Grammatical Form as Ontological Affordance Representational Constraints on Process-Relational Knowledge in Large Language Models

Large Language Models trained predominantly on English-language corpora represent process-relational knowledge systems — Āyurveda, Madhyamaka, Traditional Chinese Medicine — through a persistent substance-ontological default. This paper asks what that default consists in and where it resides. It argues that natural languages differ in which relational structures their grammars encode explicitly and at low cost, and that this difference is a matter of affordance rather than determination: no grammar prevents any ontology from being expressed, but grammars differ in what they make expressible cheaply and by default. The claim concerns logico-cognitive form and not doctrinal content. It is therefore not refuted by the observation that Vaiśeṣika articulated a substance ontology in Sanskrit; on the contrary, the Vaiśeṣika dravya is individuated operationally through inherence (samavāya) and quality-bearing, which is not the self-subsistent ousia of the Aristotelian tradition. A thesis about form that treats every articulation as confirming evidence would, however, be unfalsifiable, and §4 accordingly supplies two criteria of refutation: one internal to the Indian grammatical tradition (the Śākaṭāyana–Gārgya dispute over deverbal derivation) and one contemporary and measurable (structural analysis of clause architecture).Empirically, the position is anchored in SDI-GOC v3.1 [1]. Across matched neutral and relationally framed conditions, lexical polarity differs by +0.605 and exploratory structural polarity by +0.038; the interaction is +0.566 (95% CI [+0.459, +0.686]). Structural scores remain substance-dominant throughout. This output-level dissociation operationalises, but does not by itself prove, the distinction between describing a process-relational framework and constructing clauses within one. The paper does not claim that English syntax causes the default. Interpretability findings that localise English privilege to the output pathway [21], and that identify grammatical concepts as cross-lingually shared features [19,20], are engaged directly; they are compatible with an affordance thesis about output construction and incompatible with a thesis about deep representation. We propose ontological alignment as a fourth category of alignment and a research agenda directed at the layer where the phenomenon is measured.