Perceptual estimation obeys Occam's razor.

Area of Science:

• Cognitive Psychology
• Computational Neuroscience
• Perception

Background:

• Unsupervised category learning models suggest humans create few categories (Occam's razor).
• This principle is rooted in statistical inference and aims for simpler explanations.
• Category formation may influence perceptual estimation.

Purpose of the Study:

• To investigate if category invention affects perceptual estimation.
• To test if humans favor simpler perceptual categories even when sensory data suggests otherwise.

Main Methods:

• Participants estimated the number of colored circles on a screen.
• Circle colors were drawn from distributions with varying degrees of overlap.
• A rational analysis using Bayesian computations was performed.

Main Results:

• When color distributions overlapped significantly, estimates were biased towards intermediate values, indicating category fusion.
• This suggests participants ignored color and formed a single perceptual category.
• Reduced overlap led to less bias, showing sensory evidence can override simplicity bias.

Conclusions:

• Human perception favors simpler explanations, grouping stimuli into fewer categories.
• The degree of perceptual overlap influences whether simplicity or sensory evidence dominates category formation.
• Bayesian computations can explain these observed patterns in category-driven perception.