AI Is Racing to Save the 3,000 Languages That Will Die This Century. The Root of Language Death Goes Back to Empire.

Key Takeaways

• •A language dies approximately every two weeks — by 2100, an estimated 3,000 of the world's 7,000+ languages will be extinct
• •Meta's No Language Left Behind AI translates 200 languages including many low-resource and endangered ones
• •US Indian boarding schools (1879-1960s) systematically punished Native children for speaking indigenous languages
• •AI tools now help communities create dictionaries, pronunciation guides, and language-learning apps for languages with no written tradition

Source: Ethnologue / UNESCO

Timeline

Marie Smith Jones died on January 21, 2008, in Anchorage, Alaska. She was 89 years old. With her death, the Eyak language — spoken for thousands of years along the copper river delta of southern Alaska — ceased to exist. Its last fluent speaker was gone. An entire way of encoding reality, of describing the relationship between humans and salmon and glaciers, of telling jokes that only work in Eyak, vanished from the Earth.

This happens every two weeks. On average, one of the world's approximately 7,000 languages dies every 14 days. Linguists estimate that between 50% and 90% of today's languages will be extinct by the end of this century. That means somewhere between 3,000 and 6,000 languages — each one a unique cognitive architecture for understanding reality — will disappear within our lifetimes.

AI can't stop language death. But it's giving endangered language communities something they've never had before: tools powerful enough to document, teach, and revitalize their languages at the speed that's needed.

THE ROOT

When an elder dies, a library burns. When a language dies, a civilization's way of understanding the world disappears forever. No amount of archaeology can recover it.

Languages have always died. What's different about the current extinction wave is its speed and its cause. Most language death in recorded history can be traced to a single pattern: empire.

When Rome expanded across Western Europe, Latin didn't just coexist with local languages — it replaced them. Gaulish, the Celtic language spoken across what is now France, was dead by the 6th century. Etruscan, the language of the civilization that preceded Rome in Italy, left behind inscriptions that scholars still struggle to fully decipher. Celtiberian, Lusitanian, Dacian — all gone. Latin didn't erase these languages through violence alone. It erased them through administration, trade, law, and prestige. If you wanted a government job, you spoke Latin. If you wanted to trade, you spoke Latin. Within a few generations, parents stopped teaching their children the old language because it had no economic value.

The same pattern repeated worldwide. Spanish colonizers suppressed Quechua, Nahuatl, and hundreds of other indigenous languages across the Americas. British colonial education systems imposed English in India, Africa, and Southeast Asia. French colonial policy explicitly aimed to replace local languages in West Africa and Indochina.

But the most systematic language destruction happened in North America. In 1879, Captain Richard Henry Pratt opened the Carlisle Indian Industrial School in Pennsylvania with the explicit mission to "Kill the Indian, save the man." Native American children were taken from their families, often by force, and placed in boarding schools where they were beaten for speaking their native languages. Their hair was cut. Their names were changed. They were forbidden from practicing their cultures.

The boarding school system operated for nearly a century. By the time it ended, many Native American languages had lost an entire generation of speakers. Languages that had been spoken for thousands of years were reduced to a handful of elders.

The Cherokee syllabary was created by Sequoyah in 1821. He was illiterate in English. He invented a writing system for his people's language in 12 years. Now AI helps his descendants teach it.

THE AI LIFELINE

Traditional language documentation is slow, expensive, and labor-intensive. A trained linguist works with a native speaker for months or years to create a grammar, a dictionary, and a collection of recorded texts. This work is essential, but there aren't enough linguists, and many endangered languages have only a handful of elderly speakers left. The clock is ticking.

AI is accelerating every step of this process. Speech recognition models can now be trained on relatively small amounts of audio data — as few as a hundred hours — to create speech-to-text systems for languages that have never had one. Machine learning can identify grammatical patterns in transcribed texts, helping linguists build grammars faster. Neural machine translation can create rough translation tools between an endangered language and a major language, even with limited training data.

Meta's No Language Left Behind project, released in 2022, demonstrated that AI translation could work across 200 languages, including many with minimal digital presence. Google's 1,000 Languages Initiative is building speech and text models for the world's most-spoken 1,000 languages — covering 98% of the world's population but also including many languages that are on the brink.

On the ground, the most impactful work is being done by indigenous communities themselves, using AI tools to build language-learning apps, create interactive dictionaries, and produce educational content. The Māori community in New Zealand has built AI-powered tools for the te reo Māori language. First Nations communities in Canada are using AI to create pronunciation guides from elder recordings. Cherokee Nation is building on Sequoyah's 200-year-old syllabary with modern speech recognition.

Sequoyah's story is worth noting. Born around 1770 as a Cherokee silversmith who could not read or write English, he spent 12 years developing a complete writing system for the Cherokee language — 85 characters representing every syllable. By 1825, the Cherokee Nation was more literate than the surrounding white population. It was one of the most remarkable intellectual achievements in human history. Now AI tools help teach his syllabary to a new generation.

WHAT'S LOST WHEN A LANGUAGE DIES

Languages are not just communication tools. They are cognitive frameworks — different ways of parsing reality. The Hopi language has no grammatical tense, suggesting a fundamentally different relationship with time. The Guugu Yimithirr language of Australia uses cardinal directions (north, south, east, west) instead of relative directions (left, right), and its speakers develop extraordinary spatial awareness as a result. The Pirahã language of Brazil has no words for specific numbers, only "few" and "many."

When a language dies, we don't just lose words. We lose a way of thinking. We lose metaphors, stories, songs, jokes, and knowledge systems that evolved over thousands of years. We lose medicinal plant knowledge encoded in vocabulary that has no equivalent in English. We lose navigational knowledge, ecological knowledge, and astronomical knowledge that exists only in the oral traditions of small communities.

The root of language preservation isn't technology. It's recognition — the same recognition that Abbé de l'Épée brought to sign language, and that Sequoyah brought to Cherokee. The recognition that every language, no matter how few its speakers, encodes something irreplaceable about the human experience.

AI can't save a language that no one wants to speak. But for the communities that are fighting to keep their languages alive — and there are hundreds of them — AI is the most powerful ally they've ever had.

Every two weeks, a library burns. AI is trying to copy the books before the flames reach them.

How did this make you feel?

Recommended Gear

Keep Reading

AI for Good

AI Is Now Detecting Wildfires Within 10 Minutes of Ignition. The Root of Fire Lookouts Goes Back to the Roman Empire.

7 min read

AI for Good

AI Can Now Translate Sign Language in Real Time. The Fight for Deaf Communication Rights Is 250 Years Old.

7 min read

AI for Good

Google's AI Can Now Predict Floods 7 Days in Advance for 80 Countries. The Root of Flood Warning Goes Back to Ancient Egypt.

6 min read