Department of Language Modelling
NCN grant

Tell your story. Social communication in autism: a computational approach

The project uses deep learning and natural language processing to analyse the speech and narratives of autistic people, as a step towards objective computational measures of language in autism.

Lead Investigator: Izabela Chojnicka

Grant Number
2020/39/D/HS6/00809
Start Date
1 January 2023
End Date
31 December 2024

Project factsheet
English name: Tell your story. Social communication in autism: a computational approach
Polish name: Co słychać? Komunikacja społeczna w autyzmie: podejście obliczeniowe
Project type: A National Science Centre SONATA 16 grant (2020/39/D/HS6/00809)
Duration: 23 June 2021 – 22 January 2025 (originally until 22 June 2024)
Host institution: Faculty of Psychology, University of Warsaw
Principal investigator: Izabela Chojnicka
IPI PAN investigator: Aleksander Wawer

Project summary

Our speech is a key to the mind. There is more to what we say than many of us may think. From a written text or statement, we are able not only to recognise the unique style of the writer, but also, with the use of computational methods, the gender, age or socio-economic status of the author. Recent research shows that artificial neural networks and natural language processing techniques identify elusive and/or non-obvious speech characteristics. Artificial intelligence is entering new areas of our lives, and it is also slowly making its way into psychology and psychiatry.

This project aims to use deep learning models, a branch of artificial intelligence, and natural language processing in the analysis of the speech of people with autism in comparison to typically developing people. The results are expected to increase our understanding of social communication in autism, in particular the pragmatic aspects of language.

Since most research on narrative skills in autism has been conducted in male-predominant groups, the project further investigates sex differences in narratives produced by people with autism. Is the speech of autistic girls different from that of autistic boys? Does it differ from the speech of typically developing girls, or is it more like the speech of typically developing boys? Does it depend on the situational context – do people with autism speak differently when they tell about personal experiences and when they tell a story depicted in pictures? Can these differences be detected using fully automated methods?

The research is intended to be a step towards the development of quantifiable and objective computational measures of linguistic characteristics in autism. Computational methods, including models based on deep neural networks, could improve clinical diagnostics of autism and screening. From a technical point of view, their use only requires a computer or a smartphone, and data processing can be simplified by automatic speech transcription. In the future, detecting autism using computational models may be quick and effective on a large scale, supporting the clinician in the decision-making process.

Publications

Earlier joint work of Izabela Chojnicka and Aleksander Wawer on the language of autistic people includes: