A study published in Nature Mental Health used four natural language processing (NLP) models to analyse interviews with more than 200 children aged 9 to 13 about stressful events in their lives. The models proved effective at predicting whether these children would develop mental health conditions six years later.
Researchers concluded that the stylistic features of the children’s speech mattered more than the content itself. What counted was how the respondents constructed their sentences, with particular significance placed on the use of short connecting words such as “and,” “to,” and “but.”
Adolescence is a period when depression and anxiety often emerge. If these disorders take root, they can be very difficult to treat. The years before a diagnosis is made are critically important, yet they remain poorly understood, as do the methods for identifying which children are at risk.
Previously, assessing the likelihood of mental health issues involved face‑to‑face clinical interviews or complex techniques such as hormone measurements or cell studies — approaches that are costly and impractical. In contrast, speech analysis is a simple and accessible method; all that is needed is a basic audio recording.
The children’s stories themselves were also informative. Accounts of stress, violence, or feelings of complete isolation were more common among those who later experienced mental health disorders. Conversely, mentions of support, participation in sports or clubs, or having received help from psychologists were associated with a more positive outlook and greater resilience to stress.
About the Author
