In a fractured world, a shared language is medicine. AUMA carries the 150-year dream forward — no nation's tongue, no empire's imprint. And it goes further: language is programming. Every assertion in AUMA is a typed claim. Every word declares how you know.
Every system of thought runs on a language. The language shapes what can be expressed — and what cannot. When AI inherited English, it inherited English's ambiguity: memory, hallucination, citation, guess all look identical in the output. That is not a model failure. It is a language failure.
AUMA is the fix at the grammar level. Every claim must declare its source. Evidentiality — the linguistic feature that marks how a speaker knows what they know — is a required marker on every assertion, not an optional politeness. You cannot speak AUMA without saying how you know. Hallucination becomes a syntax error.
And because KIRA, PALADIN, and AUMLOK all produce structured AUMA output, the grammar does real work across the entire stack. KIRA retrievals come back tagged with their source. PALADIN's risk classifications feed certainty markers. AUMLOK identity claims wrap in their own evidential scope. The language is the type system — not by design choice, but because it's the language everything runs in.
In the late 1800s, a Polish ophthalmologist named Ludwik Zamenhof published a constructed language called Esperanto. His thesis was simple: if humans could meet on neutral linguistic ground — a language belonging to no nation, no empire, no native speaker — they might learn to live together in peace.
Esperanto reached two million speakers. Then the 20th century happened. World wars, colonial languages, then English as lingua franca. The dream stalled, but the math was sound: people who learned Esperanto first learned subsequent languages 30–50% faster. The cognitive case held even after the political case lost.
AUMA picks up where Zamenhof left off — with 138 years of additional linguistic theory, modern computational tooling, and a new motivation: artificial intelligence needs a language that was designed to be honest. Natural languages weren't. Esperanto wasn't either. AUMA is.
The dream lives on.
Natural languages carry the noise of their history — irregular verbs, gendered nouns, exceptions stacked on exceptions. AUMA was designed in one pass, optimized for clarity, learnability, and machine-interpretability.
nuna
(I see), odi (I hear), padi
(I deduce), intu (I sense). You literally
cannot make an unsourced claim. The grammar enforces
what English politeness fails to.
23 is du-des-tri — literally
"two-ten-three." Numbers expose their place value as
you speak them. Learn 1–100 in twenty minutes. Math
feels like a language, not a separate subject.
Some natural languages — Tibetan, Quechua, Cherokee — require grammatical evidentiality. The speaker must mark, with a particle on the verb, how they know what they're saying. AUMA takes this further: every assertion in the language carries an evidence particle, and the particles are machine-readable.
When AUMA the model is trained on AUMA the language, those
markers do real type-system work. A retrieved memory
comes back tagged nuna. An inference comes back
tagged padi. A user assertion gets tagged
odi. The model cannot produce a claim without
specifying its provenance — not because we asked nicely,
but because the grammar refuses to compose otherwise.
This is the technical answer to the hallucination problem: make hallucination a syntax error.
AUMA is learned in three twenty-eight-day phases — 84 days total. Total regularity throughout: every rule you learn on day one holds without exception on day eighty-four.
Learning AUMA is one of the fastest cognitive upgrades you can give your brain. The radical regularity alone rewires how you process structure and meaning — people who learn a logical constructed language first acquire subsequent natural languages up to 50% faster. Your brain gets better at language in general.
But start small. One friend. Two people who share a vocabulary that belongs to no one else. That is already something — a private shorthand, a bond, a secret grammar for your partnership. From two it grows to a small group: a study circle, a family, a team. At that scale AUMA becomes social and alive and yours.
This is how every living language has ever spread. Word of mouth. Person to person. The dream that Zamenhof had in 1887 — that humans could meet on neutral ground — does not require a critical mass to begin. It requires two people who decide to try.
The language is open source. The dictionary, grammar specification, and evidentiality particle system are all on GitHub. The AI fluent in it is in beta at Auma.one.