of speech acts can be particularly useful when examining data of autistic interlocuters, as their discourse may have different functions from neurotypical interlocutors.

Austin (1962) first distinguished among types of speech acts focusing on performatives. Searle (1979) extended this work by including categorisation of declaratives. Searle’s (1979) speech act categories are assertives, commissives, directives, expressives, and declarations. While Searle’s acts are still used, many approaches have taken the categories and enhanced them by introducing more specific acts. Francis and Hunston (1992) base their approach on spoken discourse, with a focus on applicability to conversations. Their list of acts includes specific purposes such as “reply-greeting” and “reply-summons” (Francis & Hunston, 1992, pp. 125-127). These acts combine into moves, “the smallest free unit” of conversations (Sinclair & Coulthard, 1992, p. 4), which are closer to Searle’s (1979) speech acts (Francis & Hunston, 1992). Francis and Hunston (1992) also specify that speech acts are identified on a lexico-grammatical level and can contribute to meaning-making on higher levels, including conversational structure and interaction.

In order to analyse CMC texts, Herring, Das, and Penumarthy (2005) developed the CMC Act Taxonomy, based on Bach and Harnish (1979) and Francis and Hunston (1992). The taxonomy consists of 16 acts characteristic of CMC: Inquire, Request, Direct, Invite, Inform, Claim, Desire, Elaborate, Accept, Reject, React, Repair, Apologize, Thank, Greet, And Manage. Ge-Stadnyk (2021) added “Congratulate” as an additional category. Inform and Claim are the most used acts across social media platforms including Soundcloud (Ishizaki et al., 2013), lab chat rooms (Barlow, 2019), Facebook groups (March, 2020), and Twitter (Nemer, 2016), while React and Direct were also common on Soundcloud (Ishizaki et al., 2013) and Facebook groups (March, 2020). These findings are consistent with research using other speech act approaches, like Nastri et al. (2006), who identified assertives and expressives as the most common speech acts in ‘away’ messages online. Similarly, on Facebook Carr et al. (2012) identified the same top two speech acts in status updates, suggesting that social media is used for evaluative language (expressives / Claim) and informative language (assertives / Inform). However, the identification of these speech acts is on a lexico-grammatical level (Francis & Hunston, 1992) or as Herring (2004) refers to it, the “meaning level.” The question arises how Claim and Inform contribute to conversations on an interactional level. This is particularly important when analysing autistic conversations in order to understand the functions of autistic conversational preferences.

In order to go beyond the lexico-grammatical focus, we propose to complement the CMC Taxonomy by adding an analysis focused on what Herring (2004) refers to as social behaviour. Following SPAAC (2003) and Dayter (2016), we propose that this coding focuses on C-Units. C-Units are “independent clausal or non-clausal units. Functionally, they represent units which can be assigned to a given communicative function, represented by its speech act attribute” (SPAAC, 2003, p. 2). C-Units map onto CMC acts. The determination of C-Units in SPAAC is partly based on pauses, as SPAAC is based on spoken discourse. For our written data, pauses are not available, therefore we relied on SPAAC’s (2003) grammatical division of C-Units into independent clauses.

The first round of coding used the CMC act taxonomy (Herring et al., 2005) to analyse speech acts. The second round of coding analysed the interactional functions of the C-Unit within the context of the thread. This second coding focused on the three most common CMC act categories in our dataset: Inform, Claim, Elaborate. Inform and Claim are classified based on the “verifiability” of the content (Herring et al., 2005), while Elaborate is based on the interaction with a preceding C-Unit. However, definitions of Inform, Claim, and Elaborate currently do not account for their functions in the interactions. Introducing a secondary analysis of the interactional functions of these categories can provide insights into how autistic Twitter users apply CMC acts in interactional tweets and to what extent neurotypical frameworks of SNS are applicable to autistic users.

The following section outlines the seven interactional functions identified in the data. These functions are based on (1) existing research on Twitter communicative and interactional functions and (2) existing research on autistic conversational characteristics. These functions are specific to our dataset. For each function we provide an example with a paraphrased preceding tweet to demonstrate the interactional context. The preceding tweets are paraphrased as they are from non-participants who did not give consent to be included in the study. The coded C-Units are italicised, and the categories are presented alphabetically. (Participant codes in parentheses are explained in Table 1 in the next section.)

1. Answer.

The first category includes direct answers to questions in preceding tweets and consist of short phrases or single-word Replies. Q&A’s are common on Twitter (Soulier et al., 2016) and are often a preferred method of interaction for autistic people (Koteyko et al., 2022; Seligman et al., 2003 in Maciejewska, 2019).

Identification: repetition of some lexis from the preceding tweet and/or direct response to a question in the preceding tweet.

e.g., Preceeding tweet asks for confirmation about a quote tweet stating that Mark Hamill left the Colour the Spectrum campaign.

Coded response: “He ‘liked' the tweet” (T4)

2. Attitude.

This category draws on Zappavigna’s (2012, 2018) work on Appraisal on Twitter and captures the sharing of opinions and perspectives and aligning with others through evaluations. The presence of this code shows how autistic people express opinions and emotions and disproves old autism research that suggested autistic people struggle with empathy and emotionality (Baron-Cohen, 2001).

Identification: attitudinal language, including affect, appreciation, and judgement (Martin & White, 2005). It may include an object of evaluation or constitute an expressive act (Searle, 1979) without a clear object.

e.g., Preceeding tweet about the VLOOKUP function in Excel being useful but difficult to understand.

