Automatic Quantification of Atypical Topic Repetition in Single Daily-Conversation Resulting from Alzheimer's Disease

Identifying signs of Alzheimer's disease (AD) in everyday situations has become increasingly important. Previous studies have succeeded in quantifying language dysfunctions and identifying AD from speech data typically during neuropsychological tests. However, no study has yet investigated atypical topic repetition within single daily-conversations, although it was reported as a prominent characteristic by previous observational and descriptive studies. In this study, we analyzed daily conversational data collected from a monitoring service and compared topic patterns in single conversations of seniors with and without AD. We first found that all features extracted from manual transcription to measure topic repetition showed significant increases in the AD group. Moreover, these features fully automatically extracted from voice data by using a speech-to-text algorithm could capture atypical topic repetition with comparable and large effect sizes of those extracted from manual transcriptions. The results indicate that quantifying atypical repetition could help automatically detect AD in everyday situations.