Grammar as Cognitive Partition

If language evolved to synchronize mental states rather than transmit propositions, we should expect grammar to reflect the structure of mental states, not the structure of the world.

This is exactly what we find.

The Case System as Evidence

Every language with a robust case system partitions sentences along the same axes:

Case	Encodes
Nominative	Agent of action
Accusative	Patient of action
Dative	Recipient / experiencer
Genitive	Possession / origin
Locative	Spatial / conceptual ground

These aren't categories of physics. They are categories of intentional structure: who acts, who receives, what grounds the event.

The hypothesis: grammatical cases are a lossy compression of mental state geometry. Different languages compress differently, but they compress the same underlying space.

Formal Statement

Let $\mathcal{M}$ be the space of mental states, and $\mathcal{G}$ a grammar. Define the semantic map $\phi: \mathcal{M} \to \mathcal{G}$ as the function mapping a mental state to its grammatical encoding.

The PSM claim: $\phi$ is surjective but not injective — every grammatical form corresponds to some mental state, but multiple mental states may collapse to the same form.

This gives the translation problem a precise shape:

$$T_{AB} = \phi_B^{-1} \circ \phi_A$$

is not a function — it is a relation. Translation is choosing among preimages.

Empirical Predictions

This framing makes testable predictions:

1. All languages should partition the Agent/Patient axis — all minds model agency, so all grammars must encode it.
2. When social structures change (shifting gender roles, for instance), grammatical gender drifts. Drift follows cognition.
3. Languages with similar case systems should be easier to learn for speakers of each, even controlling for lexical overlap.

The third prediction is supported by WALS data and partially explains the Anglophone language-learning disadvantage.