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

2025-12-10

AI Summary

This study investigates language models (LMs) as cognitive probes to uncover the inductive biases underlying human language acquisition—specifically, how learners distinguish “possible natural languages” from “impossible” ones. Method: We propose a staged, iterative optimization of LM architectures, integrating controlled interventions, configurational analysis, and cross-linguistic learnability assessment to systematically model the boundaries of linguistic possibility. Contribution/Results: For the first time, we repurpose large language models as interpretable cognitive tools—not merely generative systems—demonstrating that optimized LMs reliably discriminate grammatically well-formed yet unlearnable linguistic structures. Empirical results provide computationally grounded, empirically testable evidence for linguistic nativism, bridging formal computational modeling with theories of universal grammar. This advances the scientific study of language by rendering abstract principles of linguistic universals operational, measurable, and falsifiable.

Technology Category

• Natural Language Processing: (Large) Language Models
• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: (Computational) Cognitive Architectures

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

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.

Problem

• Language models investigate possible vs impossible natural languages
• They uncover inductive biases in human language learning
• Refine models to better discriminate and link to cognition

Innovation

• Language models probe possible vs impossible languages
• Iterative refinement of LM architectures for discrimination
• Linking LM biases to human language learning