A computational linguistic analysis of speech in elderly individuals with cognitive decline: the “Anchise-2022”

Abstract

In this article, we introduce the “Anchise-2022” corpus, the latest update of the “Anchise” corpus, with approximately 600 manually transcribed conversations, mainly between healthcare providers and elderly individuals with cognitive decline (dementia, Alzheimer's disease). We also analysed over 200 of these transcripts, to check if and to what extent we can correlate characteristics of such kind of speech production with the progression of dementia as measured by Mini-Mental State Examination (MMSE). To this purpose, we adopted computational linguistic analysis using Natural Language Processing (NLP) techniques, statistical-descriptive and inferential analysis. The results obtained allow us to assert, in line with existing literature, that certain parameters exhibit a degree of correlation with MMSE values. Additionally, it is feasible to utilize automatic classification techniques to infer the degree of linguistic/cognitive impairment.