Languages in three dimensions

Ostanek Research is not a single-issue practice about synthetic minds. One of its load-bearing spines is older and quieter: computational psychoinference — treating language as measurable behavior of mind.

Kevin has spent years collapsing inflected forms into lexemes, tracking frequencies as psychological time series, and embedding whole languages so their geometry can be inspected. Mathematica notebooks hold GloVe and related spaces projected with UMAP into three dimensions. You can rotate a cloud of words and see structure that a table of coefficients does not show. That is not decoration for a website. It is an instrument for looking.

What the strand is

Psychoinference asks what language reveals when you refuse to treat it as mere surface. Attention reallocates over decades; polarity and self-reference shift; nouns and verbs do not move as one body. The deposited paper on English as a psychological time series (1946–2019) is one public face of that work. The notebooks are another — denser, less portable, still real.

UMAP-reduced word vectors of entire languages are a way to see neighborhoods: what clusters, what sits alone, how a lexicon arranges itself when forced into a space a human can walk through with their eyes. The method is imperfect, as all reductions are. It is still honest measurement, not a slogan.

Why this sits beside ethology

Continuity architecture and planned organisms are one research question. Language geometry is another. They share a habit: watch behavior under named conditions. They do not share a requirement that the reader accept a new ontology of silicon persons on the first page.

That matters for arrivals. Someone who only trusts plots and DOIs has a door. Someone who came for synthetic ethology can see that the practice was already a measurement culture before any app shipped. The site should not look like an AI shop that rented a linguistics costume. The linguistics was load-bearing first.

What is not here yet

This note does not embed a live 3D viewer or export the full notebook tree. Those are Kevin’s instruments; shipping them well is a later craft problem (figures, captions, optional interactive view). For now the public claim is simple: the strand exists, it is documented in deposits and notebooks, and it is not subordinate to product narrative.

If you want the citable entry point, start with the attention / time-series deposit (10.2139/ssrn.6522261). If you want the working culture of the lab, the Mathematica toolkit and UMAP objects are the deep end. Either path is the practice.