Adding Types, But Not Tokens, Affects Property Induction.

Area of Science:

• Cognitive Psychology
• Machine Learning
• Behavioral Science

Background:

• Property induction relies on sample composition, but prior research overlooked redundant tokens.
• Real-world evidence often includes repeated entities (tokens), unlike previous experimental samples.

Purpose of the Study:

• Investigate how adding novel types and tokens to a sample affects property generalization.
• To understand the differential impact of types versus tokens on inductive reasoning.

Main Methods:

• Experiments 1-3: Participants generalized novel properties from samples of birds/flowers with varying types and tokens.
• Experiment 4: Assessed recognition and generalization from repeated tokens.
• Modified a Bayesian model of induction to incorporate type and token effects.

Main Results:

• Increasing sample types tightened generalization, favoring highly similar stimuli.
• Increasing sample tokens had little effect on property generalization.
• Repeated tokens aided recognition but were weighted less in generalization inference.

Conclusions:

• Sample composition, specifically the ratio of types to tokens, significantly influences property generalization.
• The study provides a refined computational model accounting for type-token dynamics in induction.