Exploring Autism
I wrote this Literature Review for my principles and practices class at the University of Rochester. The professor asked us to pick a topic of our choice and synthesize existing literature in a narrative form to create a new argument. I chose to explore autism, through the perspectives of theory of mind, language and emerging technologies in assessment and intervention. I have included my Literature Review below.
Exploring Autism Spectrum Disorder: Theory of Mind, Language, and Emerging Technologies in Assessment and Intervention
Abstract
Autism Spectrum Disorder (ASD) is a neurodevelopmental condition characterized by deficits in social communication, social interaction, and restricted, repetitive behaviors. Central to ASD are deficits in Theory of Mind (ToM) and Language. Language impairments in ASD vary widely, from pragmatic challenges to structural deficits. Assessment of ASD remains a critical yet challenging process, with tools such as ADI-R and ADOS being time-consuming and expensive, highlighting the need for innovation. Emerging technologies, including artificial intelligence (AI) and machine learning (ML), may enhance ASD diagnosis and screening. Interventions for ASD, particularly those targeting ToM and language development, show promise but often fail to generalize to real-world contexts. Technology-based interventions may be able to assist with the scalability and generalizations of these interventions. Recent technological advances, including the use of Large Language Models (LLMs), may provide new opportunities for improving assessment and intervention, particularly in areas like social skills training and language development. This paper reviews the literature on ToM, Language, Assessment, and Intervention for ASD and considers the potential for new technologies to transform the landscape of ASD care.
Keywords: Autism Spectrum Disorder, Theory of Mind, Language, Technology, Assessment, Intervention, Artificial Intelligence, Large Language Models
Exploring Autism Spectrum Disorder: Theory of Mind, Language, and Emerging Technologies in Assessment and Intervention
Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder, with symptoms often appearing between the ages of 0 to 24 months; ASD has a population prevalence between 1-2% for both children and adults (American Psychiatric Association, 2022). Two identifying characteristics of ASD are deficits in socializing, both social communication and social interaction deficits and, secondly, repetitive and restricted interests and behaviors (American Psychiatric Association, 2022). According to Anagnostopoulou et al. (2020), while the average age of symptom onset in ASD is between 0 and 24 months, the average age of diagnosis is 4 in the USA. Many researchers agree that early diagnosis and intervention are crucial for positive outcomes (Dahiya et al., 2021).
Since landmark research done by Baron-Cohen et al. in 1985, researchers have hypothesized that Theory of Mind (ToM) plays an integral role in the development of ASD, specifically that children with ASD fail to develop the ability to imagine the beliefs of others easily. ToM deficit is complex in how it emerges developmentally and in how it is related to ASD development. However, one point of interest is how ToM development links to language development and then subsequent language impairments in individuals with ASD, which lead to the social communication deficits that ASD is defined by (Bamicha & Drigas, 2022).
Assessments and interventions are key parts of ASD treatment. Researchers are exploring how technology could aid those with ASD, specifically with technology-assisted screening and diagnosis (Dahiya et al., 2021) or technology-assisted interventions (Dyrda et al., 2020). This literature review seeks to collate appropriate research around ToM and language use in ASD and then specifically focus on the assessment and intervention of ASD and how researchers are considering new technologies. Ultimately, the paper ends with questions and ideas for future research on how AI, specifically Large Language Models (LLMs), could deliver superior assessment and intervention, especially around language use in those with ASD.
Theory of Mind in Autism Spectrum Disorder
A landmark experiment conducted by Baron-Cohen et al. in 1985 explored the relationship between ToM and Autism. Baron-Cohen begins by hypothesizing that children with ASD do not simply have intellectual impairments such as mental retardation; they lack metarepresentational development and ToM, which means they fail to impute beliefs into others. Baron-Cohen defined ASD as "a profound disorder in understanding and coping with the social environment, regardless of IQ" (Baron-Cohen, 1985, pg 37), with a primary feature as the ASD individual's inability to develop social relationships. Baron-Cohen suggested that individuals with ASD treat others like objects. Baron-Cohen states that "second-order representations" are a key part of ToM development, and these typically come about at around age 2, which leads to a child's ability to play. Baron-Cohen hypothesized that children with ASD lack ToM development, which leads to their inability to attribute mental states to others, which leads to an inability to play, and a lack of social interest, leading to a lack of social development, which is a hallmark of ASD.
