:: History    
     :: Committees    
     :: Contact Us    
     :: Guidelines    
     :: Online Submission    
     :: Poster Preparation    
     :: Guidelines    
     :: Online Registration    
     :: Schedule    
     :: Tutorial    
     :: Oral Presentations    
     :: Invited Speakers    
     :: Poster Presentations    
   MY AGENDA    
     :: Instructions    
     :: My Agenda    
     :: General Information    
     :: Lodging    
     :: Directions/Parking    
     :: Convention Center    
     :: General Information    
     :: Online Application    
     :: Become An Exhibitor    
     :: Recent Publications    
     :: Past SRCLDs    
     :: Search    
 SRCLD Presentation Details 

    Moving towards accurate early prediction of language disorders via parental report measures by combining machine learning and network science approaches  
Ariellle Borovsky - Purdue University
Donna Thal - San Diego State University
Laurence Leonard - Purdue University

SRCLD Year: 2020
Presentation Type: Special Session
Presentation Time: Thu, May 28, 2020 at 04:00 AM
Abstract View Full Summary
We seek to develop predictive models of low language (LL) outcomes in preschool and school age by analyzing parental report measures of early language skill using machine learning and network science approaches. We initially harmonize two longitudinal datasets of young children including demographic and standardized measures of early language skills (the MacArthur-Bates Communicative Developmental Inventories; MBCDI) as well as a later measure of LL (verified via CELF-P2 in one dataset, and parental report of school-age language disorder in the other). We additionally use item-level MBCDI data to calculate several graph-theoretic measures of lexico-semantic structure in toddlers’ early expressive vocabularies. We use machine-learning techniques to construct predictive models to identify toddlers who will have later preschool and school-aged language delays in these datasets. Our analyses illustrate that this approach yields robust and reliable predictions of later LL outcome with classification accuracies in single datasets exceeding 90%. Demographic variables and grammatical and lexico-semantic measures ranked highly in predictive classification, suggesting promising avenues for early screening initiatives and delineating the roots of language disorders.
Funding: NIH DC01368, HD052120 and DC000482.
Author Biosketch(es)




Supported in part by: NIDCD and NICHD, NIH, R13 DC001677, Susan Ellis Weismer, Principal Investigator
University of Wisconsin-Madison - Department of Communication Sciences and Disorders