Autism spectrum disorder – what does it really mean?
Autism is a neurodevelopmental condition that shapes how a person experiences the world and interacts with others. Around 1% of the global population is estimated to be on the autism spectrum, though the actual number may be higher. People on the spectrum may face challenges with interpreting emotions, understanding figurative language, or navigating conversations. How, then, does this influence the way they tell stories?
This question was explored by Prof. Izabela Chojnicka from AutismLabUW at the Faculty of Psychology, University of Warsaw, and Dr. Aleksander Wawer from the Institute of Computer Science at the Polish Academy of Sciences. As part of their study, they analyzed hundreds of essays written by students in the eight grade of primary school, looking for patterns characteristic of narratives produced by individuals on the autism spectrum.
AI in the service of science
To examine differences between narratives written by students with ASD and their neurotypical peers, the researchers used advanced natural language processing (NLP) techniques – a field focused on analyzing and interpreting human language with the help of AI – as well as deep learning methods.
In the first phase of the study, they analyzed 333 exam essays: 195 written by autistic students and 138 by non-autistic students. They applied linguistic tools such as the LCM model, which measures the level of abstraction in language, and the Gunning fog index, which assesses text readability. They also examined emotional content, tracking how often words with positive or negative connotations appeared, as well as overall essay length.
Less emotion, more facts
The results revealed clear differences. Students with ASD were more likely to use literal language and less likely to include positively charged emotional words or verbs describing emotional and cognitive states. Their narratives tended to focus on facts and actions rather than on characters’ inner experiences.
By contrast, neurotypical students more often incorporated reflections and emotional depth into their writing. Interestingly, while students with ASD used fewer positively valenced words, the frequency of negatively toned words was similar across both groups.
Another notable finding: although essays written by students with ASD were generally shorter, they exhibited greater linguistic complexity, defined here as the use of longer words. At the same time, teachers grading the essays using standard criteria did not observe significant differences in overall writing quality. This suggests that traditional assessment methods may overlook subtle but meaningful differences in narrative style.
AI as a new diagnostic tool?
These findings are important not only for advancing scientific understanding but also for their practical implications. Computational analysis of narrative could contribute to the development of new diagnostic tools.
“Currently, the primary diagnostic tools in autism research are developmental interviews, observations, and paper-and-pencil tests and questionnaires. Intensive work is now underway on computational methods that could support diagnosticians in a way similar to standardized tools. Our findings suggest that the analysis of linguistic pragmatics may become one such indicator, useful in screening and diagnosis,” says Prof. Izabela Chojnicka.
Artificial intelligence could therefore help identify individuals who may require further evaluation by analyzing how they construct written narratives. However, it is important to emphasize that such tools would complement – not replace – the full diagnostic process.
What does this mean for education?
Understanding the distinctive features of narratives produced by individuals with ASD may also have important implications for education. Recognizing that such texts may be shorter, more literal, and less emotionally expressive could help teachers assess student work more fairly.
“Increasing our understanding of pragmatic competencies in autism allows us to better understand individuals on the spectrum, which may ultimately influence educational systems,” Prof. Chojnicka notes.
This research opens the door to developing better educational materials and language exercises tailored to the needs of students on the autism spectrum. It also suggests the need for more nuanced assessment methods – ones that take into account differences in narrative style rather than measuring all students against the same neurotypical standards.
What’s next?
The study shows that artificial intelligence can be a powerful tool for supporting both the diagnosis and education of individuals on the autism spectrum. The next step will be to develop analytical tools that could eventually become a standard component in assessing linguistic and communication competencies in people with ASD.
In the future, analyzing written texts may become one element of autism diagnosis, helping specialists identify individuals who need additional support. That would mark a meaningful step toward more inclusive education – and a deeper understanding of how people on the autism spectrum experience and describe the world.
The article was originally published in Polish on the Serwis Naukowy UW website on March 12, 2025.
