Creating a diagnostic assessment model for autism spectrum disorder by differentiating lexicogrammatical choices...

Area of Science:

• Linguistics
• Psychology
• Computer Science

Background:

• Differentiating autism spectrum (AS) from non-AS conditions in adolescents and adults presents challenges due to AS heterogeneity and diagnostic tool limitations.
• Traditional diagnostic tools like the ADOS-2 have limitations in capturing the full spectrum of AS presentations.
• A multidimensional diagnostic approach is needed to improve accuracy.

Purpose of the Study:

• To explore the utility of lexicogrammatical analysis in differentiating AS from non-AS conditions in adolescents and adults.
• To develop and evaluate machine learning-based diagnostic models utilizing linguistic features.
• To compare the diagnostic effectiveness of interview and story-recounting spoken language tasks.

Main Methods:

• Collected spoken language data (interviews, story-recounting) from 64 individuals with AS and 71 non-AS individuals (aged 14+).
• Applied machine learning techniques to analyze lexicogrammatical choices in the collected texts.
• Developed two diagnostic models: one based on linguistic tags and another combining tags with textual analysis.

Main Results:

• The combined model achieved high diagnostic effectiveness: 80% accuracy, 82% precision, 73% sensitivity, and 87% specificity.
• Interview-based texts proved more diagnostically effective than story-recounting texts.
• Lexicogrammatical analysis identified distinctive linguistic patterns in individuals with AS, reflecting altered social language use.

Conclusions:

• Lexicogrammatical analysis is a promising adjunct to traditional methods for diagnosing AS in adolescents and adults.
• Natural language processing can detect linguistic patterns to enhance diagnostic accuracy for AS.
• Altered social language use is a key differentiator between AS and non-AS conditions.