AUTOMATIC SCORING OF A NONWORD REPETITION TEST

Area of Science:

• Speech-language pathology
• Computational linguistics
• Developmental psychology

Background:

• Nonword repetition (NWR) tests are crucial for identifying language impairments.
• Manual scoring of NWR tests by speech-language pathologists is time-consuming and subjective.
• Automated evaluation methods are needed to improve efficiency and objectivity.

Purpose of the Study:

• To investigate the feasibility of using automated speech-based techniques to evaluate NWR tests.
• To develop and validate a machine learning model for predicting NWR test scores.

Main Methods:

• Utilized Automatic Speech Recognition (ASR) to transcribe children's verbal responses to nonword stimuli.
• Applied machine learning algorithms to ASR output to predict gold-standard scores.
• Evaluated the model on a dataset of 101 children, including those with autism spectrum disorders (ASD), specific language impairment (SLI), and typically developing (TD) children.

Main Results:

• The proposed automated approach demonstrated significant success in predicting NWR test scores.
• Achieved an averaged product-moment correlation of 0.74 between observed and predicted scores.
• Reported a mean absolute error of 0.06 on scores ranging from 0.34 to 0.97.

Conclusions:

• Automated speech-based evaluation of NWR tests is a feasible and effective approach.
• This technology has the potential to assist clinicians in diagnosing language impairments more efficiently.
• Further research can refine these methods for broader clinical application.