VOCES
ES

Revitalizing Languages Through Co-Created AI

An open, non-profit technology ecosystem dedicated to preserving and empowering low-resource languages.
Authorized and validated by the speaker community.

11
languages in development
5
Latin American countries
100%
open access, non-commercial

What is VOCES?

An interdisciplinary and intercultural program

VOCES seeks to protect, revitalize and promote Indigenous languages through digital empowerment and the development of technologies that create the conditions to use, learn, teach and pass on a language, making it part of everyday life.

It is not about archiving a language, but about making it present where life happens today: on the phone, at school, in public services and on the internet.

“A language existing in the digital world is not a cultural luxury: it is public infrastructure for its speakers' future.”

Our vision

  1. 01

    Cutting-edge tools

    Developing state-of-the-art language technology in the service of Indigenous peoples.

  2. 02

    Language institutions

    Strengthening Indigenous peoples' own language institutions so their work can be sustained over time.

  3. 03

    Enabling factors

    Laying the technological groundwork that lets peoples carry their language into the digital future.

Protect

Document, transcribe and preserve the living corpus of each language: voice and text.

Revitalize

Tools to learn, teach and pass the language on to those who come next.

Promote

Use it every day: at school, in healthcare, in public services, in the media and online.

The challenge

Every two weeks, an Indigenous language dies

That is 26 languages a year. Half of the world's languages could disappear this century, and each one takes with it an identity, a worldview and a heritage that cannot be recovered.

556

Indigenous languages in Latin America and the Caribbean

38.4%

are in danger of disappearing

47%

of the languages endangered today were not 15 years ago

66

languages have fewer than 99 speakers today

Source: Latin American Atlas of Endangered Indigenous Languages — IIALI · FILAC · SEGIB (2024) · UN DESA · World Bank (2019).

The digital funnel

For every 20 pages browsed on the web, one is in Spanish. It takes 125,000 to find one in Guarani, one of the Indigenous languages with the most presence online.

No Indigenous language of the Americas reaches the 0.1% threshold at which web content is measured. Artificial intelligence inherits that gap — and widens it.

  1. 7,613living languages in the world
  2. ~4,000with a writing system
  3. ~500with a presence on the internet
  4. 243available on Google Translate

Fewer than 5% of the world's languages can make the digital leap without deliberate intervention. The same technology that opened the gap can close it.

What we do

Tools that let a language work in the digital world

Our work starts from what speaker communities ask of us. These are the lines of work through which we respond.

Machine translation

Translation engines trained for each language, able to work with very little data and respectful of its structures and cultural expressions. They translate to and from Spanish.

Voice: transcription and synthesis

Audio transcription and read-aloud in the language. It is the most important line of work for communities where transmission is oral and much of the corpus lives in the memory of its speakers.

Workshops, training and courses

On-site training with speakers, teachers and community authorities: how to use the tools, how to correct them and how to decide over the community's own data.

Language datasets

Building and curating parallel corpora, lexicons and recordings, with governance in the hands of the community.

Intercultural validation

Every tool is reviewed with speakers and ancestral authorities before it opens: relevance before speed.

In development

Simultaneous translation

Live translation for meetings, ceremonies and public-facing services at institutions.

In development

Education and health apps

Digital readers, school textbooks and apps to learn and teach the language, and for healthcare.

Oral and written archive

Digitizing and recording stories, songs and testimonies in digital archives that are accessible and protected by their own communities.

In development

Integration API

Programmatic access so schools, public services and third parties can build the tools into their own systems.

Translators

The ones you can already use

From community pilots to public tools in real use. Each translator opens only once its speaker community has validated it.

In controlled release

They already work with real users from the community, while the intercultural validation that decides when and how they open is completed.

Mapuzungun translator
Controlled access

Mapuzungun

Wallmapu · Chile

Led by the Instituto de la Lengua y Cultura Mapuche Aukiñ Mapu, with communities from several territories taking part.

Bribri translator
Controlled access

Bribri

Costa Rica

Developed with ADITIBRI.

Ckunsa translator
Controlled access

Ckunsa

Atacama · Chile

Developed with the Consejo Lingüístico Ckunsa.

Iskonawa translator
Controlled access

Iskonawa

Ucayali · Peru

Developed with Estación Chana.

In the early stage

Processes in early conversation. A language is named here only once its speaker community authorizes it.

Early stage

Coming soon

Mexico

Early stage

Coming soon

Costa Rica

Early stage

Coming soon

Ucayali · Peru

Early stage

Coming soon

Chile and Peru

Early stage

Coming soon

Chile and Bolivia

Early stage

Coming soon

Bolivia

Where we work

11 languages in 5 countries

Each dot is a speaker community that chose to work with us. The work moves at the pace of each territory, not of the project's calendar.

Work in progress Early stages
Where VOCES worksBribriIskonawaCkunsaRapa NuiMapuzungun

How we work

The community decides. We build.

Our hallmark is that the tools are co-created with speaker communities, respecting their protocols and their timelines. No version opens until its speakers have validated it.

VOCES is a non-commercial collaborative alliance, aligned with the International Decade of Indigenous Languages 2022–2032.

Fieldwork with the community

Our own language models

Translation, voice and writing trained with and for each language, designed to work even with very little data.

Data governance

The data belongs to the communities: they decide what is collected, how it is used and who has access.

Participatory work

Built together on site with speakers, educators and community authorities, respecting protocols and timelines.

The collaboration flow

Eight stages, gone through once and then repeated in cycles of continuous improvement, always with the community at the center.

  1. 01

    Initial authorization

    The community decides whether it wants to work with us, and on what terms.

  2. 02

    Illustrative mock-up

    A first concrete version that shows what is possible, before asking for anything.

  3. 03

    Collaboration begins

    Material provided by the community, corrections from speakers and translation of selected sentences.

  4. 04

    Incremental improvements

    Model training, gathering of additional material and tailored adjustments.

  5. 05

    Intercultural validation

    Testing in real settings, with real users and the relevant institutions.

  6. 06

    Controlled release

    The community sets the strategy: restricted, public or mixed access.

  7. 07

    Sustainability or handover

    The tool stays in place, with local capacity to maintain it.

  8. 08

    New tools

    On top of translation, voice and writing, applications are built to revitalize, give access to and promote the language: education, health, media and public services.

A public good

Open access and non-commercial

Non-rival, non-excludable

One community using it takes nothing away from another: the same model serves schools, public services, researchers and creators.

An enabling factor

It is the foundation on which others build new tools that we cannot yet anticipate.

The market won't provide it

There is no commercial return in a language with a few thousand speakers: if public policy doesn't do it, nobody will.

VOCES makes available the enabling factors needed to meet each community's real needs.

Contact

Let's talk

If you represent a speaker community, an institution or a research team and would like to work with us, write to us.

Where we are
Centro Nacional de Inteligencia Artificial (CENIA)
Estudios Aplicados de Antropología UC
Santiago, Chile