Proceedings of the Annual Meeting of the Cognitive Science Society

As demonstrated in previous studies, Alzheimer’s diseaseleads to a degradation of vocabulary and communication skills.Novels by writers who are known to have suffered from thisdisease were compared with respect to their lexical richnessand syntactic complexity. Those written after the break-outof the disease have shown to use a considerably smaller lex-icon and a reduced syntactic complexity of the sentences.This makes us assume that writings of individual authors canbe classified automatically into “pre-Alzheimer’s period” and“Alzheimer’s period”. But the writing style of an author ishighly individual. Can we still detect whether any given novelis written by an author who suffers from Alzheimer’s? To as-sess this, we use a corpus of novels by three well-known writ-ers who were diagnosed with Alzheimer’s: Iris Murdoch, TerryPratchett and Agatha Christie. Using a mostly stylistic set offeatures we are able to distinguish between novels written un-der the influence of the disease and novels written by healthywriters with more than 82% accuracy. The classification ofthe novels of a given author into “pre-Alzheimer’s period” and“Alzheimer’s period” is accomplished with more than 86% ac-curacy. We also prove that our feature set is versatile enoughto be able to distinguish between authors in general and bookswith high precision.