Coded response: “They [sic] are a true game changer!” (TF20)

3. Explain.

This category includes reasonings for opinions or information. It is differentiated from the Elaborate CMC Act in that it focuses solely on C-Units which start with causal conjunctions. As such, it helps specify the type of Elaboration. Explain also covers autistic preference for detail and justification of thoughts used to avoid conflict (Belek, 2018).

Identification: syntactic structures like “X because Y” and “X as Y” (Explain starts at the causal conjunction). Coded content may also include illustrations in the form of emojis (Ge & Herring, 2018) or images (Martinec & Salway, 2005).

e.g., Back-and-forth thread about whether white people can experience racism.

Coded C-Unit is part of a longer tweet:
“Your white race is the dominant culture so you wouldn’t experience the thing I described in reverse. So no ethnic minorities would ever refuse to be friends with you based on your race. It only happens to ethnic minorities. Your example sucks as it’s based on bad logic” (TF24)

4. Information.

This category includes C-Units that solely provide factual information without any evaluative content such as adverbs or emojis – for example opening hours, a telephone number, or a link (Ishizaki et al., 2013). While this is a narrow distinction from the Inform act in the CMC Act taxonomy, having a separate category for solely factual information allows us to code for the autistic preference for “specific and focused interactions” (Lawson, 2020, p. 521).

Identification: solely factual information, which may include additional wording, e.g., “here is an article,” but does not contain evaluative language.

e.g., Thread started by person being evicted with late notice. Participant TF12 responds with information on who to contact:

“Contact Shelter, or ask someone to do so on your behalf. They have a helpline on 08088004444. Open until 8.” (TF12)

5. Humour.

This category includes (attempts at) jokes, which Zappavigna (2012) shows are common on Twitter. It has been separated from Attitude, as the Humour category may include evaluative language which is not meant to accurately represent the author’s perspective and can be classified as “non bona fide” (Dresner & Herring, 2010).

Identification: the presence of features such as emojis, gifs, or words (‘lol’, ‘haha’) that indicate a humorous tone.

e.g., Longer thread between TF4 and another person. The other person firstly shared a new fact they learned, and TF4 responds: “I knew that.” The other person then shares another fact and TF4 responds: “I knew that too” (TF4)

6. Qualify.

This category includes any C-Units that provide a form of qualification to a succeeding or preceding C-Unit. This category captures both autistic identity management and tone, with both used as clarification (Belek, 2018).

Identification: a claim to authority (e.g., “I am autistic”) or through an adverb (e.g., “genuinely”).

e.g., Preceding tweet talks about how driving can be particulary difficult for autistic people. Participant TF5 responds to discuss how driving is not even necessary to learn: “I'm 50 & never learnt to drive because I've never needed or wanted to. Cars are harmful to the environment and I struggle with the smell of petrol.” (TF5)

7. Small Stories.

This category is based on the Small Stories framework (Bamberg & Georgakopoulou, 2008). It accounts for autistic storytelling used as a “self-contained anecdote” (Fasulo, 2019, p. 621).

Identification: tellings of personal experiences (see also Dayter, 2015), reporting of other people’s actions and speech, and discussions of possible happenings in the future (Georgakopoulou, 2007).

e.g., Participant TF13 builds on a preceding tweet from a friend who talks about hearing a persistent ringing in her ears. While the C-Unit by itself is not a story, it builds on the preceding tweet to create a shared story: “Me too, it’s been there for months” (TF13)

Participants were recruited as part of a larger project on autistic adults’ social media usage, conducted in collaboration with the UK charity Autistica. Recruitment strategies and consent forms were co-designed with autistic adults, and the project is overseen by a lay advisory board of autistic members in addition to a scientific advisory board. Ethical approval was granted by Queen Mary University of London’s University Research Ethics committee (ref 2020/58). Participants were selected with the aim of achieving representation across socio-economic and gender categories, as well as social media usage. Out of 31 Twitter users in the study, 25 used Replies. Two participants’ Reply Threads were not collected as they were no longer available, leaving 23 participants. Table 1 provides an overview of the participants, their gender, Twitter usage, and age.

For the larger project, each participant’s Twitter activity was observed from March 2021 to May 2021, and all tweets were collected using R. In total, we collected 25,516 tweets. Some participants tweeted significantly more than others and would have skewed the findings without weightings. Weighted by participants’ usage, Quote tweets – retweets where the retweeters added text – were used least at 4% per participant on average, followed by standard broadcast tweets, which display on someone’s home timeline, which on average accounted for 17% of a participant’s tweets. Retweets accounted for 31%, and replies were used most at 41% (Koteyko et al., 2022). Replies were therefore used with a much higher relative frequency than in, for instance, Page’s (2012) study, which found Replies and Broadcast tweets used with similar frequencies (42% and 48%, respectively).

In order to analyse functions of Replies, full threads were examined. Threads were collected manually, as the Twitter API does not allow for automated thread collection. Each participant’s Replies were randomised, and the threads of the first 10 Replies per participant were screenshotted. Where threads were no longer available, for instance because tweets had been deleted, the next randomised Reply was used. A total of 192 threads were collected (some participants engaged in fewer than 10 threads over the observation period) containing 256 tweets produced by study participants. Topics discussed in the threads were diverse, with 74 different topics in total. Some of the most common topics discussed in more than 10 separate tweets and by at least two participants were: Autism (39 tweets), Politics (23 tweets), Art (16 tweets), Education (16 tweets), and Gender (11 tweets).

As we screenshotted full threads to include context in the analysis of our participants’ Replies, tweets of non-participants were also captured. However, as these Twitter users did not consent to participate in our study, these tweets were not analysed and are not presented in this article. Instead, where relevant, their tweets have been paraphrased.