Baron-Cohen proved this theory with his famous Sally-Anne Experiment. He studied 20 children with ASD, 14 children with Down syndrome (DS), and 27 typically developing (TD) children. Baron-Cohen conducted the experiment with two dolls, Sally and Anne, each with a basket. The children were shown Sally putting a marble in her basket, then leaving the scene, at which time researchers transferred the marble to Anne's basket, within the view of the child, and the child was asked, when Sally returns, "Where will Sally look for her marble?" (Baron-Cohen, 1985, pg 41). Autistic children continually pointed to the basket where the marble was, not where Sally would have thought it was. Baron-Cohen concluded that autistic children did not understand the difference between their own and the doll's knowledge. Thus, ASD children lacked second-order representation and failed to employ the theory of mind, probably because they could not represent the mental states of others.
Bamicha and Drigas, 2022 studied the link between ToM and ASD further in their literature review and confirmed that social deficits in ASD may derive from ToM deficits. Bamicha and Drigas highlight the complexity of the relationship between ASD and ToM development. ToM development begins in TD children around two and continues until about 5. During this time, children begin to mimic others and begin to understand "false beliefs," which is an understanding that others can have beliefs different from their own. Bamicha and Drigras also highlight multiple theories and types of ToM, including an unconscious and automatic ToM, which develops early in life, and a deliberate and controlled ToM, which develops later; they also highlight the difference between cognitive and emotional ToM. Bamicha and Drigas highlight the importance of social development for knowledge acquisition; for example, ToM aids in observation, perception, attention, memory, cognitive flexibility, and inhibition control. Without ToM development, the ASD child lacks knowledge that they may have derived from social learning.
Bamicha and Drigas, 2022 state that a lack of ToM may also impact an individual's cognitive development, including the ability to cultivate self-awareness, higher-order thinking, and executive functioning. Individuals with ASD are challenged to predict others' behavior and have a tendency to interpret things literally. Individuals with ASD also find it hard to express emotions in language and may also be impaired in reading emotions in others, including facial expressions. Individuals with ASD have lower empathy, although Bamicha and Drigas note that there are different types of empathy, such as emotional empathy and cognitive empathy. Lack of emotional empathy may lead the ASD individual to create systems to manage information to compensate for the lack of empathy and to predict behavior better. ASD individuals may also obsess with structure as a coping mechanism.
Bamicha and Drigas, 2022 also highlight that language capacity in individuals with ASD is challenging in social contexts. Individuals with ASD may find it difficult to initiate or maintain a conversation. Individuals with ASD also tend to interpret language literally. Individuals with ASD also have a reduced ability to switch between contexts and have a reduced ability to inhibit information; instead of taking in only what is necessary from a context, they may focus on all the details. Individuals with ASD also have reductions in memory, specifically visual-spatial working memory and autobiographical memory.
Language Use in Autism Spectrum Disorder
Schaeffer et al. (2023) examine language use in those with ASD. They state that the number one reason why parents seek out an ASD diagnosis for their child is delayed language development. Their research links lack of language development to social impairments we find in those with ASD, although there is a wide array of language use in ASD individuals, varying from no impairments to high impairments. Schaeffer et al. (2023) tell us that to study language better, we need to break it down into separate domains, lexicon, structure, and pragmatics, and score each domain independently. They break down autism as it relates to language into three distinct profiles: ASD-LN, where structural language is normal yet pragmatics is impaired; ASD-LI, where structural language is impaired; and MV, where the individual has minimal verbal abilities.
