Key points are not available for this paper at this time.
Purpose: Phonological production impairments are prevalent in children with developmental language disorder (DLD) and hearing impairment (HI). This study aims to quantify and compare phonological errors in Swedish-speaking children using a novel automated assessment tool and provide an automatic machine learning classification algorithm of children with DLD and HI to age-matched controls based on phonological errors. Methods: 72 Swedish-speaking children (29 with DLD, 14 with HI, and 29 typically developing) participated. Phonological production was elicited using a 72-item confrontation naming task. A novel tool was developed to calculate a composite phonological error score and specific scores for different phonological errors (deletions, insertions, substitutions, and transpositions) from written speech productions. This tool leverages the International Phonetic Alphabet (IPA) and a form of the Normalized Damerau-Levenshtein Distance for accurate error analysis. Results: The composite score successfully differentiated between children with DLD and typically developing children, highlighting its sensitivity in detecting phonological impairment. Machine learning models can accurately differentiate between children with and without language disorders. However, children with DLD and HI differed in the phonemic deletion errors, which suggests that their phonemic production is relatively similar. Conclusions: Children with DLD and HI exhibit significantly higher phonological error rates compared to typically developing peers. Children with HI and DLC are comparably impaired in phonology (as manifested by the compositive phonological score). These findings highlight the potential of machine learning for early identification and targeted intervention in language disorders, improving outcomes for affected children and demonstrated the potential of a multilingual tool for scoring phonological errors.
Sundström et al. (Wed,) studied this question.