In 2022, after historic storms hit remote villages across Western Alaska, the Federal Emergency Management Agency hired a California-based contractor to help residents access disaster aid. Their job was to translate applications for financial assistance: The Yukon-Kuskokwim Delta is a constellation of small Alaska Native communities, and nearly half the region’s population — some 10,000 people — learn to speak Yugtun, the Central Yup’ik dialect, before they learn English. Farther north, approximately 3,000 people speak Iñupiaq.
But when the translations came through and journalists at the local public radio station, KYUK, tried to read them, they found that the material was nonsense.
“They were Yup’ik words all right, but they were all jumbled together, and they didn’t make sense,” said Julia Jimmie, who is Yup’ik and works as a translator at KYUK. “It made me think that someone somewhere thought that nobody spoke or understood our language anymore.”
Three years later, the region is reeling from another storm: Typhoon Halong, whose remnants displaced more than 1,500 residents and killed at least one person in the village of Kwigillingok in mid-October. And despite recent changes to FEMA policy, translation is once again raising questions — this time, about the role of AI in Indigenous communities.
Prisma International, a Minneapolis-based company, posted an ad seeking “experienced, professional Translators and Interpreters” of Yup’ik, Iñupiaq, and other Alaska Native languages on Oct. 21, the day before the Trump administration approved a disaster declaration for the storm.
The company has contracted with FEMA more than 30 times over the last few years, according to government records. Its website says that Prisma's tools “combine AI and human expertise to accelerate translation, simplify language access, and enhance communication across audiences, systems, and users.” According to the job listing, the Alaska Native language translators would be asked to “provide written translations using a Computer-Assisted Translation (CAT) tool.”
A spokesperson for FEMA declined to say in late October whether the agency planned to contract with Prisma in Alaska, and the company did not respond to multiple requests for comment by phone and email. But the job posting notes a preference for applicants with experience translating or interpreting “for emergency management agencies, e.g. FEMA,” as well as knowledge of the recent storm and a connection to local Indigenous communities. Multiple Yup’ik language speakers in Alaska confirmed they had been contacted by a company representative, who described Prisma as “a language services contractor for the Federal Emergency Management Agency.”
Julia Jimmie was among those contacted. She said she would have been happy to translate for FEMA, but she had more questions about working with Prisma.
As AI expands into new areas of everyday life, including translation, it’s prompted both excitement and skepticism in Indigenous communities. Many Native tech and culture experts are intrigued by its potential, particularly when it comes to language preservation. But they warn that the technology risks distorting cultural knowledge and could threaten language sovereignty.
“Artificial intelligence relies on data to function,” said Morgan Gray, a member of the Chickasaw Nation and a research and policy analyst at Arizona State University’s American Indian Policy Institute. “One of the bigger risks is that if you’re not careful, your data can be used in a way that might not be consistent with your values as a tribal community.”
Though the U.S. government does not formally regulate AI or its use, the concept of “data sovereignty” — a tribal nation’s right to define how its data is collected and used — is increasingly part of international discussions about Indigenous intellectual property. Free, prior and informed consent for the use of Indigenous cultural knowledge is written into the United Nations Declaration on the Rights of Indigenous Peoples. UNESCO, the U.N. body that oversees cultural heritage, has called for AI developers to respect tribal sovereignty in dealing with Indigenous communities’ data.
“A tribal nation needs to have complete information about the way that AI will be used, the type of tribal data that that AI system might use,” Gray said. “They need to have time to consider those impacts, and they need to have the right to refuse and say, ‘No, we’re not comfortable with this outside entity using our information, even though you might have a really altruistic motivation behind doing it.’”
It’s still unclear whether Prisma has contacted tribal leadership in the Y-K Delta. The Association of Village Council Presidents, a consortium of 56 federally recognized tribes in the region, did not respond to a request for comment.
On its website, Prisma says that clients can opt to use only human translators, noting that its AI use is governed by an AI Responsible Usage Policy. But the details of that policy are not readily available online, and the company did not respond to requests for clarification.