Following the procedure described in the previous section, we first applied the CMC Act Taxonomy (Herring et al., 2005), where we coded for the primary CMC Act expressed in each C-Unit, followed by the interactional function coding of the three most common CMC Acts: Inform, Claim, Elaborate. While the interactional functions are not mutually exclusive, we coded only for the primary interactional function based on the context of the thread. The coding included multimodal features such as emojis and hashtags, which were coded together with the corresponding C-Unit when they repeated the content or sentiment from the C-Unit and were coded as a separate C-Unit when they presented new content. Only nine photographs were present in the data. In six instances, the photograph was the entire Reply, which was coded as a C-Unit. The other three instances were coded together with the preceding C-Unit.

The coding was conducted by an autistic researcher (Author1) using NVivo 12 and was performed twice for both the CMC acts and the interactional functions. Author2 acted as a moderator between each coding round, when both authors discussed codes that Author1 was unsure about in the first coding round. Following the discussion, Author1 coded the data again without referring to the codes from round 1. For each coding type (CMC act and interactional functions), the two rounds of coding (first round by only Author1 and second round by Author1 after moderation with Author2) were checked against each other for agreement to ensure intra-coder reliability. The CMC act coding showed consistency between the two rounds of coding, with an 80% agreement rate. The interactional function coding was also consistent, with an agreement rate of 83%. Inconsistencies were resolved in discussion between the two authors.

The 192 collected threads consisted of 256 tweets containing 596 C-Units. To answer the first part of RQ1: What do autistic adults use the Twitter Reply function for?, we analysed all C-Units for the CMC Acts they contained. The most common acts were (1) Claim (25%), (2) Inform (17%), and (3) Elaborate (13%); see Table 2.

As autism was the most common topic in our dataset, with 39 tweets mentioning autism or a related term, we separate our findings by topic: autism and other topics. While the prevalence of autism as a topic may be skewed due to our participant sample (autistic people who are keen to participate in autism research), it provides insight into how autistic people use Twitter to discuss their neurotype. In threads about autism, 64% of interactants (Twitter users to whom our participants replied) publicly identified as autistic, through hashtags like #ActuallyAutistic or identification in profiles, compared to 10% of interactants in other topic threads. This difference demonstrates that autism is a key topic on autistic Twitter, similar to its previously identified prevalence in online autistic communities (Davidson, 2008; Parsloe, 2015). By taking into consideration this difference in interactants between autism and other topics, our analysis provides insight into autistic adults’ discursive practices online.

Table 2 provides an overview of the CMC acts and their frequencies in the dataset.

The two most common categories, Inform and Claim, mirror findings from existing research (Barlow, 2019; Ishizaki et al., 2013; March, 2020; Nemer, 2016). However, the third most common category, Elaborate, is generally not found as a key category in written SNS communication (Barlow, 2019; Ishizaki et al., 2013; March, 2020; Nemer, 2016). While further research with large and representative samples is needed to understand the relationship between demographic characteristics and SNS proficiency, our pilot study comparing the percentage of participants per group (gender and frequency of SNS use) with the percentage of Elaborations for that group suggests a potential impact of SNS experience, while gender did not show any differences in use. Participants who used Twitter daily used fewer Elaborations than expected (Elaborations = 31%, participants = 43%), while participants who used Twitter monthly or less often employed more Elaborations than expected (Elaborations = 35%, participants = 22%). It is possible that daily Twitter users may have adjusted to the brevity expectations of the platform, while monthly users with less experience may be more likely to adhere to standard written communication practices.

Inform, Claim, and Elaborate are also categories which do not consider the C-Unit’s interactional function (see the previous section). Thus, each category was coded using the interactional function codes: Answer, Attitude, Explain, Humour, Information, Qualify, and Small Stories. Table 3 shows the frequencies of these functions, with Attitude and Small Stories being used most overall.

The effect of the topic of the tweet on both the CMC Act and the interactional function was checked using an Independent-Samples Mann-Whitney U Test in SPSS with the significance set at alpha = .05. While the difference between the autism and non-autism tweets in the CMC Acts was not statistically significant (alpha = .114), the topic of the tweets did have a statistically significant effect on the interactional function categories (alpha = .000). This seems to have specifically been caused by Answers for autism topics and Small Stories for other topics, which showed the most difference. Our qualitative analysis (below) also shows that tweets about autism performed different interactional functions than tweets about other topics.

We also identified what interactional functions occur most commonly for each of the three CMC Acts that we analysed further, as shown in Table 4.

Not unsurprisingly, Attitude was the most common function expressed by Claim, Explain was most common for Elaborate, and Small Stories was most common for Inform, followed by Answer. However, the functions are spread across the three acts (excepting Explain and Information), demonstrating that the three CMC Acts can fulfil different interactional functions.

Finally, as autistic individuals are known to follow patterns and schemas (Maciejewska, 2019), we analysed the usage of the categories per participant. The CMC Act coding did not show any one category being used more than half of the time, with only two participants with more than one Reply using one category (React – TF1, Claim – TF17) in 50% of instances. However, the interactional function coding showed category preferences, with 17 out of 23 participants using one category at least 50% of the time, although this could be explained by the fact that there are fewer interactional functions (7) than CMC Acts (17). The three categories that were most used were:

Answer – T4 (67%), TF18 (67%), TF19 (73%), TF26 (57%)

Attitude – T5 (58%), TF1 (80%), TF2 (54%), TF8 (68%), TF9 (50%), TF17 (75%), TF20 (67%), TF21 (60%), TF24 (56%)

Small Stories – T1 (74%), TF4 (58%), TF11 (50%), TF13 (62%)

In this section and the following discussion, the codes are expressed as CMC act [interactional function], for example: Claim [Attitude].

