Linguistic Rule Generalisation Creates the Same Distributional Structure That Feeds It.

Area of Science:

• Cognitive Science
• Linguistics
• Computational Linguistics

Background:

• Language expressivity arises from applying rules to new items.
• Linguistic rules vary in their generalizability.
• Frequency distributions influence rule generalization.

Purpose of the Study:

• Investigate how frequency distribution properties affect linguistic rule generalization.
• Determine the specific properties influencing generalization behavior.
• Model learners' generalization strategies.

Main Methods:

• Artificial language learning experiment with adult participants.
• Analysis of suffix generalization based on input frequency distributions.
• Urn model simulation to test generalization explanations.
• Bayesian inference modeling of learner behavior.

Main Results:

• Learners exhibit probability-matching behavior.
• A significant preference for suffixes associated with more low-frequency stems was observed.
• Traditional explanations focusing on type count or skew failed to predict behavior.
• The low-frequency preference accurately predicted participants' generalization patterns.

Conclusions:

• Linguistic rule generalization is influenced by a preference for low-frequency stems.
• This preference, not just type count or skew, drives generalization.
• Rule generalization is a self-sustaining process, feeding back into the distributional structure of language.