Three ideal observer models for rule learning in simple languages.

Area of Science:

• Cognitive Science
• Developmental Psychology
• Computational Linguistics

Background:

• Children generalize language rules beyond direct exposure.
• Rule learning in infancy is a key model for understanding generalization.
• Existing research shows performance differences in rule learning across age groups and rule types.

Purpose of the Study:

• To model the representational and inferential assumptions underlying generalization in rule learning.
• To develop ideal observer models that capture diverse experimental findings in infant and adult rule learning.
• To provide a computational baseline for future investigations into generalization mechanisms.

Main Methods:

• Introduction of three progressively complex ideal observer models for rule learning.
• Model 1: Learns a single rule.
• Model 2: Learns a single rule from noisy input.
• Model 3: Learns multiple rules from noisy input.

Main Results:

• The ideal observer models successfully capture a wide range of experimental results in rule learning.
• The models account for findings previously interpreted as evidence for domain-specificity or learning limitations.
• The computational approach provides a unified framework for understanding generalization across different conditions.

Conclusions:

• Ideal observer models offer a powerful framework for understanding generalization in rule learning.
• These models suggest that observed differences in learning may stem from assumptions about input noise and rule complexity.
• The proposed models serve as a valuable baseline for future research on cognitive generalization in infants and adults.