By Clare Duffy, CNN

New York (CNN) — Like many of the young people in her Indigenous Anishinaabe community in the upper peninsula of Michigan, Danielle Boyer grew up speaking only a little of her people’s native language, Anishinaabemowin.

“In our community, generational language loss is very rapid,” she told CNN. “A lot of our grandparents speak it and then our parents speak a little bit of it and then we speak even less of it.”

But Boyer, 24, is now seeking to reverse that trend with her language teaching robot, the SkoBot. Inspired by a talking Elmo toy, SkoBot is designed to be an interactive way for children to learn Anishinaabemowin.

Boyer’s project is part of a growing push to preserve and revitalize endangered languages with the help of robotics and artificial intelligence, often led by young people looking to connect more deeply with their Indigenous roots. The effort comes at a time when the United Nations estimates that an Indigenous language dies every two weeks and that half of the world’s languages will disappear by 2100.

That’s largely because, in recent centuries, waves of colonizers globally discouraged or even outlawed the teaching and learning of Indigenous languages — eradication that researchers and technologists like Boyer, as well as organizations like UNESCO, are now working to undo.

“When you lose your language, you lose such a key component of your culture and your ways,” Boyer told CNN’s Terms of Service podcast, adding that her community lost many elders who spoke the language during the pandemic. “It’s the way that we communicate about the world around us. It’s the way that we tell stories.”

Jared Coleman, a member of the Big Pine Paiute Tribe of Owens Valley, California, who is now an assistant professor at Loyola Marymount University in Los Angeles, said he was initially inspired to study computer science as an undergraduate in hopes of creating a program like Rosetta Stone for his native language, Owens Valley Paiute. Although he learned a few words in the language during his childhood, Coleman grew up “off reservation,” so language classes were harder to access.

“The last person to speak the language fluently in my family was my great grandfather,” Coleman told CNN. “My great grandpa went to a boarding school where it was prohibited to speak the language, so my grandma didn’t get taught the language. That’s the sad history of the language in my family, and it’s the same for a lot of people in my tribe and in a lot of other tribes.”

Teaching a language with the help of AI

Boyer worked with two mentors, members of different Indigenous communities, to develop the SkoBot — a robot about the size of a coffee mug that looks like a woodland animal and sits on the shoulder of the wearer to facilitate easy back-and-forth conversation.

When the user says a word in English, the SkoBot uses AI speech recognition technology to identify the word and play the corresponding, pre-recorded audio file of the same word in Anishinaabemowin. If the users says “hello” to the SkoBot, it will respond with, “Boozhoo.”

The recorded audio files feature the voices of children from the community because the SkoBots are intended to be used for language learning by children in classrooms. The project brings together two of Boyer’s passions: language revitalization and STEM education for Indigenous youth.

“We bring the SkoBots into classrooms, and the students build the robots themselves, which is really exciting,” she said. “They get to design their own aspects of it, they get to wire the robot, and then from there … you speak to it.”

Coleman and his team used two versions of OpenAI’s large language models, GPT-3.5-turbo and GPT-4, to create an AI system trained on words from Owens Valley Paiute. Because of the model’s existing understanding of sentence structure in other languages, it was able to take those words and string together basic sentences.

Based on that research, Coleman released an online Owens Valley Paiute dictionary, sentence builder and translator and hopes to continue building more advanced language-learning tools.

He said the tools are intended “first and foremost” to help other members of his community, but he also hopes people visiting the region in California will take an interest in learning the language, as well.

“We have a lot of tourists going through for hiking and fishing and skiing at Mammoth Mountain,” he said. “My hope is that people will be interested in learning about the First People of the land that they’re playing on and visiting.”

Considerations for language preservation in the AI era

Some technologists working to revitalize their communities’ languages say they’re being intentional about how to apply AI to the problem in light of how Indigenous communities’ resources have often been extracted without compensation or consent.

Boyer, for example, said she decided to use real, pre-recorded voices in the SkoBot rather than AI-generated audio because “languages are living things … language learning should never happen purely with a robot or on your phone, it should always happen with a community member.” The community members who made the recordings retain ultimate ownership of them as part of an ethical AI framework Boyer and her advisors developed for the project, she said.

“The basic goal around it is for our youth to be able to introduce themselves first and foremost in our languages because that’s really important for … your sense of belonging,” Boyer said. “The other component of it that I’m really passionate about is the documentation part of it, making sure that my language is recorded and well-documented but in a way that’s not being exploited by companies that are not from our communities.”

Similarly, Coleman said that while he’s learned from writings and recordings of his community’s elders speaking the language, including his great grandfather, he hasn’t directly uploaded any verbatim sentences to train his AI model. He said he wants to prevent those comments from potentially being used by AI companies in ways he can’t control.

“Some of the recordings belong to different people … other families that are represented there might have different opinions on how those recordings should be used, some of them are sacred songs, some of them are telling sacred stories, so we’ve been very careful about what we use to train the models,” Coleman said.

He continued: “We want to be very careful in the way that we do use them because there’s some things that you do that you can’t go back on.”

They’re also mindful of accuracy. Both Boyer and Coleman recall instances of seeing mainstream AI chatbots spit out inaccurate reflections of their communities’ languages, something they said could lead to incorrect understandings of their cultures.

“Language is so much more than just its words,” Coleman said. “It encodes an entire culture and an entire history along with it.”

The-CNN-Wire
™ & © 2025 Cable News Network, Inc., a Warner Bros. Discovery Company. All rights reserved.