Word learning as category formation.

Area of Science:

• Cognitive Science
• Developmental Psychology
• Computational Linguistics

Background:

• Understanding how children generalize word meanings from limited examples is a core challenge in word learning.
• The
• suspicious coincidence effect
• (SCE) suggests more training data leads to narrower word meanings, often explained by statistical inference models.

Purpose of the Study:

• To investigate the independent effects of training object number and presentation timing on word generalization.
• To develop a unified computational model explaining word learning phenomena.
• To test the proposed model against human behavioral data.

Main Methods:

• Reanalysis of existing experimental data on word learning.
• Development of the Naïve Generalization Model (NGM), a computational model based on local, incremental category formation.
• Testing the NGM against human behavior across seven experimental conditions varying presentation timing, object number, and item hierarchy.

Main Results:

• Both the number of training objects and their presentation timing independently influence word generalization, contrary to prior assumptions of interaction.
• The Naïve Generalization Model (NGM) successfully accounts for these independent effects within a unified framework.
• Model performance aligns with human behavior across various experimental conditions, both qualitatively and quantitatively.

Conclusions:

• Word generalization is influenced by independent factors of training set size and presentation timing.
• The Naïve Generalization Model provides a parsimonious explanation for word learning, grounded in local category formation.
• Rational word learning behavior may emerge from simple, mechanistic processes rather than complex global statistical inference.