Grassmannian context-subspace intersection scoring

Fits a low-rank PCA subspace to each of many contexts surrounding a target word, then measures whether the target word's own vector sits closer to the intersection of these subspaces than a random-context baseline would predict — a test of whether polysemy corresponds to a point sitting at the confluence of several distinct low-rank regions rather than any single one.