Connecting Preschoolers' Spontaneous Speech Patterns to Future Language Skills: A 3-Year Concurrent and Longitudinal...

Understanding how children's spontaneous language behavior relates to standardized metrics of speech and language remains a crucial challenge in studies of language acquisition and clinical assessment. Traditional lab- and clinic-based paradigms are time- and resource-intensive and may not fully capture a child's underlying language competencies. This study aimed to investigate whether automated analysis of naturalistic, child-centered audio recordings can index the developmental trajectory of speech-language abilities from 2 to 5 years of age.

A longitudinal design was employed with N = 155 preschoolers, who were followed from 2 to 5 years of age. Deep learning speech classification methods were used to compute canonical proportion-the proportion of a child's speech that is produced as canonical syllables (well-formed syllables with a consonant-vowel transition, such as "ba"), a key marker of speech motor control development and phonological representation building-from child-centered audio recordings collected at 2-4 years of age. Standardized lab-based assessments of speech and language development were administered concurrently and 1 year later at 2-5 years of age.

Canonical proportion measures significantly predicted multiple dimensions of later speech-language development. The strongest longitudinal associations were observed for consonant articulation skill and vocabulary size. Weaker, although still significant, relationships were found for metaphonological skills, including phonological awareness and phonological working memory.

Findings suggest that children's spontaneous, everyday speech production patterns provide moderate indices of subsequent language development across multiple domains. Automated measures such as canonical proportion hold promise for expanding and diversifying approaches to studying language acquisition and for complementing traditional lab-based assessments.