Schaeffer et al. (2023) go on to describe how, from a pragmatic language point of view, individuals with ASD may still have good language skills as long as things are literal because ASD individuals derive meaning from the language, not social cues, but once communication uses idioms, metaphors or irony, ASD individuals face difficulties. ASD individuals may also have difficulties knowing which pronouns to use. Individuals with ASD also tend to tell stories differently, being more linear instead of hierarchical. Schaeffer et al. (2023) then look at structural language, including phonology, phonetics, syntax, semantics, and morphosyntax, stating that about fifty percent of ASD individuals have impairments here. Finally, Schaeffer et al. (2023) look at another portion of ASD individuals who remain minimally verbal, around 30%. Schaeffer et al. (2023) also confirm the link between ToM and Executive function (EF) in ASD.
Vogindroukas et al. 2022 also describe the diversity of language behavior in ASD, with ASD individuals experiencing impairments in four key areas: semantics, pragmatics, phonology, and morphosyntax. Vogindroukas et al. 2022 state that language development begins in infants at around nine months through joint attention development. An infant requires stable and available adults to practice interaction with for joint attention to develop correctly. Vogindroukas et al. 2022 describe that semantic difficulties can arise when individuals with ASD have a smaller operating vocabulary than TD. Individuals with ASD may use words but not understand their meaning. Individuals with ASD may also use nonsensical terms because they do not know how to convey the actual meaning they wish to express via language.
Vogindroukas et al. 2022 also state that pragmatic difficulties seen in individuals with ASD often arise from underlying ToM or EF deficits; this is why individuals with ASD may speak strangely, for example, not providing comments or not asking for information during conversation. Pragmatic impairment may also lead to echolalia, pedantic speech, and misunderstanding of figurative language. ASD individuals may also speak too softly, loudly, or emotionlessly, use too many formal words, or use too many unusual phrases, including ones they have created themselves. Once again, individuals with ASD may interpret others too literally. They may sound overly philosophical or pompous or give "hyperinformation" that is unnecessary.
Phonological difficulties may arise, such as those in sound created for language. Phonological processing requires that one is self-aware, and this self-awareness may be limited in ASD individuals due to ToM deficiency. Individuals with ASD also have difficulty with morphosyntax, that is, the grammar relationship between subject and object, as well as tense and pronouns. Individuals with ASD may also have problems with speech, for example, being too monotone, robotic, jerky, or sing-songy. They may also have slower speech rates and a greater range of tone.
Cardillo et al. (2020) also look closely at pragmatic language and how it is related to ToM and EF in those with ASD. They investigate two primary parts of pragmatic language: comprehension of non-literal language and the ability to make inferences. Cardillo et al. (2020) explain that language plays a key role in social relationships and that vocabulary and syntax are not enough; an individual needs context and the ability to interpret non-verbal or non-literal meanings, this being defined overall as pragmatic language ability. They state that individuals with ASD have problems with pragmatic language. They give examples, for example, that individuals with ASD have problems understanding poetry, non-literal language, irony, and figures of speech; they cannot interpret ambiguous meanings and may struggle with topic shifts and with taking another's perspective. Two main perspectives are that, one, ToM deficits cause pragmatic difficulties, and secondly, EF deficits cause pragmatic difficulties.
The Cardillo et al. (2020) study looked at 143 individuals, 73 with ASD and 70 controls. They tested pragmatic language via metaphors and inferences. ToM was tested verbally and via facial recognition tests; they tested EF by inhibition, task switching tasks, and working memory tests. The results indicated that groups with ASD were deficient in pragmatic language, although the severity of ASD does not correlate with pragmatic language ability, ToM, or EF. However, strong correlations emerged between PL and ToM for both ASD and TD. To summarize, Cardillo et al. (2020) found that impairments in pragmatic language are crucial characteristics of those with ASD, and deficits of ToM and EF also occur in those with ASD.
