Est. 2018 — Borderless Collective

Where language, systems & intelligence converge.

ELSA Network is a distributed collective of linguists, engineers, designers and researchers building the open infrastructure that connects human expression with machine understanding — across borders, disciplines and generations of technology.

142Countries
9 400+Members
310Open projects
38Labs & hubs
Computational Linguistics Decentralized Collaboration Open Source AI Human-Machine Interfaces Language Preservation Design Systems Computational Linguistics Decentralized Collaboration Open Source AI Human-Machine Interfaces Language Preservation Design Systems
About the network

A quiet federation of people who refuse to let language be a barrier.

ELSA Network began as an informal correspondence between three research labs on two continents — a linguist in Lisbon, a systems architect in Osaka, and a designer in Montréal — who believed that the next decade of technology would be written not only in code, but in the way we teach machines to listen.

Today we are a federation of independent labs, universities, open-source maintainers and independent studios. We do not sell products. We build public goods: datasets, specifications, protocols and reference implementations that anyone can fork, audit and extend without permission.

Our members publish research, run local community hubs, translate resources into underrepresented languages, and mentor early-career practitioners. Every project we fund ends up under an open license. Every paper we co-author ends up in the public record. Every dataset we curate is documented, consent-aware and reversible.

We work slowly and deliberately. We believe that speed without care produces brittle systems — and that language, being the medium of human thought, deserves infrastructure as considered as the bridges we build for ourselves.

Open by default

Every artifact we produce — code, data, guidelines — is published under a permissive licence. No NDA, no paywall, no gatekeeping.

Consent-aware data

We only work with corpora that have been gathered with explicit community consent, documented provenance and a defined removal path.

Local roots, global reach

Our 38 hubs operate autonomously. They choose their own projects, their own languages, their own pace — and share everything back to the network.

Long-horizon thinking

We fund five-year research programs, not quarters. Our working assumption is that the meaningful work will outlive all of us.

Programs

Four pillars. One commitment to the public record.

Everything the network does falls into one of four long-running programs, each stewarded by a rotating council of members.

01 / LINGUA

Language Preservation Lab

Working alongside 87 indigenous and minority language communities to build open lexicons, speech corpora and teaching tools that remain under community ownership — not corporate extraction.

02 / FORGE

Open Model Foundry

Releasing small, auditable, energy-efficient language models with fully documented training pipelines, so that independent researchers can reproduce and extend our results without a nine-figure compute budget.

03 / ATLAS

Protocols & Standards

Drafting the boring, thankless specifications — provenance formats, consent metadata, evaluation protocols — that make the next generation of AI systems legible, testable and accountable.

04 / COMMONS

Community & Fellowships

Funding early-career engineers, linguists and designers for twelve-month residencies inside network hubs, with a strict no-strings publication requirement at the end.

The network

A living map of collaborators in 142 countries.

ELSA is not a company. It is a lattice of independent labs, universities and studios that agree on a shared public commons. Membership is free; contribution is the only currency.

9 412Verified members
38Local hubs
310Active repositories
187Languages represented
Selected projects

Work shipped by the network this year.

Open Dataset

Tongue/1

A 42-language parallel corpus of everyday speech, gathered in partnership with community radio stations across four continents. Fully consent-tagged and released under CC-BY-SA.

Specification

Provenance v2

A lightweight metadata format for tracking the origin, licensing and removal path of every training sample. Now adopted by eleven independent laboratories.

Reference Model

Ember-7B

A small multilingual model trained on a fully reproducible 900-billion-token pipeline, released with every checkpoint, log and evaluation harness intact.

Tooling

Fieldnotes

An offline-first app used by 1 200 field linguists to record, transcribe and align endangered-language interviews without an internet connection.

Journal

News from inside the network.

Field reports, research notes and long-form essays written by members.

No news yet.

Open call

Build the commons with us.

Membership is free and open to anyone with something to contribute — a dataset, a paper, a translation, a well-argued objection. We review applications every two weeks.

Apply to join the network →
Contact

Reach the network.

Our inbox is monitored daily by a rotating group of members. We respond to every serious enquiry within five working days.

General enquiries [email protected]
Research & papers [email protected]
Press & media [email protected]
Fellowship applications [email protected]
Mailing address ELSA Network Collective — Hub Lisboa
Rua da Boavista 84, 1200-069 Lisboa, Portugal

We do not operate a phone line. Every conversation with the network begins in writing — a deliberate choice, so that our records remain readable, searchable and permanent.