AI Project for Early Childhood Language Assessment Aims to Speed Up Intervention for Children with Developmental Delays

Evaluating a child’s early childhood language development can start with as little as six minutes of observing structured playtime between child and parent. But that’s the easy part.  

After the toys are packed up and the families are headed home, specialists still need to review the recorded session and code the results—counting gestures, vocalizations, along with single or multiple words spoken—and compile the data.  

At Juniper Gardens Children’s Project, researchers leading a pilot project involving artificial intelligence in child language evaluation are hoping to shorten the time it takes to analyze the data by from language development assessments. They are working at training artificial intelligence to code Early Communication Indicator (ECI) assessments to help children access therapeutic intervention faster. 

The ECI is a standardized play-based tool used across the world to evaluate children ages 6 months to about 3 and a half years. The ECI requires trained staff to administer and score assessments.  

"It takes really skilled practitioners to accurately count the number of times that a child uses expressive communication during those six minutes,” said Jay Buzhardt, a member of the research team that developed and disseminates the ECI at Juniper Gardens. 

Coding the results can easily take four times as long as the testing itself. These coded evaluations are necessary for clinicians to determine if a child may need therapy and to identify what kind of treatment plan is needed. However, lack of trained evaluators can make it difficult to obtain an assessment in a preschool or pediatric office. 

For children who are falling below expected benchmarks, this creates a bottleneck that can delay implementing therapy, monitoring progress and making data driven decisions. 

“The reason we are really investigating the use of AI ... is that it would reduce or eliminate the need to have trained assessors,” Buzhardt said. “Importantly, it would reduce and eliminate the burden of scoring these assessments.” 

As anyone who has spoken to an automated phone system knows, computers can struggle to interpret adult speech. Coding the expressions of a young child who is just starting to form words is even more of a challenge. As Buzhardt explained, “This is something that modern AI systems struggle to do because they are trained on adult speech." 

The first step of the pilot project involves training an existing AI model to near-mastery levels on an initial set of ECIs. Then the researchers will address improving the algorithm to demonstrate reliability that is equivalent to human coders. A second step will involve investigating the usability and feasibility of individuals, including educators and parents, to record videos, submit them to their existing web platform, and then using AI-coded ECI scores to analyze the videos. 

The project, funded through a Research Rising Pilot Award depends on an interdisciplinary team of researchers at Juniper Gardens and a local group of AI developers based in Overland Park. This work was inspired by Dwight Irvin’s work in using speech recognition tools to measure adult-child interactions in community spaces. He is an affiliate investigator at the University of Florida. 

“This groundbreaking work will help solve the scoring burden individuals can encounter when using the ECI and, in all likelihood, increase practitioner usage, ensuring more children gain the language skills needed for school and life.” Irving said.  

The project is led by Rebecca Davis at Juniper Gardens, and she is optimistic about the progress. She said they have given the program authentic videos to transcribe and score.  

“They’re starting to score correctly for the definitions we’re giving them,” she said. "We're now talking through the process of providing them more, and then starting to check accuracy and troubleshoot errors.”  

Children respond more rapidly to therapeutic interventions the earlier they receive them.  

The researchers hope using AI in early childhood intervention assessment coding will minimize the need for training and painstaking coding by practitioners and provide nearly immediate scores, effectively making it easier for children with developmental delays to receive timely interventions, including early speech and language therapy. 

Ultimately, that could speed up access to interventions that could help children with and without intellectual and developmental disabilities, as early in life as possible.