A plea that does not answer its objections is only a sermon.

We have put these ideas to computer scientists, linguists, officials and entrepreneurs. Here are the strongest objections we heard, and what we answer.

“Big tech will cover every language anyway.”

Coverage is growing fast: in November 2025, Meta released open speech recognition for more than 1,600 languages. But for oral languages, text does not exist and will not appear on its own; speech must be gathered from those who carry it, with consent, time and trust. Above all, coverage says nothing about governance: who owns the voices, who draws the value, who decides what is correct. Understanding a language is not the same as respecting it.

“Translation gives access to the world’s knowledge.”

True, and we want more of it, into Darija and Amazigh too. What we refuse is translation as the mandatory step of all processing. Translating into a language to bring it knowledge adds. Passing a language through another to understand it subtracts.

“English is the world’s common language.”

For science, trade and diplomacy, probably. Our question is everyday speech, public services, health, education and memory, where most of humanity does not speak English. Latin was Europe’s learned language for centuries, yet peasants never had to pray, bargain or seek care in Latin.

“It is far too expensive.”

Count every cost: public services part of the population cannot use, dependence on foreign suppliers, national data valued elsewhere, and heritage no budget can buy back. Methods that learn from untranscribed speech cut annotation needs, and tools built once serve everywhere. That is the reason for an alliance.

“Languages die; that is life.”

Languages almost never die a natural death. They die because schools, administrations, markets and now machines were designed for other languages. Hebrew, Māori and Welsh came back when given the means. And even if a language falls silent, a duty remains: to listen before it does.

“This is identity politics.”

Morocco’s 2011 Constitution defines a national identity that is one and plural. Unity is not uniformity. Inverse AI lets everyone reach everything, other languages included, starting from their own: openness that asks no one to deny themselves to come in.

“Parents themselves prefer French or English.”

That choice deserves respect. But it comes from a world where the languages of home close doors. We seek to change the conditions in which the choice is made. And a child who masters a first language learns the next ones better.

“Orality is the past; we need literacy.”

Literacy is a right. But hundreds of millions of adults cannot read today. A technology that speaks their language gives them access now, and can become a bridge to writing. Orality is not the enemy of reading. It is the path to it.

“Who speaks for the communities?”

Perhaps the most serious objection. There is no perfect answer, but there are safeguards: plural councils of guardians (women and men, elders and young, residents and diaspora); reasoned, public decisions; individual rights that remain; rules that can be revised.

“A community veto will slow science.”

Experience mostly shows the opposite. Te Hiku Media gathered hundreds of hours of Māori in a few days because its community trusted it. Trust is not a brake on science. It is its fuel.

What remains open

  • LUMINA has no results of its own yet. Every hypothesis is still to be tested in the field.
  • Evaluation by speakers costs time and money; its reliability must be measured.
  • Who represents a community is itself a political act.
  • Long-term funding is not secured.

We do not claim to have solved everything. We claim the questions are well posed, and that it is time to start, together.