TOWARD ADAPTIVE LANGUAGE-BASED PSYCHOMETRIC ASSESSMENT OF COGNITIVE DECLINE

Advances in large language modelling have created new opportunities for psychometric assessment based on naturalistic language use. Conventional psychometric instruments rely on fixed items and standardized norms, which may limit sensitivity to subtle cognitive changes and introduce bias related to educational level, literacy, and professional background. Language modelling approaches enable the analysis of spontaneous speech and free-text responses, offering rich markers of cognitive functioning across semantic, lexical, syntactic, and discourse levels. In the context of neurocognitive disorders, such methods show particular promise for the early detection of mild cognitive impairment and early signs of dementia, where alterations in lexical diversity, semantic coherence, narrative structure, and ease of lexical and conceptual evocation often precede overt functional decline. Additionally, impairments in the retrieval and contextualization of landmark events embedded in an individual’s sociocultural space may provide sensitive indicators of disruption in autobiographical and semantic memory systems. Furthermore, language model–driven adaptive testing allows assessment tasks to be dynamically adjusted to an individual’s educational attainment and to vocabulary domains associated with specific professions, reducing construct-irrelevant variance and improving diagnostic accuracy. By tailoring tasks and questions (altering difficulty and semantic content) while preserving underlying cognitive targets, language-based assessments may enhance both sensitivity and validity. This abstract reviews emerging psychometric frameworks integrating language modelling for cognitive assessment, discusses validation and ethical challenges, and outlines pathways toward personalized, longitudinal, and clinically relevant measurement systems for cognitive decline.