Events Calendar

Seminar-Algorithms for detecting atypical language use in autism spectrum disorders

Speaker: Jan van Santen, Center for Spoken Language Understanding, Institute on Development and Disability, Oregon Health & Science University

Talk Title: Algorithms for detecting atypical language use in autism spectrum disorders

Abstract: The DSM-5 lists repetitiveness and impaired reciprocal social interaction as core symptoms of autism spectrum disorder (ASD), but does not list language impairment. Yet, language use is often atypical in ASD, being a natural modality for core symptoms to manifest. Standard language measures are not optimal for capturing these characteristics because they are too structured: Analysis of natural language samples is needed. However, such analysis is time consuming and inexact when conducted manually. Computational methods are needed.

We developed and applied algorithms to transcripts of Autism Diagnostic Observation Schedule (ADOS) sessions with children ages 4-8 with high-functioning ASD, Specific Language Impairment (SLI), or Typical Development (TD). The ASD group was divided into children with SLI (ALI) and without (ALN). Results showed ASD-specific atypicalities in verbatim and topical repetitiveness, discourse marker use, type of disfluencies, and other features of language use. These results attest to the feasibility of computing ASD-specific characteristics from natural language samples, tapping into multiple aspects of core ASD symptoms. Their usefulness is demonstrated by the intricate pattern of differences and similarities between the ASD and SLI groups and the ALI and ALN groups.

Biography: Jan van Santen obtained his PhD in Mathematical Psychology at the University of Michigan in 1979. He worked initially on visual perception and image processing at New York University and Bell Labs, and then switched to speech technology in 1985. He developed the prosody generation components of the Bell Labs text-to-speech system. In 2000 he became the Director of the Center for Spoken Language Understanding, now part of the Oregon Health & Science University. Here, he became one of the pioneers of a growing new field: the application of Natural Language Processing algorithms to neurological and neurodevelopmental disorders for diagnostics, remediation, and assistive communication, with special emphasis on autism spectrum disorders. In his spare time, he runs a startup, BioSpeech, that works on algorithms for processing biological sounds, including not only speech but also snoring and rodent calls.

He has written over 100 peer-reviewed papers, was an editor of Speech Communication and of the Journal of Mathematical Psychology, was the editor of a book, Progress in Speech Synthesis, and has seven patents.


Host: Shrikanth Narayanan & Daniel Bone

Location: Ronald Tutor Hall of Engineering (RTH) - 211

Audiences: Everyone Is Invited

Contact: Tanya Acevedo-Lam/EE-Systems