Andrés-Roqueta et al. (2020) also looked closely at pragmatic language in individuals with ASD, comparing language use with those with developmental language disorder (DLD). Andrés-Roqueta et al. (2020) describe how classically, pragmatic language deficits are considered by researchers as derived from deficiencies in ToM, EF, central coherence, and lack of social motivation, but also potentially in structural languages such as vocabulary and syntax. Andrés-Roqueta divides pragmatic language into two subdomains: linguistic pragmatics and social pragmatics. The study directly tests the hypothesis that linguistic pragmatics depends on structural language ability, whereas social pragmatics depends on ToM skills. The experiment included 20 individuals with ASD and 20 individuals with DLD and found that individuals with ASD performed poorly on ToM and social pragmatics tasks. However, individuals with DLD did not, but both individuals with ASD and DLD performed poorly on linguistic pragmatic tasks. Highlighting the difference ToM deficit makes in pragmatic language use in those with ASD compared to those with DLD.
Assessment Use in Autism Spectrum Disorder
Given the understanding gained from our exploration of ToM and Language in ASD, we can begin to have a fuller appreciation for the potential for the application of improvements, especially around ASD assessments. Kaufman (2020) provides a critical perspective on the ASD assessment landscape. To begin, Kaufman agrees with the critical nature of early identification of ASD. However, he also stresses the importance of recognizing current biases in ASD assessment, including that many people who do not have ASD may be diagnosed with ASD due to the profit motives of the manufacturers and providers of assessments; he also points out incorrect usage of assessment tools. Kaufman states that researchers designed such tools for highly impaired individuals, yet clinicians are using them on individuals with lower impairment.
Kaufman particularly takes issue with the ADI-R and ADOS assessments, stating that within the field, clinicians and researchers tout them as the “gold standard,” yet limited evidence is provided for their level of quality. Kaufman enumerates the biases within the ADI-R and ADOS assessments and critiques the positioning of these assessments and their delivery by a multidisciplinary team as unfounded and exclusionary. Kaufman believes in the power of assessments and their importance in the field of ASD, although he advocates for a diversity in assessment tools, the development of new assessments, and the ability for alternative assessments to be delivered by solo clinicians to be accepted by the field. Kaufman believes that doing this will increase availability and accuracy in ASD diagnosis.
Anagnostopoulou et al. (2020) also describe the critical role assessment plays in ASD diagnosis and advocate for the use of new technologies, specifically artificial intelligence (AI), to help in the shortening of the lead time for diagnosis. Anagnostopoulou et al. (2020) also comment on the current ADOS and ADI-R assessments, suggesting that they need to be improved, are too time-consuming to administer, and need more clinicians with the ability to administer the tests. Anagnostopoulou et al. (2020) describe using Machine Learning (ML) in novel scenarios, such as with gene data, brain imaging data, or by building an AI model from ADOS data to help better categorize patients or even reduce the number of questions needed for assessment. Other studies used accelerometers attached to individuals to compute their movement patterns and diagnose ASD via movement measurement. Other researchers used ML to evaluate video recordings of individuals to diagnose their ASD, and another used ML to classify responses of individuals viewing facial expressions to detect emotion.
Natural Language Processing (NLP), a form of AI, was reviewed by Anagnostopoulou et al. (2020) to see if computers could diagnose ASD via a chatbot interaction or to better summarize information from medical forms. Anagnostopoulou et al. (2020) described other researchers who used fuzzy logic maps to determine the comorbidity between ASD and Attention Deficit Hyperactivity Disorder (ADHD) or other researchers who used mobile apps, some using ML and others using questionnaires and videos to assess for ASD. Anagnostopoulou et al. (2020) conclude that there seems to be much potential for using new technologies in ASD assessment, especially AI-based technologies, which could lead to earlier detection of the disorder.
