Login
Computer Science @ Rochester
Friday, January 18, 2013
12:45 PM
CSB703
Emily Tucker Prud'hommeaux
Oregon Health & Science University
Automated evaluation of clinically elicited spoken language for diagnostic screening
Abstract: The application of natural language processing technology to medical data has historically focused on information retrieval from clinical records, but there has recently been progress in using NLP techniques to analyze spontaneous spoken language elicited during the administration of diagnostic instruments. Analyzing such spoken language data can reveal the presence of diagnostic markers for neurological disorders and other language features with diagnostic screening utility. In this talk, I apply NLP techniques to clinically elicited spoken language samples in order to distinguish between individuals with neurological disorders and neurotypical controls. First, I discuss the use of HMM-based and graph-based word alignment for evaluating the quality of narrative retellings and picture descriptions of elderly subjects with dementia. I then turn to automated syntactic and semantic analysis of spoken language data in children with autism. Finally I describe how these approaches could be used to enhance more mainstream language assessment tasks, including machine translation output evaluation and automated essay scoring.