Sunghye Cho
Linguisitc Data Consortium (LDC)
UPenn

Automated acoustic and lexical analyses of natural speech in patients with Alzheimer’s disease

Speech production is a complex behavior, involving activation of multiple regions of the brain; thus, examining speech production provides opportunities to identify disease markers. Since Alzheimer’s Disease (AD) accounts for up to 80% of patients with dementia, affecting over 50 million people worldwide, much attention has been paid to cognitive changes of AD, including the linguistic domain. In this talk, I will present a series of AD-related projects that identify specific language biomarkers of AD, Mild Cognitive Impairment (MCI), and logopenic variant primary progressive aphasia (lvPPA), which is a non-amnestic type of AD, using our automated lexical and acoustic pipelines. I will also show correlations with our language variables and traditional clinical ratings, disease severity predition, and longitudinal changes in patients’ speech. I will also present results of automatic classification of patient groups with our language features. Lastly, I will present the application of our automated pipelines to speech samples of standardized neuropsychological tests.