Detecting markers of Alzheimer’s disease in spoken and written discourse

The pathological build-up of Alzheimer’s disease begins years prior to overt expression of symptoms, by which stage it is too late for disease-modifying therapies to have an effect. Identifying individuals at risk of later developing the disease could pave the way for finding effective medications. Although often considered a disorder of memory, discourse has been found to change due to the presence of Alzheimer’s disease pathology in the brain, and these changes can be seen early on, including in Mild Cognitive Impairment. Advances in Natural Language Processing (NLP) - a branch of computer science which aims to analyse human language at scale - and Machine Learning, mean that changes in discourse can be explored objectively and quickly. Recent studies have classified groups of Alzheimer patients and healthy controls with up to 82% accuracy. However, much research to date has used the same dataset and picture-description task to elicit discourse, and has not followed participants over time to see how discourse may change according to disease. In this talk two studies are outlined which aim to investigate: 1) the utility of different speech elicitation tasks and discourse markers for group classification, both cross-sectionally and longitudinally; and 2) markers of Alzheimer’s disease in written discourse and how they change across the lifespan, in a collection of texts gathered from members of the public. Cross-sectional results for spoken discourse are presented, along with future analysis plans.