Dignity in Your Own Language

By Anushka Appala and Dr. Janio Rosales

A woman in Alta Verapaz walks into a municipal office to ask about a benefit she is owed. She speaks Q'eqchi'. The clerk speaks Spanish. The form is in Spanish. The sign explaining the requirements is in Spanish. She understands almost none of it, and there is no one to translate. She nods, takes the paper she cannot read, and goes home. Nothing about her case was decided against her. She simply could not enter the conversation.

This happens every day, in dozens of languages, across a country where the State speaks — in practice — only one of them. It is not a small inconvenience. It is the difference between a citizen and a supplicant.

Language is not a feature. It is the door.

Most technology treats language as an add-on: build the product, then bolt on a translation layer for the markets that justify the cost. We reject that ordering. For a public platform in Guatemala, the first question is not what does the system do but whom can it reach. If it cannot reach a Q'eqchi'-speaking grandmother in her own words, then for her it does nothing at all.

The numbers make the case plainly. 43.6% of Guatemalans are indigenous — roughly 6.5 million people (INE Census 2018). Guatemala recognizes 25 official languages: 22 Mayan languages plus Xinca, Garífuna, and Spanish (Academia de Lenguas Mayas / Ley de Idiomas Nacionales). These are not fading dialects. The four largest Mayan languages alone are spoken by millions: Q'eqchi' by about 1.3 million people, K'iche' by about 1.1 million, Mam by roughly 600,000, and Kaqchikel by about 410,000 native speakers (2019 census / Ethnologue). A state that serves its people in Spanish only has, by construction, excluded a plurality of them from the first sentence.

So NaciluzIA is multilingual as a first principle, not a later release. The AI understands a request in Q'eqchi', explains a process in Q'eqchi', and drafts the paperwork a Spanish-speaking official will read and sign. The citizen and the State finally meet in a language the citizen actually speaks — and the binding decision still rests, as always, with a named human being.

Why voice comes before the screen

There is a second exclusion beneath the first, and it is easy to miss from a desk in a capital city. Guatemala has more mobile connections than people — mobile subscriptions equal about 110% of the population, some 20.4 million SIMs — yet roughly 39% of the population is offline (DataReportal / GSMA, Digital 2025: Guatemala). Nearly everyone holds a phone; four in ten cannot reliably use the internet on it. Layer literacy and language onto that, and a text-first, form-first, Spanish-first service is a wall with a very small door.

This is why NaciluzIA is voice-first. A person should be able to speak, in their own language, and be understood — not required to type, to read dense instructions, or to navigate a menu built for someone else. Voice meets people where the largest share of them already are: on a phone, speaking, not scrolling. The written record still exists, tamper-evident and auditable. But the way in is a human voice, answered in kind.

The hardest test is the one worth passing

Serving 25 languages well is genuinely difficult, and we will not pretend otherwise. A Mayan language is not Spanish with different words; it carries its own grammar, its own regional variation, its own oral traditions and spelling conventions that no scraped dataset captures cleanly. Getting this right cannot be automated away.

So we do not try to. Every indigenous-language interaction is validated in partnership with the Academia de Lenguas Mayas, the institution charged with the stewardship of these languages. Linguists and native speakers check what the system understands and how it responds, closing the loop between a probabilistic model and the lived reality of a language. The AI proposes; people who speak the language decide whether it got it right.

We choose the hard test on purpose. If a platform can deliver a public service in Q'eqchi' — with limited data, real dialectal variation, and a demanding standard of respect — then it can deliver that service almost anywhere. The most difficult inclusion problem in Guatemala is also the most valuable one to solve for the region, because the same discipline that reaches a grandmother in Alta Verapaz reaches indigenous and minority-language communities across Central America.

Dignity is not a translated afterthought. It is being understood the first time you speak, in the language you dream in, by a State that finally listens. That is what we are building — and it is why we started with the languages the systems usually skip.

Cada decisión, a la luz.

— Anushka Appala and Dr. Janio Rosales