How Children’s Linguistic Style Predicts Future Mental Health Problems

Natural language processing models that analyze the exact words children use when talking about stressful events prove better at predicting future mental health problems than a panel of human experts, according to a study published in Nature Mental Health. Researchers at Stanford’s School of Humanities and Sciences found that evaluating recorded interviews of more than 200 children aged 9 to 13 accurately predicted whether those children would develop mental conditions six years later.

Linguistic Style Outperforms Content in Predicting Adolescent Psychopathology

How children construct their sentences matters significantly more than the actual stories they tell about stress, according to the findings. Researchers discovered that the use of small connector words—such as “and,” “to,” and “but”—serves as a stronger indicator of future psychological distress than a child’s direct description of an event. This discovery shifts focus away from the specific nature of a stressor and toward structural speech patterns.

“We believe this study provides a robust proof of concept for the development of scalable tools that identify markers of risk before individuals are diagnosed,” said Chase Antonacci, the study’s lead author and a neuroscience doctoral student at Stanford’s School of Humanities and Sciences (H&S). Antonacci noted that adolescence marks the period when depression and anxiety most often emerge. Once these disorders take hold, they prove notoriously difficult to treat, leaving clinicians searching for reliable ways to identify at-risk youth during the critical window leading up to diagnosis.

Comparing Speech Analysis to Traditional Risk Indicators

Traditional clinical methods offer varying degrees of predictive utility, but language analysis provides a unique advantage in accessibility. According to Ian Gotlib, the study’s senior author and a professor of psychology in H&S, speech analysis requires no specialized equipment or invasive testing.

“These factors—cortisol, stress reactivity, and telomere length—all have some predictive utility, but speech is something that is inexpensive and scalable,” Gotlib said. “It’s easy, it’s accessible, and it may be a stronger predictor of the development of problems than any of these other factors alone.”

To conduct the research, Gotlib’s lab utilized audio-recorded interviews lasting about 1.5 hours each. A panel of experts initially reviewed the interviews using a traumatic events screening inventory, also known as TESI, to rate stressor severity ranging from financial insecurity and parental divorce to abuse and natural disasters. The panel assigned a single cumulative number to each child’s stress experience. However, researchers recognized that reducing these clinical interviews to a single metric lost considerable richness and nuance, prompting the application of four natural language processing models to the audio recordings.

Content and Context: Violence, Resilience, and Therapy Mentions

While linguistic style proved most predictive, the actual content of the children’s speech also revealed vital links to future problems or resilience. Statements associated most strongly with risk described extreme physical violence, such as being punched or choked, or harsh social exclusion, such as feeling that an entire school was against them.

Conversely, resilience linked directly to statements concerning social support and participation in activities like sports and school clubs. Notably, mentions of mental healthcare itself—specifically references to therapists or counselors—emerged as one of the strongest protective signals within the data.

The research team plans to test the models using a larger dataset as a crucial next step. “If these findings hold, it means we may be able to just take smartphone recordings of children talking, analyze that speech, and identify which children are at risk, years before they might develop a disorder,” Gotlib said.

Frequently Asked Questions

What did the natural language processing models analyze in the children’s speech?

The models analyzed the structural style of the children’s speech—specifically how they constructed sentences and used small connector words like “and,” “to,” and “but”—rather than just the content of their stress descriptions.

How far in advance can these models predict mental health issues?

According to the study published in Nature Mental Health, the models accurately predicted whether children aged 9 to 13 would develop mental health conditions up to six years later.

Why is speech analysis considered better than traditional testing methods for children?

Unlike methods requiring blood draws, cortisol measurement, or telomere length analysis, speech is inexpensive, easily accessible, scalable, and non-invasive, making it feasible to implement across large groups of children.