Exploring a corpus-based approach for detecting language impairment in monolingual English-speaking children

Area of Science:

• Computational linguistics
• Developmental psychology
• Machine learning

Background:

• Language impairment affects a significant number of children.
• Accurate and early detection is crucial for effective intervention.
• Traditional methods for language assessment can be time-consuming and subjective.

Purpose of the Study:

• To develop and evaluate an automated method for predicting language impairment in children using narrative analysis.
• To leverage corpus-based approaches from natural language processing (NLP) and machine learning (ML).

Main Methods:

• Extraction of diverse linguistic features from narrative samples.
• Training of language models and machine learning algorithms (e.g., predictors).
• Evaluation on two datasets comprising spontaneous narratives from 118 older children (13 years) and play sessions from over 600 younger children (6 years).

Main Results:

• The automated method significantly outperformed a baseline approach.
• Achieved high prediction accuracy with F-measures exceeding 85% on two language tasks.
• Demonstrated a 48% F-measure on a third language task.

Conclusions:

• Corpus-based NLP and ML methods are effective for detecting language impairments in children's narratives.
• These findings support further research into NLP for communication disorder assessment.
• The framework is adaptable for analyzing narratives in languages other than English.