In our quantitative findings we found that participants used more Elaborate than was found in previous analyses of CMC Acts in written SNS interactions (Barlow, 2019; Ishizaki et al., 2013; March, 2020; Nemer, 2016), regardless of what topic participants were discussing. Therefore, we explored the use of Elaborate by our participants in more detail with the aim of investigating whether autistic preferences may be fulfilled through Elaborate (RQ2). Elaborate was primarily used to add Explanations or Attitude to preceding C-Units. It mostly followed Claim (in 21 instances), followed by Inform (16 instances), and Reject (16 instances).

[1] Beautiful painting would never have thought it was Kew Gardens the colours aren't quite what we see in England not that that deters from such a great painting (TF1)

In example 1, participant TF1 replied to a picture of a painting. He responded with a Claim [Attitude]: “Beautiful painting,” followed by another Claim [Attitude]: “would never have thought it was Kew Gardens,” which is then followed by an Elaborate [Explanation]: “the colours aren’t quite what we see in England.” Elaborate provides an explanation for TF1’s Claim that he “would have never thought it was Kew Gardens.” The explanation provides explicitation (Belek, 2018) for his Claim, clarifying that his surprise is not based on a negative evaluation of the painting but on a comparison between the colours in the painting and the colours commonly found in England. The explicitness manages the audience reception (Bondi, 2018) and can also be tied to the autistic characteristic to provide more information than expected by neurotypical norms (Belek, 2018; Surian et al., 1996). As in this example, Elaborate is used commonly as a way of preventing potential miscommunication throughout the data, most noticeably when used in combination with Reject.

[2] I'm #ActuallyAutistic and I loved my time at school - at least, the structure and routine. We are not all the same. The silent corridors and well-behaved kids on here sound good to me (TF5)

TF5 responded to a tweet which referenced a new Education policy focused on “managing behaviour.” The initiating tweet quotes from a government website about the policy’s focus on correcting behaviour, and the initiator negatively evaluates the policy with regard to neurodivergent children. TF5’s response begins with an Inform [Qualify]: “I’m #ActuallyAutistic,” followed by a Reject: “and I loved my time at school – at least, the structure and routine,” a Repair: “we are not all the same,” as TF5 is trying to correct the assumption in the preceding tweet that all autistic children have the same experiences and needs, and finally Elaborate [Attitude], on the Repair: “The silent corridors and well-behaved kids on here sound good to me.” The opening Qualification is an instant appeal to authority based on TF5’s personal experience as an autistic child in school. TF5 disagrees with the claim made in the preceding tweet about all neurodivergent children struggling with strict school policies, which they first Reject and then Repair to educate the initial tweeter, followed by a reinforcing elaboration at the end. The Elaborate also provides specificity for the Rejection (“the silent corridors and well-behaved children”) and highlights TF5’s personal opinion (“sounds good to me”), mitigating the face threat of a direct rejection and pre-empting potential conflict.

The additional information in Elaborate appears aimed at preventing miscommunications and conflicts and clarifying the tone of preceding C-Units. Autistic adults may struggle to understand others and make themselves understood (Bottema-Beutel & Frisch, 2020), particularly in interactions with allistic people (Crompton et al., 2020). The tendency to provide more information than expected has been viewed as an autistic characteristic with no indication of any specific function (Surian et al., 1996). However, similar to how repetition has been shown to fulfil functions in autistic conversation (Maciejewska, 2019), additional information also has a clear interactional purpose in our findings: to prevent potential conflicts and provide further context to ensure understanding.

In addition to Elaborate, we identified other features that demonstrated that participants take up offline autistic communicative competencies and preferences in Twitter Replies (RQ2). Notably, the data contain repetition and reliance on questions and answers, regardless of the topic of discussion. While autistic conversation can also be characterised by topic shifts, there were no instances of that in our data. This is not unexpected, as Sterponi and Shankey (2014) suggest that topic shifts may be used to avoid uncomfortable situations, which on social media can be avoided through disengagement. There was only one instance of direct repetition, which was again unsurprising, as echolalia is generally a spoken feature. We did identify heavy use of questions and answers, however, as well as repetition of functions by individual participants. Our participants used the Claim, Inform, and Elaborate categories as Answers in 73 instances. These answers range from one word (example 3) to multiple sentences (example 4).

[3] Dumbledore ❤️ (TF19)

[4] Ours can't tell that a food will be spicy until after they've tasted it, but then they shake their heads, go "pa-pa-pa" and run away, rather than eat more. (TF12)

In example 3, participant TF19 responds to a question regarding Harry Potter characters, while participant TF12 in example 4 responds to a question about cats. Participants responded to a range of questions both from accounts with a small number of followers (example 4) and larger accounts (example 3). By responding to questions, autistic adults can use social media in a way that requires no initiation and where the expected response is scaffolded from the initiating tweet – “whatever answers do, they must do this with something already begun” (Goffman, 1976, p. 257). In addition, the questions can function as a type of ice breaker, which Fox et al. (2019) found to be positive for neurodivergent SNS users.

Finally, individual participants appeared to have preferences for interactional functions, as demonstrated earlier. These preferences can be linked to autistic preferences for routine (Maciejewska, 2019), as autistic people like the familiarity of previously discussed topics, routines, and structures. Most notable is participant T4, who responds to tweets with the same links and, at times, the same sentences.

[5] Without my autistic brain I wouldn't be me, so it does define me. The assumption that our inherent selves are not worth being “defined” by, is part of the stigma associated with our condition. (T4)

[6] What's the problem with being “defined” by our autism? Without my autistic brain I wouldn't be me. (T4)

In both examples, participant T4 repeats the same phrase: “without my autistic brain I wouldn’t be me.” Although responding to tweets by different people on different occasions, she uses the exact same sentence to do so, demonstrating the preference for familiar schema (Maciejewska, 2019).

