THE 'GENERATIVE-AI CHALLENGE': RE-EVALUATING CORE LINGUISTIC THEORIES IN THE AGE OF LARGE LANGUAGE MODELS

The advent of high-capacity Large Language Models (LLMs) like GPT-4 presents a profound challenge to contemporary linguistic theory. These models, trained on vast textual corpora, demonstrate a remarkable capacity for generating syntactically coherent and contextually appropriate human language. This proficiency appears to emerge without the explicit encoding of innate grammatical structures or deep semantic grounding, principles long held as foundational to human linguistic competence, particularly within the Chomskyan framework. This article explores the critical tension between the statistical, performance-based nature of LLMs and the competenceoriented, rule-based assumptions of generative grammar. We argue that while LLMs do not invalidate core linguistic theories, they necessitate a critical re-evaluation of the evidence used to support them, particularly regarding the 'Poverty of the Stimulus' argument. This paper analyzes the implications of LLM success for our understanding of syntax, semantics, and the fundamental nature of linguistic knowledge.