A computational theory of child overextension

Area of Science:

• Developmental Psychology
• Computational Linguistics
• Cognitive Science

Background:

• Overextension, where children use words broadly, is key to early language development.
• Existing research lacks a unified formal explanation for this common childhood linguistic phenomenon.

Purpose of the Study:

• To develop a computational framework explaining children's word overextension.
• To model overextension using probabilistic inference across semantic, neural, and psychological data.

Main Methods:

• Developed a computational framework integrating lexical semantics, deep neural networks, and psychological data.
• Operationalized overextension as probabilistic inference within a conceptual space.
• Tested the framework against child speech data and experimental findings on production/comprehension asymmetry.

Main Results:

• The framework successfully explains overextension patterns in young children across various noun-referent pairs.
• It accurately predicts the observed differences between children's overextended word production and comprehension.
• The model demonstrates a unified approach to understanding word meaning extension.

Conclusions:

• A computational theory for word meaning extension in early childhood has been established.
• The findings support a single-system model for both language production and comprehension in children.
• This framework provides a formal account for a critical aspect of early lexical innovation.