Shortening the MacArthur-Bates Communicative Developmental Inventory Using Machine Learning Based Computerized...

Area of Science:

• Developmental Psychology
• Computational Linguistics
• Machine Learning

Background:

• Early identification of developmental disorders in infants and toddlers is crucial for effective intervention and cost reduction.
• The MacArthur-Bates Communicative Development Inventory (MB-CDI) is a standard tool for assessing early lexical development but is lengthy and time-consuming.
• Existing assessment methods face limitations in efficiency and adaptability.

Purpose of the Study:

• To develop a shortened version of the MB-CDI using machine learning and computerized adaptive testing (ML-CAT).
• To evaluate the accuracy and efficiency of the ML-CAT in predicting lexical development scores.
• To compare the performance of ML-CAT against existing methods, including Item Response Theory (IRT).

Main Methods:

• Utilized a machine learning approach combined with computerized adaptive testing (ML-CAT).
• Adapted the sequence of words presented based on individual subject responses to shorten the assessment.
• Validated the ML-CAT across five different languages.

Main Results:

• ML-CAT reliably predicts H-MB-CDI scores using an average of just 10 words with 94%-96% accuracy.
• ML-CAT demonstrated superior performance compared to non-adaptive methods and IRT-based models.
• ML-CAT showed improved accuracy in handling atypical talkers (outliers) compared to IRT methods.

Conclusions:

• ML-CAT offers a more efficient and reliable method for assessing lexical development in infants and toddlers.
• The shorter assessment length reduces the burden on subjects and caregivers, potentially increasing reliability.
• ML-CAT facilitates wider and more frequent community-based screening for developmental disorders.