Large Language Models and Quantitative Linguistics: Simulation within the Scientific Framework of Language

Quantitative Linguistics (QL) has long sought to uncover universal patterns in language through mathematical modelling, establishing linguistic laws grounded in the scientific method. A core principle of QL is that language operates under measurable regularities, linguistic laws based on theoretical models that can be empirically tested. However, the emergence of so-called ‘Large Language Models’ (LLMs) has sparked a profound epistemological debate: are they truly language models? LLMs do not originate from explicit linguistic theories, nor do they undergo the formal validation procedures characteristic of QL. Rather, LLMs constitute a powerful simulation technology relying on data-driven optimization, producing emergent linguistic representations without a structured theoretical basis. While LLMs may capture statistical regularities, the question remains whether they truly reflect linguistic laws, or are merely artefacts of high-dimensional interpolation. This work examines the epistemological tension between the hypothetic-deductive approach of QL and the inductive nature of LLMs, arguing that hyperrealistic simulations such as LLMs can be valuable tools for linguistic inquiry, but only within a framework that adheres to scientific rigour. Integrating these perspectives may refine linguistic theory while deepening understanding of the limitations and potential of AI-driven models. This article is a discussion paper aimed at epistemological clarification rather than presenting new empirical findings.