No abstract

Background: Several recent studies have used automated linguistic analysis of naturalistic speech in a picture-description task (e.g. Cookie Theft) combined with machine learning approaches to distinguish the speech of those with dementia from healthy age-matched controls. Extension of these techniques to predict continuous variables related to cognitive status offers a means to track the severity of dementia over time and evaluate the effectiveness of interventions. Here we examine the relationship of speech variability to executive function, the ability to manage conflicting information in speeded task performance. Higher executive function confers protection against the clinical manifestation of dementia despite underlying neurodegeneration. Method: In a new sample of 76 healthy adults aged 65-75, we measured executive function using an extensive test battery, and elicited spoken picture descriptions in the same individuals. The battery included N-back, Simon task, verbal fluency, Color Word Interference (CWI or Stroop test), Trails A & B, the elevator subtests of the Test of Everyday Attention, and Logical Memory. Two picture description narratives were recorded and subjected to automated analysis generating over 400 linguistic features, grouped into 8 composite measures.