Understanding Before Syntax: Reframing Innateness in Language and AI

Recent advances in large language models (LLMs) have reopened foundational questions about the nature of language and the locus of innateness. Classical accounts, most prominently those associated with Chomsky's Universal Grammar and its minimalist successors, posit an innate syntactic structure as the basis of language acquisition. However, the ability of LLMs to generate fluent language without an explicitly encoded syntactic module challenges this assumption and invites a theoretical reorientation. This paper advances the thesis that what is innate in humans is not syntax, but a more fundamental faculty of understanding — defined here as the capacity to map signals to intentions within a shared context, enabling appropriate inference and response. We propose a three-layer model in which (1) understanding, rooted in social-intentional cognition, constitutes the innate core; (2) communication emerges as a pre-rational, embodied process involving gesture, gaze, affect, and contextual coordination; and (3) syntax arises as a rational stabilization of these interactions. We further introduce a critical distinction between genuine understanding — grounded in embodied perception, affective salience, and social-intentional cognition — and functional understanding, which LLMs achieve through large-scale pattern matching. Syntax is thus an externalizable crystallization of understanding, not its engine. Empirical anchors from developmental psychology, human interaction, and AI benchmarks support this hierarchy. Pre-linguistic infants demonstrate joint attention and intentional attribution prior to any productive syntax; everyday communication routinely relies on non-syntactic cues; and LLMs exhibit systematic limitations on reformulation-resistant Theory of Mind benchmarks — the signature of functional but not genuine understanding. These findings suggest that while syntax is externalizable and reproducible, genuine understanding remains grounded in embodied, relational, and context-sensitive processes. This reframing relocates the innateness debate from syntactic structure to the capacity for intentionality, with significant implications for both cognitive science and the future of artificial intelligence.