A Japanese LDA model for automatic clustering analysis of semantic verbal fluency tests

Area of Science:

• Computational psycholinguistics
• Neuroscience
• Cognitive psychology

Background:

• Clustering analysis is vital for semantic structure examination in verbal fluency tests (VFTs).
• Computational psycholinguistics enhances reproducibility but lacks language-specific models.
• A validated computational approach for Japanese VFTs is needed.

Purpose of the Study:

• To construct and validate a Japanese Latent Dirichlet Allocation (LDA) model for semantic clustering in VFTs.
• To enable objective quantification of word associations and automatic semantic cluster detection in Japanese.
• To compare LDA-based clustering with traditional manual methods.

Main Methods:

• Developed a Japanese Latent Dirichlet Allocation (LDA) model.
• Conducted semantic VFTs with healthy young Japanese adults.
• Performed clustering analyses using both the LDA model and manual human coding.

Main Results:

• The LDA model successfully identified semantic clusters, comparable to human coders.
• Response intervals within identified clusters were significantly shorter than those outside clusters.
• LDA-based clustering identified larger clusters, suggesting detection of subtle semantic relationships missed by humans.

Conclusions:

• The Japanese LDA model is a valid tool for semantic clustering in VFTs.
• The model enhances objective quantification and automatic detection of semantic relationships in Japanese.
• This work promotes the use of computational linguistic approaches in Japanese VFT research.