Large Language Models Demonstrate the Potential of Statistical Learning in Language.

Area of Science:

• Cognitive Science
• Computational Linguistics
• Artificial Intelligence

Background:

• The debate on language acquisition degree from linguistic input alone has persisted for centuries.
• Computational modeling of language has been limited by the complexity of human language, restricting studies to small linguistic fragments.

Purpose of the Study:

• To investigate the extent to which language can be acquired solely from linguistic data.
• To leverage Large Language Models (LLMs) as computational tools for empirical research on language acquisition.

Main Methods:

• Utilizing advanced Large Language Models (LLMs), which are deep learning architectures trained on extensive natural language datasets.
• Analyzing the capabilities of LLMs in performing a wide range of linguistic tasks.

Main Results:

• LLMs have shown the capacity to acquire human-like grammatical language without relying on a pre-existing, built-in grammar.
• These models demonstrate significant linguistic task performance, despite current semantic and pragmatic limitations.

Conclusions:

• LLMs provide robust computational models for empirically testing the role of statistical learning in language acquisition.
• The findings suggest that a substantial portion of human language ability may be derivable from linguistic experience alone.