The instantiation principle in natural categories

Area of Science:

• Cognitive Psychology
• Computational Linguistics
• Artificial Intelligence

Background:

• The instantiation principle posits that category representations contain detailed information about specific instances.
• Many categorization theories, like classical and prototype theories, filter out exemplar-level data.
• The instantiation principle can be integrated into both exemplar and abstraction models.

Purpose of the Study:

• To empirically assess the validity of the instantiation principle in human categorization.
• To investigate whether detailed instance information is retained in category representations.
• To evaluate a computational model's ability to predict typicality judgments based on instantiation.

Main Methods:

• A parameter-free exemplar-based model was developed to simulate the instantiation principle.
• The model was applied to typicality judgments for 16 simple and 14 complex categories across four superordinates.
• Three studies were conducted to analyze mean typicality, standard deviations, and distribution skew.

Main Results:

• The exemplar-based model accurately predicted mean typicality judgments (correlations > 0.9).
• The model showed good performance in predicting standard deviations (fits 0.6–0.9).
• Prediction of typicality distribution skew was successful (fit 0.87), and removing atypical exemplars harmed prediction.

Conclusions:

• Results strongly support the instantiation principle.
• Human category representations appear to incorporate detailed information about specific instances.
• The findings challenge theories that assume filtering of exemplar-level data in categorization.