Automated Scoring of Tablet-Administered Expressive Language Tests

Area of Science:

• Pediatric developmental psychology
• Computational linguistics
• Speech-language pathology

Background:

• Speech and language impairments affect up to 10% of children.
• Expressive language disorders are frequently undiagnosed.
• Objective and reliable assessments are crucial for early intervention.

Purpose of the Study:

• To develop and validate computational models for automated scoring of pediatric expressive language tasks.
• To assess the accuracy and reliability of AI-driven scoring compared to traditional methods.
• To provide tools for more efficient and objective developmental language evaluations.

Main Methods:

• Utilized a tablet-based framework for administering and recording expressive language tasks.
• Employed machine learning, specifically deep neural networks, for automated scoring.
• Compared automated scores against traditional paper-and-pencil scoring for four distinct tasks.

Main Results:

• Automated scoring achieved high accuracy (83-99%) at the item level for all four tasks.
• Automated scores showed strong and significant correlations with manual scoring (ρ = 0.76-0.99).
• Models demonstrated effectiveness with both clean and verbatim speech transcripts.

Conclusions:

• Automated computational methods can reliably and objectively administer and score expressive language tasks.
• AI-driven assessments hold significant potential for improving pediatric developmental language evaluations.
• This technology can aid in the timely identification of expressive language disorders in children.