The Experiential–Factual Divide: A Universal Geometric Property of Human Language, AI Models, and Brain Organisation

We present converging evidence from three independent methodologies that human language encodes a fundamental geometric partition between experiential concepts (pain, love, death, memory, identity, divinity, self-referential processing) and factual concepts (mathematics, geography, physics, history, chemistry). Study 1: Static word embeddings (GloVe + fastText) across 7 typologically diverse languages (EN, UK, ZH, ES, FR, DE, JA). Cohen's d = 0.65–1.12, all p < 0.001. Concreteness confound controlled. Study 2: 10/10 large language models from 9 organisations universally replicate the geometric partition via Grassmann subspace distance. Present in BASE models before RLHF. Absent in randomly initialised networks. Study 3: Real fMRI validation — Brainomics Localizer dataset (n=20, Yeo 7-network atlas). Sentence reading vs. calculation: t(19) = 3.56, p = 0.002. Direction matches LLM geometric predictions. The experiential–factual divide is a universal property of human communicative cognition, inherited by language models from training data, and reflected in the functional architecture of the human brain. Part of the DSAOP series. All data and replication code included.