Simplicity and informativeness in semantic category systems.

Area of Science:

• Cognitive Science
• Linguistics
• Psychology

Background:

• Semantic category systems, like color and kinship terms, exhibit an optimal balance between simplicity and informativeness.
• This balance is thought to arise from competing pressures: simplicity from learning and informativeness from communication.
• An alternative hypothesis suggests learning itself might favor informativeness.

Purpose of the Study:

• To investigate competing hypotheses on human inductive bias in semantic categorization.
• To determine whether learners prioritize simplicity or informativeness when acquiring semantic systems.
• To explore the underlying mechanisms of semantic category learning.

Main Methods:

• Formalization of competing hypotheses using a Bayesian iterated learning model.
• Simulation of language emergence under different learning bias assumptions.
• Experimental testing of learner biases in isolation from communicative tasks.

Main Results:

• Strong evidence supports the simplicity account of inductive bias in learning.
• Simplicity principles in learning can mimic a bias for informativeness.
• Learners' biases, when isolated, favor simplicity over informativeness.

Conclusions:

• Semantic categories are likely learned via domain-general principles.
• The observed balance of simplicity and informativeness does not necessitate domain-specific learning mechanisms.
• This research reframes our understanding of how humans acquire and generalize semantic information.