Language models as tools for investigating the distinction between possible and impossible natural languages

Area of Science:

• Cognitive Science
• Computational Linguistics
• Artificial Intelligence

Background:

• Understanding the principles that define possible human languages is crucial for explaining language acquisition.
• Investigating the inductive biases that guide human language learning remains a significant challenge in cognitive science.

Purpose of the Study:

• To explore the potential of language models (LMs) as tools for distinguishing between possible and impossible natural languages.
• To uncover the inductive biases underlying human language learning by refining LM architectures.

Main Methods:

• Proposing a phased research program involving iterative refinement of LM architectures.
• Utilizing LMs to discriminate between various types of linguistic structures, including those deemed impossible for human languages.

Main Results:

• Demonstrating that language models can be effectively employed to probe the boundaries of natural language.
• Establishing a framework for linking computational findings from LMs to hypotheses about human cognitive processes.

Conclusions:

• Language models offer a promising avenue for empirical investigation into the nature of human language and learning.
• The proposed research program provides a pathway to bridge computational linguistics and cognitive science, enhancing our understanding of language acquisition.