Automated Coding of Children's Spoken Language is Associated with Internalizing Symptoms and Cortisol Reactivity to Stress

Automated coding has the potential to streamline otherwise laborious manual coding of psychologically relevant human behaviour. For example, automated coding of adults' written language (e.g., social media posts) shows that incidental language patterns are associated with internal psychological processes, including symptoms of depression. However, it is unclear whether automated coding of children's incidental spoken language is related to their symptoms and other indices of risk. We therefore examined associations between 146 children's (Mage = 9.53 years, SD = 0.88; 82 girls) spoken language, transcribed for automated coding, and their internalizing symptoms and cortisol stress reactivity. Children completed self-reports of depression and anxiety, and cortisol reactivity to stress was measured using salivary cortisol samples during the Trier Social Stress Test for Children (TSST-C). Emotional and other linguistic indices derived using Linguistic Inquiry and Word Count software package were meaningfully correlated with children's depressive symptoms and children's cortisol reactivity, with medium mean effect sizes of |r| = 0.21 (range: 0.19-0.28) and |r| = 0.23 (range: 0.20-0.29), respectively, across significant correlations, providing preliminary evidence that children's spoken language coded via automated methods is a valid marker of children's internal states. This study provides evidence that children's spoken language can be coded using automated technology, and that this coded data is related to established markers of children's psychological functioning. Compared to traditional methods of behaviour and affect coding, automated coding may be an efficient and cost-effective way of assessing linguistic indicators of children's internalizing symptoms and psychophysiological stress reactivity.