Cavus et al. (2020) also look at ML models in assessing ASD. They agree that ASD assessment is too costly and that there is a lack of trained clinicians in the field to diagnose those with ASD. Cavus et al. (2020) state that while there is research on ML, there needs to be more real-world clinical application. The researchers conducted a meta-analysis and enumerated several studies that use AI-driven tools for ASD assessment, including ones that used MRI data and genetic data. However, they indicate that the most promising promise lies in using ML for behavioral data and even models to build more efficient questionnaires. The tools reviewed delivered high accuracy in their studies. The hope is that continued research and development in this domain will lead to more real-world trials and applications.
Mukherjee et al. (2022) also agree with the challenges faced in the early identification of ASD due to assessments being lengthy and costly. Their meta-analysis reviewed 38 publications, looking at digital tools for ASD screening, focusing on scalable tools that clinicians could use on laptops, desktops, or mobile devices inside and outside of labs and in low-resource settings. Many varied technologies, from virtual reality (VR), gamified tasks, video recording, or even electronic devices like toy cars, were considered. Of all the technologies they evaluated, computer vision was most valuable, and tools that assessed individuals on social and motor tasks performed better than measurements on other dimensions.
Dahiya et al. (2021) also looked at new technology solutions for ASD assessment. Their research was motivated by requirements to assess during COVID-19. Researchers agreed that current assessments for ASD, including ADI-R and ADOS, are costly and challenging to deliver. The researchers evaluated several technology types: phone, web, live video, and recorded video, as well as different types of tools such as eye contact, directed speech, gestures, and repetitive play. Findings showed that technology-powered ASD assessments hold promise, with live video assessments implementing versions of ADI-R and ADOS showing the most promise, and video recordings also provide a viable alternative. Web and phone-based options also play a role.
Intervention in Autism Spectrum Disorder
Many researchers agree that technology can play a role in assisting in ASD assessment. Intervention is another area of active research, especially with the application of technology for improved results. Sandbank et al. (2022) conducted a systematic review of seven primary types of nonpharmacological interventions for ASD: behavioral, developmental, naturalistic developmental behavioral intervention (NBDI), Treatment and Education of Autistic and Communication Handicapped Children (TEACCH), sensory-based, animal-assisted, and technology-based. Their findings indicated that only behavioral, developmental, and NBDI showed significant results. NBDI and developmental interventions show sufficient evidence to be considered suitable treatments with behavioral interventions, often termed Applied Behavior Analysis (ABA), lacking methodological rigor and quality control.
Sandbank et al. (2022) describe that the treatment market for ASD is fragmented, with many different opinions on intervention. Many scholars call for earlier and more intensive treatments, for example, 25-40 hours a week for a year or longer, targeting infants and youth when the brain is developing. Sandbank et al. (2022) also highlight the bias involved in many of these interventions; for example, a caregiver often delivers the intervention or provides data for experimentation; this can lead to ineffective treatment and results that do not generalize outside of particular research environments. According to this review, researchers did not consider technology-based interventions significant in effectiveness. Ultimately, the authors call for more research held to higher quality standards to develop a trustworthy knowledge base of evidence of which interventions work best.
Dyrda et al. (2020) also look at ASD interventions, focusing specifically on interventions to develop ToM. Researchers evaluated several methods such as computerized mind reading training, emotion recognition in pictures training, and social-cognitive training, including how to listen to others, make friends and develop visual and auditory perception, recognize the thoughts and feelings of others, and look at things from different perspectives. Interventions that put human faces on fictional objects, such as cartoons or Thomas the Tank Engines, helped develop emotional awareness in hyper-systemizing ASD individuals. Dryda et al. also look at video modeling, computer-assisted technologies, and VR-based training interventions, recognizing these as delivering improvements. However, the paper concludes that while many ToM interventions are available and show promise, they often need to generalize more effectively to real-life social interactions.