Finally, we compared Replies about autism to Replies about other topics with the aim of answering RQ3: Do autistic adults use Replies differently when discussing autism versus other topics? When discussing autism and interacting with a majority autistic community, our participants seem to resort to autistic conversational features, shown in the differences in CMC acts and interactional functions between autism and other topics in Tables 2 and 3. In CMC act coding, we found that Claim was used 8% less when discussing autism compared with other topics. This could potentially be explained by the slight increase in Inform and Elaborate in autism-related Replies, where participants expressed fewer opinions and made more informative statements. The 8% difference is relatively small, however, and not statistically significant. In the interactional function coding, in contrast, more extensive differences were identified between autism and other topics. Small Stories, Explain, and Humour appeared more than twice as often in Replies about other topics, while Answer and Qualify appeared more than twice as often in Replies about autism (see Table 3). However, Qualify and Humour appear in small numbers in the dataset, and few conclusions can be drawn based on the eight and six instances, respectively, that appear in our data. The remaining categories provide more data to allow further analysis: Answer appears 73 times, Explain appears 27 times, and Small Stories appears 88 times.

Answer is the only category of these three that appears relatively more often in autism-related threads. An example of these interactions are Replies to an autistic “chat moderator” account, which posts questions about autism and autistic experiences that others can respond to, for instance about experiences with sensory overload and support animals. Considering autistic preference for direct and clear interactions, and a reliance on questions and answers (Dobbinson et al., 1998; Maciejewska, 2019), this finding shows not only that our participants use more Answers in inter-autistic communication, but also that there are Twitter accounts set up to accommodate this preference.

In contrast, Small Stories and Explain were more common in Replies about other topics than in Replies about autism. Explain is found primarily in threads where interactants are having a discussion.

[7] This is misleading as when people do chargebacks they get banned from the company they did the chargeback with. (TF24)

In example 7, TF24 begins his reply with a Reject, “this is misleading,” and continues with an Elaborate [Explain]. While a simple disagreement would be possible, the participant includes an explanation for why they find the initiating tweet misleading. The additional information clarifies the initial Reject, preventing potential misinterpretation, which can be common in tweets due to the character limit (Scott, 2015). Beyond the limited nature of Twitter, autistic people may be used to relying on additional information in order to interpret or understand situations (Constant et al., 2020), and they may also apply this strategy in conversations they initiate and in their own conversational turns.

Small Stories are used to add details and to connect through co-tellings, often extending across multiple C-Units and tweets, as in example 8, which consists of two Reply tweets by participant T1 in the same thread. The Reply appearing in between has been paraphrased, as it belongs to a non-participant:

[8] My own dog was brought up on a housing building site so he pays no attention to loud noises but he loves to greet anyone wearing hi-vis jackets and hard hats [laughing emoji] (T1)

hi-vis… people run on our road… he barks at them (Non-Participant)

All the builders used to make a fuss of [dog name] when he was a puppy so now he drags us towards anyone in hi-vis thinking that they will do the same! (T1)

T1 first responds to an initiating tweet about a dog barking at construction noises. The story extends across four tweets as it is co-constructed with the initiating tweet and the intermediate Reply by building on the created world (Dayter, 2015). The story contains five C-Units, including multiple descriptions of what the dog does: “pays no attention to loud noises,” “loves to greet anyone…,” “drags us towards anyone…,” and why it does so: “brought up on a housing building site,” “the builders used to make a fuss…” Similar to Explain and Elaborate, the additional information provides clarity to the reader, yet the Small Stories go beyond Explain to focus on the telling of an event. The additional details emphasise the event’s tellability through the use of internal evaluations like “fuss,” “drags,” and the exclamation point (Dayter, 2015). In addition, the co-telling goes against autistic storytelling preferences, which generally present stories as “self-contained anecdotes” (Fasulo, 2019, p. 621) with no room for co-narrators.

These findings suggest that interactions where autistic users do not talk about autism are characterised by increased information provision through Explain and Small Stories compared to autism-related interactions. Elaborations and explanations potentially aim to avoid misunderstandings which are common between neurotypes (Catala, 2021; Crompton et al., 2020). In contrast, interactions about autism took place primarily within the autistic Twitter community (64% of interactants publicly identified as autistic) and were characterised by questions-and-answers and less emotive language, as evident for example in the lesser use of Small Stories, which are generally more emotive, in inter-autistic communication. The use of Answers suggests a community set-up that is predisposed to autistic preferences. These findings support previous findings by Crompton et al. (2020) that communication between people of the same neurotype is generally more successful and requires less adaptation, as neither neurotype needs to adjust their preferences and expectations, whereas in inter-neurotype communication autistic participants converge to perceived neurotypical preferences and expectations.

Drawing on a mixed-methods approach, we have shown how the analysis of interactional functions can complement analysis of CMC acts, while also providing insights into the characteristics of autistic communication and interactions around autism compared to other topics. We have demonstrated that interactional functions identified in previous research on social media interactions map onto autistic online conversations, and that this approach allows us to investigate what makes autistic conversation unique, while at the same time showing the similarities between autistic and neurotypical interactions. In addition, our analysis shows that medium factors can facilitate autistic preferences, for instance through the availability of the Reply function, demonstrating the importance of considering medium factors for inclusive SNS use.

