Goldilocks Forgetting in Cross-Situational Learning.

Area of Science:

• Cognitive Science
• Developmental Psychology
• Computational Linguistics

Background:

• Word learning involves navigating referential uncertainty (noise).
• The precise role of forgetting in mitigating this noise is not fully understood.
• Existing models often focus on cognitive biases or developmental changes.

Purpose of the Study:

• To investigate the extent to which forgetting can act as a filter for referential noise in word learning.
• To identify an optimal balance of memory storage and loss for effective learning.
• To explore the mechanistic underpinnings of forgetting's role in language acquisition.

Main Methods:

• Utilized a computational Cross Situational Learning (CSL) model.
• Simulated word learning with varying degrees of forgetting (store-loss ratio).
• Analyzed the model's performance in terms of error reduction and learning outcomes.

Main Results:

• A U-shaped function of errors was observed, indicating an optimal "Goldilocks" zone for forgetting.
• Forgetting functioned as a high-pass filter, reducing referential ambiguity noise.
• The model demonstrated improved learning without specific cognitive biases or developmental shifts.

Conclusions:

• Forgetting, within an optimal range, significantly aids word learning by filtering noise.
• Memory limitations and linguistic experience, rather than specific biases, drive this beneficial effect.
• Integrating domain-general cognitive processes like memory is crucial for understanding language acquisition.