Using NLP to Detect Linguistic Cues in Alzheimer's Disease Patients

This paper aims to detect linguistic characteristics and grammatical patterns from speech transcriptions generated by Alzheimer's disease (AD) patients. This is considered an important application of natural language processing and deep learning techniques for computational health. The authors propose several neural models such as CNNs and LSTM-RNNs -- and combinations of them -- to enhance an AD classification task. The trained neural models are used to interpret linguistic characteristics of AD patients (including gender variation) via activation clustering and first-derivative saliency techniques. Language variation can serve as a proxy that monitors how patients' cognitive functions have been affected (e.g., issues with word finding and impaired reasoning).