2018/04/17 by Sweta Karlekar, Tong Niu, Karlekar, Sweta +3 · 2 citations
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Natural Language Processing Techniques #Topic Modeling
paper · pdf · doi:10.48550/arxiv.1804.06440
openalex publication_date 2018/04/17 · openalex created_date 2022/10/01 · openalex updated_date 2026/07/28
Alzheimer's disease (AD) is an irreversible and progressive brain disease\nthat can be stopped or slowed down with medical treatment. Language changes\nserve as a sign that a patient's cognitive functions have been impacted,\npotentially leading to early diagnosis. In this work, we use NLP techniques to\nclassify and analyze the linguistic characteristics of AD patients using the\nDementiaBank dataset. We apply three neural models based on CNNs, LSTM-RNNs,\nand their combination, to distinguish between language samples from AD and\ncontrol patients. We achieve a new independent benchmark accuracy for the AD\nclassification task. More importantly, we next interpret what these neural\nmodels have learned about the linguistic characteristics of AD patients, via\nanalysis based on activation clustering and first-derivative saliency\ntechniques. We then perform novel automatic pattern discovery inside activation\nclusters, and consolidate AD patients' distinctive grammar patterns.\nAdditionally, we show that first derivative saliency can not only rediscover\nprevious language patterns of AD patients, but also shed light on the\nlimitations of neural models. Lastly, we also include analysis of\ngender-separated AD data.\n