Speaker
Description
Alzheimer's Disease (AD) is a progressive neurodegenerative disorder that significantly impairs cognitive functions such as memory, language, and problem-solving, profoundly impacting individuals and their families. Language change is among the earliest indicators of cognitive decline, offering a valuable pathway for early diagnosis.
While linguistic markers of AD have been extensively studied in English using Natural Language Processing (NLP) techniques, research in local and low-resource languages like Norwegian remains limited. In this study, we present a Norwegian-language dataset consisting of speech transcripts from AD patients and cognitively healthy control subjects. To address the scarcity of Norwegian-language resources for dementia detection, we translated the ADReSS Challenge dataset into Norwegian using a combination of human and machine translation.
We explore various text representation methods, including both sparse and dense embeddings, and establish baseline results using a range of machine learning (ML) and deep learning (DL) models. Additionally, we evaluate the performance of multilingual and domain-specific transformer-based models such as mBERT, NorBERT, and NB-BERT.
Our findings indicate that the linguistic difficulties faced by AD patients particularly when describing cookie theft picture are effectively captured in their speech narratives. This supports the use of NLP techniques for early AD detection in low-resource language settings.
