CharMark: Character-Level Markov Modeling to Detect Linguistic Signs of Dementia

Dementia, one of the most prevalent neurodegenerative diseases, affects millions worldwide. Understanding linguistic markers of dementia is crucial for elucidating how cognitive decline manifests in speech patterns. Current non-invasive assessments like the Montreal Cognitive Assessment (MoCA) and Saint Louis University Mental Status (SLUMS) tests rely on manual interpretation and lack detailed linguistic insights. This paper introduces a first-of-its-kind interpretable artificial intelligence (IAI) framework leveraging first-order Markov Chain models to characterize linguistic patterns associated with early-stage dementia. By computing steady-state probabilities for characters in speech transcripts from dementia subjects and healthy controls, we identified distinctive character-usage patterns. The space character " ", representing pauses, and letters such as "n" and "i", showed significant differences between groups. Principal Component Analysis (PCA) visualizations highlighted natural clustering corresponding to cognitive status. Kolmogorov-Smirnov tests confirmed statistically significant distributional differences in character usage between groups. Additionally, a Lasso Logistic Regression validated the relevance of these linguistic markers. Our primary contribution remains the identification and characterization of specific character-level linguistic patterns associated with cognitive decline. These findings demonstrate the potential of character-level modeling for detecting subtle language changes linked to dementia, informing the future development of sensitive cognitive screening tools.