Determining the dimensionality in spatial representations of semantic concepts.

Area of Science:

• Cognitive Psychology
• Computational Linguistics
• Data Science

Background:

• Determining optimal dimensionality is crucial for multidimensional scaling (MDS) in semantic concept research.
• Existing dimensionality selection strategies are widely used but may not fully capture semantic richness.

Purpose of the Study:

• To evaluate traditional dimensionality choice criteria for MDS solutions.
• To compare traditional methods with a novel approach based on predicting an external criterion.

Main Methods:

• Two studies were conducted using MDS to represent semantic concepts.
• Typicality of exemplars within concepts was predicted from their distance to concept centroids.
• Model performance was assessed across varying dimensionalities.

Main Results:

• Traditional methods selected low-dimensional solutions.
• Predictions of the external criterion (exemplar typicality) improved with increasing dimensions.
• This improvement continued to dimensionalities significantly higher than conventionally used.

Conclusions:

• Traditional dimensionality selection methods may underestimate the complexity of semantic concepts.
• Higher dimensional spaces better represent the richness of semantic information.
• The proposed external criterion prediction method offers a more sensitive approach to determining MDS dimensionality for semantic data.