Predicting developmental language disorders using artificial intelligence and a speech data analysis tool

Developmental Language Disorder (DLD) affects children's comprehension and production of spoken language without any known biomedical condition. The importance of early identification of DLD is widely acknowledged. Several studies have explored DLD predictors to identify children needing further diagnostic investigation. Most of these measures might be problematic for young children and bilingual children. Based on literature reporting fragile rhythmic abilities in children with DLD, in our study, we followed a different approach. We explored how non-linguistic measures of rhythmic anticipation can be gathered by means of advanced information technology and used to identify children at risk of DLD. With this aim, we developed MARS, a web-based tool to collect such data in a playful way and to analyze them using Machine Learning. MARS engages children in rhythmic babbling exercises, records their vocal productions, and analyzes the recordings. We discuss the methodological rationale of MARS and its underlying technology, and we describe a preliminary study with N = 47 children with and without DLD. The analysis of the audio features of participants' rhythmic vocal productions highlights different patterns in the two groups. This result, although preliminary, suggests that MARS could be a valuable tool for early DLD assessment.