Language Heritage · Ordinary People
Coleman wanted a Rosetta Stone for Owens Valley Paiute. There wasn't one, so Coleman built one.
Jared Coleman · Owens Valley, California
A computer scientist teaches a language model to speak the endangered tongue of the Big Pine Paiute Tribe, one rule and one elder's sentence at a time.
The person and the place
Jared Coleman is a member of the Big Pine Paiute Tribe of the Owens Valley, California, now an assistant professor of computer science at Loyola Marymount University in Los Angeles.
The problem they cared about
Five years ago Coleman found an old archive of Owens Valley Paiute: a dictionary and recordings of Coleman's elders, including Coleman's own great-great-great-grandfather's voice. "I had no idea what he was saying," Coleman told WHYY. The archive was enough to learn from by hand, but not enough to train an AI model the usual way. Google Translate-style systems learn from "millions and millions and millions of sentences," Coleman said. "You can't do that for endangered languages, we just don't have that many sentences."
The agency moment
In 2023, finishing a PhD at USC with advisor Bhaskar Krishnamachari and researcher Khalil Iskarous, Coleman built something different: a system that learns a language's grammar rules directly instead of waiting for a training set that will never exist. Then Coleman brought in native speakers and elders to train the system by hand, the way you'd teach a person. "So I taught it," Coleman told WHYY. "I said this is the word for water. This is the word for drink."
What changed, with evidence
The result, Kubishi (Paiute for "brain"), lets people look up Paiute words, build new sentences, and translate the language. It's publicly available online, and Coleman's Kubishi Research Group keeps expanding it. It is also built not to bluff: "When you're talking with ChatGPT, you should never assume that it's accurate because it just makes stuff up all the time," Coleman told WHYY, which reported that Kubishi "doesn't spit out wrong translations when it doesn't know the answer." CNN independently confirmed the project the following year: the tools are intended "first and foremost" to help Coleman's own community, CNN reported, with hope that visitors to the Owens Valley will take an interest in the language too.
The move worth copying
Coleman didn't wait for a company with enough training data to get around to Owens Valley Paiute. When there wasn't enough data to teach a machine the ordinary way, Coleman taught it directly, one rule and one elder's sentence at a time. Know a language running low on data, and someone patient enough to teach it right? Feed the scout at goodinprogress.org.
"You can't do that for endangered languages, we just don't have that many sentences." — Jared Coleman, WHYY
Sources
Here is where this story came from. Go read the originals.
Curated by Good in Progress from the public record.
Sources last verified .
