Symbolic Recurrence: A Framework for Linguistic Biomarker Discovery in Speech

The rising prevalence of neurocognitive disorders, such as Alzheimer's disease (AD), poses a significant global health challenge. Traditional analytical methods, including clinical interviews and paper-based tests like the Mini-Mental State Examination (MMSE), Mini-Cog Test, and Montreal Cognitive Assessment (MoCA), are limited by subjectivity, memory biases, and interviewer variability. To advance our understanding of cognitive-linguistic patterns, this study introduces an interpretable artificial intelligence (AI) architecture to analyze distinctive speech characteristics between healthy individuals and those with dementia. Initially, each unique character in the speech transcripts is encoded as a distinct number, enabling fine-grained analysis of linguistic patterns. Recurrence Quantification Analysis (RQA) is then applied to generate recurrence plots, capturing the dynamic temporal structures and non-linear patterns in speech production. From these plots, we extract salient features using deep metric learning with Siamese networks, which learn to represent and differentiate essential linguistic characteristics in a meaningful embedding space. This novel architecture enables the discovery of subtle yet significant differences in language patterns between groups. Our approach reveals distinct linguistic signatures, demonstrating clear separability between healthy and dementia-related speech patterns, as evidenced by quantitative evaluation metrics. This research advances our understanding of the underlying linguistic indicators of cognitive disorders, providing insights into the characteristic patterns of language changes associated with cognitive decline. These findings contribute to the development of more nuanced and interpretable approaches for analyzing cognitive-linguistic patterns in clinical settings.