A study led by researchers at Stanford University indicates that analyzing children’s speech can forecast their risk of developing mental health conditions, such as depression and anxiety, up to six years later. The findings were published in *Nature Mental Health* on July 31, 2026.

Read More

Utilizing four natural language processing models, the study assessed recorded interviews of over 200 children aged 9 to 13 discussing stressful life events. The linguistic models outperformed human experts in predicting future mental health issues, demonstrating that the way children construct their sentences—specifically the use of connector words like "and," "to," and "but"—is a stronger indicator than the content of their speech.

Chase Antonacci, a neuroscience doctoral student and lead author of the study, remarked that the research offers a substantial proof of concept for developing scalable tools to identify early markers of mental health risks. He emphasized that adolescence is a critical period for the emergence of depression and anxiety, and current assessments lack scalability for large populations.

Previous methods of identifying mental health risks primarily relied on clinician assessments, which are not practical for widespread application. Other objective methods, such as blood tests for stress hormones or physiological measurements of stress reactions, do not offer the accessibility that speech analysis provides. Ian Gotlib, the senior author of the study and a professor of psychology at Stanford, noted that speech analysis could be a more effective predictor of future issues compared to these other factors.

Gotlib’s research team has been following a cohort of young people over several years, conducting lengthy interviews that explore various stressful experiences. Initially, stressors were rated by a panel of experts, but the researchers recognized the need to analyze the richness of speech in these recordings.

By employing natural language processing strategies, the team aimed to determine if the verbal patterns and styles of speech from children could foretell future mental health disorders. The models revealed that linguistic style holds significant predictive power, in line with prior research suggesting that patterns in word usage, such as frequent first-person pronouns and conjunctions, are linked to mental health issues.

While linguistic style proved to be the most predictive element, content also revealed crucial connections. Risks were associated with descriptions of extreme violence and social exclusion, whereas resilience correlated with mentions of social support and involvement in activities like sports or school clubs. Notably, discussing mental healthcare, such as referring to therapists or counselors, was one of the strongest indicators of protective factors.

Looking forward, Gotlib indicated that the next step includes testing these models on a larger dataset to validate the findings. He expressed optimism that simple smartphone recordings of children’s speech could eventually identify those at risk for mental health issues years in advance.

The study also included contributions from psychology doctoral students Eugenia Giampetruzzi and Sabrina Jones, undergraduate Kaitlyn Kwan, postdoctoral scholar Jessica Uy, and James W. Pennebaker from the University of Texas at Austin, who developed the software used in the research. It received support from the National Institute of Mental Health and the National Science Foundation.