
Speech-based AI predicts adolescent psychopathology years ahead
A study of 204 youths (mean age ~11) used automated natural language processing of stress narratives to forecast internalizing psychopathology up to six years later; linguistic features explained more than twice the variance of traditional risk factors, with language style (not just explicit emotion) most predictive. Transformer-based embeddings were interpreted to reveal risk and resilience themes, including narratives of physical violence and social exclusion as risk markers, and routine activities and healthcare access as protective factors. These data-driven semantic dimensions also predicted future diagnostic outcomes, outperforming expert ratings of cumulative stress, suggesting a scalable AI framework to identify objective risk markers and guide targeted interventions in adolescence.