In the three years since its contractor produced the notoriously botched translations, FEMA has sought to improve its work with Alaska Native communities. KYUK’s reporting on the scandal prompted a civil rights investigation, and the California contractor, Accent on Languages, reimbursed the agency for the faulty translations. A spokesperson for FEMA said the agency now employs only “Alaska-based vendors” for Alaska Native languages, prioritizing those in disaster-impacted areas. It also requires a secondary quality-control review of all translations. “Tribal partners are continuously consulted to determine language services needs and how FEMA can meet those needs in the most effective and accessible manner,” the spokesperson wrote in an email.
The agency’s policies on AI are less clear. The email did not directly address questions about whether FEMA has policies in place to regulate the use of AI or protect Indigenous data sovereignty, though it said that FEMA “works closely with tribal governments and partners to make sure our services and outreach are responsive to their needs.”
According to government records, FEMA has contracted with Prisma in more than a dozen states. Prisma’s website highlights a case study in which it used its “LexAI” technology to help a federal agency provide information about disaster relief in multiple languages following a wildfire. The case study says it offered translation for more than 16 languages, including “rare Pacific Island dialects.”
The company has also worked with several other federal agencies, the government database shows, but it does not appear to have contracted with the federal government in Alaska before.
In the Y-K Delta, the Yup’ik language translators Prisma contacted shared a practical concern: Would AI be able to translate their language accurately?
“Yup’ik is a complex language,” Jimmie wrote in a message. “I think that AI would have problems translating Yup’ik. You have to know what you’re talking about in order to put the word together.”
Most AI models rely on extensive data to help produce accurate translations. But that kind of data is rarely available for Indigenous languages, and AI has a poor record when it comes to translating them, often providing inaccurate sentences and even completely invented words.
Sally Samson, who is Yup’ik and a professor of Yup’ik language and culture at the University of Alaska Fairbanks, said she was skeptical that AI could master Yugtun syntax, which differs substantially from English. Her concern wasn’t just that the technology might provide misinformation, but that it would fail to convey the nuances of a Yup’ik worldview.
“Our language explains our culture, and our culture defines our language,” she said. “The way we communicate with our elders and our co-workers and our friends is completely different because of the values that we hold, and that respect is very important.”
“Our language explains our culture, and our culture defines our language.”
Indigenous software developers are actively working to address some of AI’s shortcomings around Native languages. Many hope to mobilize the technology to preserve endangered dialects. An Anishinaabe roboticist has already designed a robot to help kids learn Anishinaabemowin, while a Choctaw computer scientist created a chatbot to converse in Choctaw.
But in those cases, Indigenous people have been the ones developing the AI models and making decisions about how to use them. Crystal Hill-Pennington, who teaches Native law and business at the University of Alaska Fairbanks and offers legal consultation to Alaska tribes, said she worries about the potential for exploitation if AI is trained on the work of Indigenous translators for future use by non-Native companies.
“If we have communities that have a historical socioeconomic disadvantage, and then companies can come in, gather a little bit of information, and then try to capitalize on that knowledge without continuing to engage the originating community that holds that heritage, that’s problematic,” she said.
Native communities have centuries of experience with outsiders extracting and exploiting their cultural knowledge — but there is also recent precedent for this kind of controversy. In 2022, the Standing Rock Sioux Tribal Council voted to banish a nonprofit that had promised to help preserve their language. After Lakota elders spent years sharing cultural knowledge with the non-Native group, the organization copyrighted the material and tried to sell it back to tribal members in textbooks.
Hill-Pennington said the introduction of AI by private companies adds another layer of complexity to contemporary conversations around intellectual property.
“The question is, who ends up owning the knowledge that they’re scraping?” she said.
Standards around AI and Indigenous cultural knowledge are evolving quickly, just like the technology. Hill-Pennington said that some companies using AI may still be unfamiliar with the expectation of informed consent and the concept of data sovereignty. But, she said, those standards are becoming increasingly relevant.
“Particularly if they’re going to be doing work with, let’s say, a federal agency that does fall under executive orders around authentic consultation with Indigenous peoples in the United States, then this is not something that should be overlooked,” she said.
This article first appeared on High Country News and is republished here under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.