A whole-word computational algorithm for phonological error scoring: Differentiating children with developmental...

Area of Science:

• Speech and Language Pathology
• Computational Linguistics
• Pediatric Audiology

Background:

• Accurate phonological assessment is crucial for managing phonological disorders.
• Current methods involve time-intensive manual transcription and analysis.
• Children with developmental language disorder (DLD) and hearing impairment (HI) are at high risk.

Purpose of the Study:

• Introduce a computational phonological scoring tool.
• Evaluate its ability to identify phonological problems versus typical development (TD).
• Assess its capacity to differentiate between DLD and HI.

Main Methods:

• Developed a computational tool using Normalised Damerau-Levenshtein Distance to compute phoneme-level distances.
• Calculated a Composite Phonological Score (CPS) and specific error counts (deletions, insertions, substitutions, transpositions).
• Utilized machine learning models to analyze data from 72 Swedish-speaking children (29 DLD, 14 HI, 29 TD) in a picture naming task.

Main Results:

• Both DLD and HI groups exhibited higher CPS (more errors) compared to the TD group.
• CPS did not significantly differ between DLD and HI groups.
• Deletions were more frequent in the HI group than in the DLD group.

Conclusions:

• The Composite Phonological Score (CPS) is a valuable primary screening marker for identifying phonological impairments in children.
• Distinguishing between DLD and HI requires detailed error analysis and broader language measures beyond CPS.
• Computational phonological scoring offers richer data than standard correct/incorrect scoring, aiding in screening and differential diagnosis.