Automatic Assessment of Language Ability in Children with and without Typical Development

Area of Science:

• Speech and Language Pathology
• Computational Linguistics
• Developmental Psychology

Background:

• Expressive language assessment is crucial for identifying developmental disorders.
• Traditional language tests rely on manual scoring by speech-language pathologists, which can be time-consuming.
• Automating language assessment can increase efficiency and accessibility.

Purpose of the Study:

• To develop and validate a fully automated method for expressive language assessment.
• To predict gold standard language test scores using vocal responses.
• To establish a technological foundation for computerized voice-based language assessments.

Main Methods:

• Utilized a sentence repetition task (SRT) to elicit vocal responses from children.
• Developed a test-specific automatic speech recognition (ASR) system for transcription.
• Employed a regression model to predict scores from transcriptions.

Main Results:

• The automated method demonstrated feasibility in predicting gold standard scores.
• Achieved an average mean absolute error of 6.52 on a score range of 0-90.
• Tested on 104 children, including those with neurodevelopmental disorders.

Conclusions:

• Fully automated voice-based scoring is viable for language assessment.
• This technology has significant clinical implications for speech-language pathology.
• The developed system supports the computerization of various voice-based language tests.