By complementing the CMC taxonomy with interactional functions, we have been able to identify autistic users’ strategies that respond to interactional demands in the CMC context. The analysis of CMC Acts has shown the predominance of Claim, Inform, and Elaborate, with the prominence of Elaborate likely to be characteristic of autistic users and their Twitter Replies. This characteristic can be linked to autistic strengths around clarity in conversations (Belek, 2018), as opposed to viewing autistic conversational features as misreading conversational expectations (Surian et al., 1996). We suggest that it is a strategy to prevent misunderstandings and miscommunication, which are common in autistic-allistic interactions (Bottema-Beutel & Frisch, 2020; Crompton et al., 2020). We have also shown how characteristics of offline autistic conversations, including repetition and formulaic language (Maciejewska, 2020), are present in Twitter Replies through the repetition of content and sentences, as well as in individual preferences for interactional functions. Additionally, autistic adults provide more Small Stories and Explanations when discussing non-autism topics, while they interact with other autistic Twitter users through question-and-answer sequences. This suggests that autistic adults share an understanding not just of offline autistic conversations (Crompton et al., 2020) but also of online autistic conversations where autistic-to-autistic interactions can be successful with less information and coordination, as autistic interlocuters have a “low demand for coordination” (Heasman & Gillespie, 2019, p. 916).

On the methodological level, we have shown how the analysis of interactional functions in conjunction with CMC acts provides a more detailed understanding of interactions taking place on social media platforms. Two of the most common acts, Inform and Claim, are defined by their verifiability (Herring et al., 2005), which leaves room for further analysis of interactional functionality. More generally, the CMC Act focus is on the lexico-grammatical level of analysis, while potential autistic uses of language can only be investigated by analysing the interactional functions. By applying our interactional functions, future research can use the CMC Act taxonomy to investigate both CMC Act functions and interactional effects. Our findings show that each of the three acts we analysed – Inform, Claim, and Elaborate – are used for a variety of functions, meaning that the interactional function coding provides additional information on these acts. Future research will need to consider whether the functionality analysis can be applied to other categories of the CMC Act taxonomy, as we have only applied it to our data’s most common categories. The functionality categories may also need to be expanded to accommodate the features and content of other platforms and other users, as our research focus was on autistic Twitter Replies only.

The empirical findings presented here support going beyond considerations of psychosocial outcomes for autistic SNS users, to ask linguistically informed questions about why autistic people interact on SNS in the ways that they do, and about the role of socially and technologically configured expectations in their interactions. By demonstrating the interrelation between platform architecture and discursive practices of autistic users (e.g., the prevalence of Answers in autistic chat) our approach speaks directly to the pioneering agenda of Ochs and Solomon (2010), who argued that autistic sociality “is not an oxymoron” and should be examined with reference to “sociocultural ecologies that demonstrably promote or impede its development” (2010, p. 69). The application of our framework to other SNS popular with autistic people can reveal a full set of interactional functions important for adapting digital networking technologies for autistic use. In addition to the focus on user actions, users' beliefs about each specific SNS platform and the social networks they may imagine they are part of (Tagg & Seargeant, 2016) are equally key for understanding autistic sociality online and should be analysed through interviews and collaborative workshops as part of participatory research design (Fletcher-Watson et al., 2018).

We would like to thank our participants, our lay Advisory Board, including Sofia Billan, our scientific advisory board, Autistica, and Autistic Nottingham for collaboration and support. The project was funded by the ESRC (grant number ES/T016507/1).

Austin, J. L. (1962). How to do things with words: The William James Lectures delivered at Harvard University in 1955. Oxford University Press. https://doi.org/10.1093/acprof:oso/9780198245537.001.0001

Autistica (2021). What is autism. Retrieved November 29, 2021 from https://www.autistica.org.uk/what-is-autism/what-is-autism

Bach, K., & Harnish, M. (1979). Linguistic communication and speech acts. MIT Press.

Bagatell, N. (2010). From cure to community: Transforming notions of autism. Journal of the Society for Psychological Anthropology, 38(1), 33-55. 

Bamberg, M., & Georgakopoulou, A. (2008). Small stories as a new perspective in narrative and identity analysis. Text&Talk, 28(3), 377-396.

Barlow, J. B. (2019). Emergent roles in computer-mediated synchronous virtual groups. In Proceedings of the 52^nd Hawaii International Conference on System Sciences (pp. 759-768). IEEE Press.

Baron-Cohen, S. (2001). Theory of mind in normal development and autism. Prisme, 34, 174–183.

Belek, B. (2018). Autism and the proficiency of social ineptitude: Probing the rules of “appropriate” behavior. Ethos, 46(2), 161-179.

Bondi, M. (2018). Try to prove me wrong: Dialogicity and audience involvement in economics blogs. Discourse, Context & Media, 24, 33–42. https://doi.org/10.1016/j.dcm.2018.04.011

Bottema-Beutel, K. (2017). Glimpses into the blind spot: Social interaction and autism. Journal of Communication Disorders, 68, 24–34.

Bottema-Beutel, K., & Frisch, M. K. (2020). Understanding and addressing social communication difficulties in children with autism. In G. Vivanti, K. Bottema-Beutel, & L. Turner-Brown (Eds.), Clinical guide to early interventions for children with autism, best practices in child and adolescent behavioral health care (pp. 41-60). Springer.

Brownlow, C., Bertilsdotter Rosqvist, H., & O’Dell, L. (2015). Exploring the potential for social networking among people with autism: Challenging dominant ideas of “friendship.” Scandinavian Journal of Disability Research, 17(2), 188–193.

