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

TL;DR

This paper proposes using language models as investigative tools to distinguish between possible and impossible natural languages, aiming to uncover the inductive biases underlying human language learning.

Contribution

It introduces a phased research approach to iteratively refine LM architectures for better discrimination of language possibilities, linking to human cognition.

Findings

LMs can potentially differentiate possible from impossible languages

Refined LM architectures may reveal inductive biases in language learning

Supports linking hypotheses to human cognition

Abstract

We argue that language models (LMs) have strong potential as investigative tools for probing the distinction between possible and impossible natural languages and thus uncovering the inductive biases that support human language learning. We outline a phased research program in which LM architectures are iteratively refined to better discriminate between possible and impossible languages, supporting linking hypotheses to human cognition.