Lecheler et al. (2020) also look at ToM-based interventions by evaluating a 12-week program that taught first and second-order ToM tasks, role-playing, modeling, or other social perspective-taking to ASD individuals. Researchers saw significant improvement in many areas, and parents reported improvement in their children. However, basic ToM did not improve after the program, which the researchers attribute to the failure to generalize. Peristeri et al. (2021) also looked at bilingualism as an intervention for ASD. They found that learning and using a second language improved ToM and EF, with having and using a second language providing an excellent means to develop second-order representations of what others are thinking and cognitive flexibility due to requiring cognition about different words in alternative languages that others are using during the bilingual speech.
Bamicha and Drigas (2022) also discussed the use of technology interventions for those with ASD, indicating that individuals with ASD may feel less stressed using computers as they are more predictable than other people and everyday environments. Bamicha and Drigas (2022) highlighted that virtual environments show potential for those with ASD, as well as mobile apps to allow individuals with ASD to stay better in control of their lives and to receive an education. Computer games may also be helpful as a safe place for an individual with ASD to try new skills without fear of failure. Robotics may also be useful to create a more predictable environment around the individual with ASD.
Conclusion
ASD is a disorder that could benefit from further technological innovation. ToM is considered a core driver of social and behavioral deficits seen in ASD, precisely the inability of individuals with ASD to understand the beliefs of others. ToM deficit leads to social difficulties, which in turn contribute to language difficulties for those with ASD. ASD individuals have trouble using language and may speak differently than their TD peers, for example, creating their own words or using words in strange ways, speaking too much about a particular topic, providing too much information, or being overly philosophical; they may also have trouble understanding language, often taking things in too much detail and too literally and being unable to absorb context correctly.
Assessment of individuals with ASD is a crucial step in facilitating treatment. Most researchers surveyed agreed that assessment for ASD today is too complex, time-consuming, and costly. Relying solely on the ADOD and ADI-R assessment instruments and recommending a multidisciplinary team for diagnosis may be contributing to problems faced in assessment. There is considerable opportunity to use technology to improve assessments, whether through video interaction or individual recordings, capturing behavioral data via digital tools, or using AI / ML to design better screening questionnaires or classify individuals based on data sets.
None of the research surveyed deeply examined using LLM chatbots for assessment. Considering the recent developments in LLM technology and the importance of language in ASD, there appears to be an area of considerable opportunity for future research to see if ASD could be diagnosed simply by an individual talking with an LLM or by an LLM reviewing video of an individual, or an LLM reviewing data sets that an individual has generated automatically.
Similarly, due to interventions' intensive and time-consuming nature, there is an excellent opportunity for technology to assist in scalability and operationalization. Individuals with ASD tend to have an affinity for computers and technology systems due to their predictable and less stressful interaction mechanisms. Computerized training, as with video modeling, has shown promise, yet none of the research surveyed looked explicitly at using LLMs for interventions. Due to the recent incredible advancements in LLM technology, new possibilities for intervention may have opened up, such as social skills training powered by LLM avatars, language use feedback within an LLM, or even LLM-powered robots to assist ASD individuals. Future research on how LLMs could be used for ASD assessments and intervention would be valuable and fill a gap in the existing literature.
References
American Psychiatric Association. (2022). Diagnostic and statistical manual of mental disorders (5th ed., text rev.). American Psychiatric Publishing.
