ATTENTION DISORDER
In writing, ADHD makes young people go from ‘I’ to ‘one’
Two psychologists from the University of Geneva have analysed the editorial style and form of short autobiographical texts to identify written markers of attention disorder.
Algorithms can detect evidence of ADHD in teenagers’ school essays. | Photo: Pavel Danilyuk / Pexels
The signs of attention deficit disorder with or without hyperactivity (ADHD) can be tracked in autobiographical stories. This is the hypothesis tested by Juan Barrios and Martin Debbané, who are researchers at the Faculty of Psychology and Educational Sciences of the University of Geneva. “The analysis of writing, with an interest in the meaning of words and their frequency, is already used to detect linguistic markers in depression or post-traumatic stress”, says Barrios. As for stylometry, it is usually used to identify the author of a text. We wanted to see if it made it possible to identify linguistic markers of ADHD in adolescents”.
They tested two groups of 24 French-speaking teenagers, one formed of young people with a confirmed diagnosis, the other representing the Geneva population. Each teenager wrote three personal memories. Their texts were statistically analysed by a stylometry algorithm and other automatic language processing techniques. The form mattered more here than the substance, with attention to functional words or sequences of characters.
A strong marker appeared: the use of the French impersonal pronoun ‘on (one)’ in the ADHD group, against ‘je (I)’ among controls. “This often reflects a retreated posture that can be related to difficulties in regulating emotions”, says Barrios, confirming what he has seen during twenty years of practice as a clinician. This marker, combined with others highlighted by the study, allowed correct predictions at rates ranging from 69–100 percent, depending on the configurations.
They are promising results, but they call for caution. “The objective is not to make a diagnosis using AI, but to provide a useful tool for professionals, helping them to prioritise”. To validate the method, larger-scale studies are necessary. And to adapt it to other languages, the model will have to be tested in each one.