Carr, C. T., Schrock, D. B., & Dauterman, P. (2012). Speech acts within Facebook status messages. Journal of Language and Social Psychology, 31(2), 176–196. https://doi.org/10.1177/0261927X12438535

Catala, A., Faucher, L., & Poirier, P. (2021). Autism, epistemic injustice, and epistemic disablement: A relational account of epistemic agency. Synthese, 199, 9013-9039. https://doi.org/10.1007/s11229-021-03192-7

Constant, A., Bervoets, J., Hens, K., & Van de Cruys, S. (2020). Precise worlds for certain minds: An ecological perspective on the relational self in autism. Topoi, 39(3), 611–622. https://doi.org/10.1007/s11245-018-9546-4

Cook, J., Crane, L., Bourne, L., Hull, L., & Mandy, W. (2021). Camouflaging in an everyday social context: An interpersonal recall study. Autism, 25(5), 1444-1456. https://doi.org/10.1177/1362361321992641

Costa, E. (2018). Affordances-in-practice: An ethnographic critique of social media logic and context collapse. New Media & Society, 20(10), 3641-3656. 

Crompton, C. J., Ropar, D., Evans-Williams, C. V., & Flynn, E. G., & Fletcher-Watson, S. (2020). Autistic peer-to-peer information transfer is highly effective. Autism 24, 1704–1712. https://doi.org/10.1177/1362361320919286

Davidson, J. (2008). Autistic culture online: Virtual communication and cultural expression on the spectrum. Social & Cultural Geography, 9(7), 791-806.

Dayter, D. (2015). Small stories and extended narratives on Twitter. Discourse, Context & Media, 10, 19–26.

Dayter, D. (2016). Discursive self in microblogging. John Benjamins.

DHSC/Department of Health and Social Care and NHS England (2020). Autism waiting time statistics. Retrieved September 12, 2021 from https://digital.nhs.uk/data-and-information/publications/statistical/autism-statistics/q1-april-to-june-2020-21/data-quality-copy

Dobbinson, S., Perkins, M. R., & Boucher, J. (1998). Structural patterns in conversations with a woman who has autism. Journal of Communication Disorders, 31(2), 113–134. https://doi.org/10.1016/S0021-9924(97)00085-3

Dresner, E., & Herring, S. C. (2010). Functions of the nonverbal in CMC: Emoticons and illocutionary force. Communication Theory, 20(3), 249–268. https://doi.org/10.1111/j.1468-2885.2010.01362.x

Fasulo, A. (2021). Critical perspectives in clinical psychology: Autistic identities. In M. Bamberg, C. Demuth, & M. Watzlawik, M. (Eds.), The Cambridge handbook of identity (pp. 463-486). Cambridge University Press.

Fletcher-Watson, S., Adams, J., Brook, K., Charman, T., Crane, L., Cusack, J., Leekam, S., Milton, D., Parr, J. R., & Pellicano, E. (2018). Making the future together: Shaping autism research through meaningful participation. Autism, 23(4), 943-953.

Fox, E., Baden, S., Greene, J., Ziegler, N., & Sharmin, M. (2019). Connection: Assisting neurodiverse individuals in forming lasting relationships through a digital medium. In Proceedings of the 21st International ACM SIGACCESS Conference on Computers and Accessibility, ASSETS ’19 (pp. 633–635). https://doi.org/10.1145/3308561.3354641

Francis, G., & Hunston S. (1992). Analysing everyday conversation. In M. Coulthard (Ed.), Advances in spoken discourse analysis (pp. 123-161). Routledge.

Ge, J., & Herring, S. C. (2018). Communicative functions of emoji sequences on Sina Weibo. First Monday, 23(11). https://firstmonday.org/ojs/index.php/fm/article/view/9413

Georgakopoulou, A. (2007). Small stories, interaction and identities. John Benjamins.

Ge-Stadnyk, J. (2021). Communicative functions of emoji sequences in the context of self-presentation: A comparative study of Weibo and Twitter users. Discourse & Communication, 15(4), 369–387. https://doi.org/10.1177/17504813211002038

Gillespie-Lynch, K., Kapp, S. K., Shane-Simpson, C., Smith, D. S., & Hutman, T. (2014). Intersections between the autism spectrum and the internet: Perceived benefits and preferred functions of computer-mediated communication. Intellectual and Development Disabilities, 52(6), 456–469. https://doi.org/10.1352/1934-9556-52.6.456

Goffman, E. (1971). Relations in public: Microstudies of the public order. Basic Books.

Goffman, E. (1976). Replies and responses. Language in Society, 5(3), 257-313.

Heasman, B., & Gillespie, A. (2019). Neurodivergent intersubjectivity: Distinctive features of how autistic people create shared understanding. Autism, 23(4), 910–921.

Herring, S. C. (2004). Computer-mediated discourse analysis: An approach to researching online communities. In S. A. Barab, R. Kling, & J. H. Gray (Eds.). Designing for virtual communities in the service of learning (pp. 316-338). Cambridge University Press.

Herring, S. C. (2007). A faceted classification scheme for computer-mediated discourse. Language@Internet, 4. https://www.languageatinternet.org/articles/2007/761

Herring, S. C., Das, A., & Penymarthy, S. (2005). CMC Act Taxonomy. Retrieved November 15, 2021 from http://ella.slis.indiana.edu/~herring/cmc.acts.html

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, 2519-2534.

Hutchby, I. (2001). Technologies, texts and affordances. Sociology, 35(2), 441-456.

Idriss, C. R. (2021). Invisible autistic infrastructure: Ethnographic reflections on an autistic community. Medical Anthropology, 40(2), 129-140.

Ishizaki, H., Herring, S., Hattori, G., Ono, C., & Takishima, Y. (2013). A computer-mediated discourse analysis of user commenting behavior on an online music distribution site. In Proceedings of Forum on Information Technology, 12, 47-52. 

