Alzheimer’s disease (AD) is the most common cause of dementia and affects wide parts of the elderly population. Since there exists no cure for this illness, it is of particular interest to develop reliable and easy-to-use diagnostic methods to alleviate its effects. Speech can be a useful indicator to reach this goal. We propose a purely statistical approach towards the automatic diagnosis of AD which is solely based on n-gram models with subsequent evaluation of the perplexity and does not incorporate any further linguistic features. Hence, it works independently of a concrete language. We evaluate our approach on the DementiaBank which contains spontaneous speech of test subjects describing a picture. Using the Equal-Error-Rate as classification threshold, we achieve an accuracy of 77.1%. In addition to that, we studied the correlation between the calculated perplexities and the Mini-Mental State Examination (MMSE) scores of the test subjects. While there is little correlation for the healthy control group, a higher correlation could be found when considering the demented speakers. This makes it reasonable to conclude that our approach reveals some of the cognitive limitations of AD patients and can help to better diagnose the disease based on speech.
Cite as: Wankerl, S., Nöth, E., Evert, S. (2017) An N-Gram Based Approach to the Automatic Diagnosis of Alzheimer’s Disease from Spoken Language. Proc. Interspeech 2017, 3162-3166, doi: 10.21437/Interspeech.2017-1572
@inproceedings{wankerl17_interspeech,
author={Sebastian Wankerl and Elmar Nöth and Stefan Evert},
title={{An N-Gram Based Approach to the Automatic Diagnosis of Alzheimer’s Disease from Spoken Language}},
year=2017,
booktitle={Proc. Interspeech 2017},
pages={3162--3166},
doi={10.21437/Interspeech.2017-1572}
}