BantuNomics Sign in
B BANTUNOMICS
Evaluate UCM
Universal Concord Matrices agreement intelligence for Bantu AI systems
Overview

Public sample · limited taste

Bemba (IciBemba) — 6 of 31 dimensions.

This is a deliberately limited slice: one language (Bemba, the canonical prototype), 6 of the 31 concord dimensions, and only a handful of noun-class rows. Read across a row and you can watch one noun class govern the noun prefix, the subject and object concord, the adjective, and the demonstrative — the agreement cascade in miniature. The remaining 25 dimensions, the other 18 languages, and all machine-readable exports are available under evaluation.

The cascade, in a few columns

One noun class, one row across.

N/A is a linguistic judgment, not missing data. Forms are shown exactly as certified. This is a taste of the 31-dimension matrix — not the matrix.

Class Number NP_Base_Prefix SP_Affirmative Object_Prefix Adjective_Agreement Demonstrative_Proximal + 25 more
1 Singular mu- a- -mu- mu- uyu 🔒
2 Plural / Honorific Sg ba- ba- -ba- ba- aba 🔒
3 Singular mu- u- -u- u- uyu 🔒
4 Plural mi- i- -i- i- iyi 🔒
5 Singular li- li- -li- li- ili 🔒
6 Plural ma- ya- -ya- ya- aya 🔒
7 Singular ci- ci- -ci- ci- ici 🔒
8 Plural fi- fi- -fi- fi- ifi 🔒

What's locked

Everything past this slice.

The full 31-dimension Bemba matrix (augments, negative concords, pronouns, the full demonstrative set, relatives, tone, nominalizer), the remaining 18 published languages, cross-language comparison, and the JSON / CSV / XLSX exports are all behind evaluation and license.

Start your free evaluation

Why it matters

This is your control surface.

Constrain generation to the class that governs a noun; detect agreement-chain errors in translation QA that a text-only model misses; build evals that test whether a model understands the concord cascade rather than word lookup. The full matrix is the machine-readable version of exactly what you see above.

Access

Start where you are.

Prove it free, validate it on your own data, or license the whole program.

01 · Free
Evaluation

Score your model against the foundational layer — the Alphabet Test and the L26 Lite suite, with saved results. Self-serve, no cost.

Start evaluating →
Most labs start here
02 · 75 days
Validation Pilot

Three languages you choose — and everything we hold for them. Measured on your own held-out data.

Scope a pilot →
03 · Program
Full Annual Subscription

Every product, every language, the full consented corpus — and everything curated while you're subscribed.

Start a Full Annual Subscription →