Anagnostopoulou, P., Alexandropoulou, V., Lorentzou, G., Lykothanasi, A., Ntaountaki, P., & Drigas, A. (2020). Artificial intelligence in autism assessment. International Journal of Emerging Technologies in Learning (iJET), 15(6), 95–107. https://doi.org/10.3991/ijet.v15i06.11231
Andrés-Roqueta, C., & Katsos, N. (2020). A distinction between linguistic and social pragmatics helps the precise characterization of pragmatic challenges in children with autism spectrum disorders and developmental language disorder. Journal of Speech, Language, and Hearing Research, 63(5), 1494–1508. https://doi.org/10.1044/2020_JSLHR-19-00263
Bamicha, V., & Drigas, A. (2022). ToM & ASD: The interconnection of Theory of Mind with the social-emotional, cognitive development of children with Autism Spectrum Disorder. Technium Social Sciences Journal, 33, 42–72. https://doi.org/10.47577/tssj.v33i1.7010
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
Cardillo, R., Mammarella, I. C., Demurie, E., Giofrè, D., & Roeyers, H. (2021). Pragmatic language in children and adolescents with autism spectrum disorder: Do theory of mind and executive functions have a mediating role? Autism Research, 14(5), 932–945. https://doi.org/10.1002/aur.2423
Cavus, N., Lawan, A. A., Ibrahim, Z., Dahiru, A., Tahir, S., Abdulrazak, U. I., & Hussaini, A. (2021). A systematic literature review on the application of machine-learning models in behavioral assessment of autism spectrum disorder. Journal of Personalized Medicine, 11(4), 299. https://doi.org/10.3390/jpm11040299
Dahiya, A. V., DeLucia, E., McDonnell, C. G., & Scarpa, A. (2021). A systematic review of technological approaches for autism spectrum disorder assessment in children: Implications for the COVID-19 pandemic. Research in Developmental Disabilities, 109, 103852. https://doi.org/10.1016/j.ridd.2021.103852
Dyrda, K., Lucci, K. D., Bieniek, R., & Bryńska, A. (2020). Therapeutic programs aimed at developing the theory of mind in patients with autism spectrum disorders: Available methods and their effectiveness. Psychiatria Polska, 54(3), 591–602. https://doi.org/10.12740/PP/108493
Kaufman, N. K. (2022). Rethinking “gold standards” and “best practices” in the assessment of autism. Applied Neuropsychology: Child, 11(3), 529–540. https://doi.org/10.1080/21622965.2020.1809414
Lecheler, M., Lasser, J., Vaughan, P. W., & Leal, J. (2020). A matter of perspective: An exploratory study of a Theory of Mind autism intervention for adolescents. Psychological Reports, 124(1), 39–53. https://doi.org/10.1177/0033294119898120
Mukherjee, D., Bhavnani, S., Lockwood Estrin, G., Rao, V., Dasgupta, J., Irfan, H., Chakrabarti, B., Patel, V., & Belmonte, M. K. (2022). Digital tools for direct assessment of autism risk during early childhood: A systematic review. Autism, 28(1), 6–31. https://doi.org/10.1177/13623613221133176
Peristeri, E., Baldimtsi, E., Vogelzang, M., Tsimpli, I. M., & Durrleman, S. (2021). The cognitive benefits of bilingualism in autism spectrum disorder: Is theory of mind boosted and by which underlying factors? Autism Research, 14(8), 1695–1709. https://doi.org/10.1002/aur.2542
Sandbank, M., Bottema-Beutel, K., Crowley, S., Cassidy, M., Dunham, K., Feldman, J. I., Crank, J., Albarran, S. A., Raj, S., Mahbub, P., & Woynaroski, T. G. (2022). Autism intervention meta-analysis for studies of young children. Psychological Bulletin, 146(1), 1–29. https://doi.org/10.1037/bul0000215
Schaeffer, J., Abd El-Raziq, M., Castroviejo, E., Durrleman, S., Ferré, S., Grama, I., Hendriks, P., Kissine, M., Manenti, M., Marinis, T., Meir, N., Novogrodsky, R., Perovic, A., Panzeri, F., Silleresi, S., Sukenik, N., Vicente, A., Zebib, R., & Prévost, P. (2023). Language in autism: Domains, profiles and co-occurring conditions. Journal of Neural Transmission, 130, 433–457. https://doi.org/10.1007/s00702-023-02592-y
Vogindroukas, I., Stankova, M., Chelas, E. N., & Proedrou, A. (2022). Language and speech characteristics in autism. Neuropsychiatric Disease and Treatment, 18, 2367–2377. https://doi.org/10.2147/NDT.S331987
