The Missing Half of Language Learning in Current Developmental Language Models: Exogenous and Endogenous Linguistic...

Area of Science:

• Cognitive Science
• Artificial Intelligence
• Developmental Psychology

Background:

• Current developmental language models (DLMs) primarily focus on exogenous linguistic input.
• Child language acquisition is significantly influenced by endogenous processes, which are often overlooked in DLMs.
• Endogenous processes include non-linguistic perception co-option, private speech, and neural replay during sleep.

Purpose of the Study:

• To propose a more comprehensive framework for DLMs that accounts for both exogenous and endogenous linguistic input.
• To bridge the gap between artificial and human language learning by integrating key aspects of child language acquisition.
• To redefine 'linguistic exposure' to include internal cognitive and neural mechanisms.

Main Methods:

• Conceptual framework integrating exogenous and endogenous linguistic input.
• Proposal for DLM enhancement through label feedback, self-generated speech, and sleep-like consolidation.
• Emphasis on interdisciplinary collaboration for empirical grounding.

Main Results:

• Current DLMs do not replicate the amplification and refinement of language through endogenous processes observed in children.
• Integrating endogenous factors can significantly improve the efficiency and naturalness of DLMs.
• A redefined 'linguistic exposure' is crucial for developing more human-like language acquisition models.

Conclusions:

• DLMs need to incorporate endogenous processes (inner speech, sleep replay, cognitive co-option) to truly mirror child language acquisition.
• Future DLMs should integrate label feedback, self-generated speech, and sleep consolidation for enhanced learning.
• Interdisciplinary research is vital to develop DLMs that reflect the complexity of human language development.