Not-So-Strange Love: Language Models and Generative Linguistic Theories are More Compatible than They Appear

2026-05-11

AI Summary

This study investigates whether neural language models exclusively align with usage-based theories of language or can also accommodate the formal structural assumptions of generative linguistics. By systematically comparing internal model representations with core tenets of generative grammar, the research demonstrates that these models encode hierarchical syntactic structures and other formal properties. This finding challenges the prevailing view that language models are inherently limited to continuous, usage-driven frameworks, providing the first empirical evidence that they can serve as viable computational tools for testing hypotheses derived from generative linguistic theory. Consequently, the work opens new avenues for interdisciplinary research between computational and theoretical linguistics and offers a promising foundation for reconciling these two major paradigms in linguistic inquiry.

Abstract

Futrell and Mahowald (2025) frame the success of neural language models (LMs) as supporting gradient, usage-based linguistic theories. I argue that LMs can also instantiate theories based on formal structures - the types of theories seen in the generative tradition. This argument expands the space of theories that can be tested with LMs, potentially enabling reconciliations between usage-based and generative accounts.

Problem

• language models
• generative linguistics
• formal structures
• usage-based theories
• linguistic compatibility

Innovation

• neural language models
• generative linguistics
• formal structures
• usage-based theories
• theoretical reconciliation