Koteyko, N., Van Driel, M., & Vines, J. (2022). Autistic sociality on Twitter: Enacted affordances and affiliation strategies. Discourse & Communication, 16(4), 385-402.

Lawson, W. B. (2020). Adaptive morphing and coping with social threat in autism: An autistic perspective. Journal of Intellectual Disability – Diagnosis and Treatment, 8, 519-526.

Maciejewska, E. (2019). Discourse analysis as a tool for uncovering strengths in communicative practices of autistic individuals. Discourse Studies, 21(3), 300–316.

Maciejewska, E. (2020). Autistic resources from a discourse-analytic perspective. Qualitative Psychology, 7(3), 348–366. https://doi.org/10.1037/qup0000167

March, L. (2020). Rebel reviewers: Social media review pages as sites of Confederate memorial discourses. First Monday, 25(12).

Martin, J. R., & White, P. (2005). The language of evaluation - appraisal in English. Palgrave Macmillan.

Martinec, R., & Salway, A. (2005). A system for image–text relations in new (and old) media. Visual Communication, 4(3), 337–371. https://doi.org/10.1177/1470357205055928

Mazurek, M. O. (2013). Social media use among adults with autism spectrum disorders. Computers in Human Behavior, 29, 1709–1714.

McGhee Hassrick, E., Graham Holmes, L., Sosnowy, C., Walton, J., & Carley, K. (2021). Benefits and risks: A systematic review of information and communication technology use by autistic people. Autism in Adulthood, 3(1), 72-84.

Nastri, J., Peña, J., & Hancock, J. T. (2006). The construction of away messages: A speech act analysis. Journal of Computer-Mediated Communication, 11(4), 1025–1045. https://doi.org/10.1111/j.1083-6101.2006.00306.x

National Autistic Society (2021). What is autism. Retrieved from https://www.autism.org.uk/advice-and-guidance/what-is-autism (accessed 1.27.21).

Nemer, D. (2016). Celebrities acting up: A speech act analysis in tweets of famous people. Social Networking, 5(1). https://doi.org/10.4236/sn.2016.51001

Ochs, E., & Solomon, O. (2010). Autistic sociality. Ethos, 38(1), 69-92.

Orgad, S. (2009). How can researchers make sense of issues involved in collecting and interpreting online and offline data? In A. Markham & N. Baym (Eds.), Internet Inquiry: Conversations about method (pp. 33-53). Sage.

Page, R. (2012). The linguistics of self-branding and micro-celebrity in Twitter: The role of hashtags. Discourse & Communication, 6(2), 181–201. https://doi.org/10.1177/1750481312437441

Parsloe, S. M. (2015). Discourses of disability, narratives of community: reclaiming an autistic identity online. Journal of Applied Communication Research, 43(3), 336-356.

Scott, K. (2015). The pragmatics of hashtags: Inference and conversational style on Twitter. Journal of Pragmatics, 81, 8–20.

Searle, J. R. (1979). Expression and meaning: Studies in the theory of speech acts. Cambridge University Press.

Sinclair, J., & Coulthard, M. (1992). Towards an analysis of discourse. In M. Coulthard (Ed.), Advances in spoken discourse analysis (pp. 1-34). Routledge.

Soulier, L., Tamine, L., & Nguyen, G.-H. (2016). Answering Twitter questions: A model for recommending answerers through social collaboration. In Proceedings of the 25th ACM International on Conference on Information and Knowledge Management (pp. 267–276). https://doi.org/10.1145/2983323.2983771#

SPAAC (Speech-Act Annotation Scheme) (2003). A Project of the University of Lancaster, University Centre for Computer Corpus Research on Language. Retrieved October 25, 2021 from http://ucrel.lancs.ac.uk/SPAAC/SPAAC%20Annotation%20Scheme1.pdf 

Sterponi, L., de Kirby, K., & Shankey, J. (2015). Rethinking language in autism. Autism, 19(5), 517–526.

Sterponi, L., & Shankey, J. (2014). Rethinking echolalia: Repetition as interactional resource in the communication of a child with autism. Journal of Child Language, 41(2), 275–304. https://doi.org/10.1017/S0305000912000682

Surian, L., Baron-Cohen, S., & Van Der Lely, H. (1996). Are children with autism deaf to Gricean maxims? Cognitive Neuropsychology, 1(1), 55-71.

Tagg, C., & Seargeant, P. (2016). Negotiating social roles in semi-public online contexts. In S. Leppänen, E. Westinen, & S. Kytölä (Eds.), Social media discourse, (dis)identifications and diversities (pp. 211-234). Routledge.

Williams, G., Wharton, T., & Jagoe, C. (2021). Mutual (mis)understanding: Reframing autistic pragmatic “impairments” using Relevance Theory. Frontiers in Psychology, 12, article 616664.

Zappavigna, M. (2012). Discourse of Twitter and social media. Continuum.

Zappavigna, M. (2018). Searchable talk. Bloomsbury Academic.

Zeidan, J., Fombonne, E., Scorah, J., Ibrahim, A., Durkin, M. S., Saxena, S., Yusuf, A., Shih, A., & Elsabbagh, M. (2022). Global prevalence of autism: A systematic review update. Autism Research, 15(5), 778-790.

Martine van Driel [m.a.vandriel.1@bham.ac.uk] is Assistant Professor of English Language and Communication at the University of Birmingham, UK. Her current research interests include autistic communication, discourse-based approaches to social media, and participatory research methods in linguistics.

Nelya Koteyko [n.koteyko@qmul.ac.uk] is Professor of Language and Communication at Queen Mary University of London. Her current research interests include discourse-based approaches to autistic communication, participatory research methodologies in linguistics, and identity in computer-mediated interaction.