Why Give Native Languages to AI?

In the Lakota language alone, in the ten-year span between 2006 and 2016, approximately 4,000 fluent speakers passed away — an approximate 66% decrease^[1]. There have been recent efforts through several Lakota revitalization initiatives (one such example^[2]), and a “reawakening” of the language is currently being witnessed. However, these efforts alone have not been enough to revive the dying language, as few to no first-language speakers are being born today. Only roughly 1% of Lakota people are fluent speakers, compared to the 113,713 self-identified Sioux members from the U.S. Census^[3]. This is further worsened by the fact that many of the remaining first-language speakers are elderly, with an average age well into their 60s^[4].

These figures represent only the Lakota language, yet similar patterns appear across nearly every other Native American language (2025 Census Report^[6]). This is a tragic reality caused by the inhumane conditions that founded the United States, as discussed in this section^[5]. Unfortunately, those events are effectively irreversible given the country’s immense socioeconomic power — the highest GDP of any nation^[7]. The Native American Language Revitalization Project does not aim to reverse these conditions, but to provide modern access to these languages within the digital ecosystem. Additionally, we hope to help mass-produce learning resources for new Lakota speakers, as discussed in this section^[8], so that human speakers can continue to become fluent in the language even after all currently living first-language speakers have taken their journey into the spirit world.

The Native American Language Revitalization Project hopes to eliminate — or at the very least mitigate — the need for people to learn English simply to participate in the modern world. Globally, English is by far the most dominant language^[9]. Although it ranks only third by native speakers^[11], English has the highest number of total fluent speakers^[10]. This is due to the immense GDP^[7] of the United States, combined with how deeply English has become embedded in global celebrity culture^[12]. Existing in the world without knowing English has become extremely difficult. The vast majority of online content — social media, films, television, and more — is available in the English language^[13]. In contrast, almost no digital content is available in Lakota (a small Wikipedia list of films^[14]). This same trend exists across every Native American language; we simply use Lakota as our example, as it is the first language we support.

We hope to make digital content accessible to Native Americans around the world in their tribal language. Artificial Intelligence enables nearly any form of media to be translated into these languages, so no one has to miss out because of a language barrier. While AI cannot replace human speakers, it can help humans learn to speak a language by allowing them to quickly correlate concepts from their favorite shows, films, and books with the corresponding words. Research shows that visual anchors^[15] can significantly increase the rate at which someone learns a language. As we continue to improve the accuracy of the languages supported by the Native American Language Revitalization Project, we plan to release tools for translating digital content instantly. We hope to provide browser extensions, keyboard widgets, and OCR technology for uploading photos, e-books, PDFs, and more, translating them into a chosen language within seconds.

During the founding of the United States, the country successfully destroyed Native American languages^[16] and induced trauma that continues to persist^[17] to this very day. Federal policies such as the Civilization Fund Act of 1819 funded boarding schools where Native children were stolen from their families, forced to have their hair cut, assigned English names, pushed into colonialist religions, and, above all, prevented from speaking any language other than English. By forcing them into English from a young age, these policies caused them to forget their native language during the most critical developmental period^[18].

While the U.S. has made some attempts to protect these languages from extinction^[19], it cannot undo the damage that has already been done. By allowing these languages to go extinct, we allow the sins committed during that era to continue causing harm hundreds of years later. The Native American Language Revitalization Project hopes to take a drastic leap forward in language revitalization, far beyond what the remaining fluent speakers can achieve on their own. The number of remaining fluent speakers is dwindling rapidly, and once all have passed on, it will be essentially impossible to recover the language.

The rise of Artificial Intelligence is having a deeply unfortunate impact on the sustainability of our planet. AI data centers are producing immense — and potentially irreversible — harm to the environment, including excessive water usage, enormous electricity demand, and harmful emissions. This is unacceptable and will affect both current and future generations. We are already seeing prices skyrocket, including RAM prices^[27] and electricity prices^[28]. The impact extends far beyond economics: people are beginning to lose their jobs^[24] and lose their homes^[25].

Through the use of extremely lightweight AI models — less than 0.1% the size of typical modern models — and techniques such as quantization^[29] and knowledge distillation^[30], our models are small enough to run on consumer-grade GPUs and, in some cases, even on phones. This means we only need to rely on water-based cooling solutions during the initial training phase. In addition, we plan to expand beyond translator AI models^[26]. We want to produce specialized AIs for narrow tasks, so that the same lightweight approach can prevent misuse and accelerate humans rather than replace them. We are also taking measures to prevent our data from being used by other AI systems^[23].

For many valid reasons, Native American tribes and speakers want to keep their data sovereign. They share the belief^[31] that it is the right of a nation to govern the collection, ownership, and application of its own data. Alongside this, some express concerns about AI being trained on Native American languages at all. These beliefs are grounded in good intentions, and we take every feasible measure to respect and protect data sovereignty, as well as to prevent other AI systems from accessing our data.

Firstly, the Native American Language Project was founded and is led by a creator who not only carries Lakota and Choctaw DNA, but who also grew up on a Native American reservation throughout his middle school and high school years. Additionally, this is not the only project attempting to use AI to preserve Native American languages; other efforts — such as First Languages AI Reality^[33], Lakota BERT^[34], and the Lakota AI Code Camp^[35] — are pursuing similar goals.

Secondly, we have taken concrete measures to respect and protect data sovereignty. We use rate limits, CAPTCHAs, and paywalls^[26] to prevent other AI companies from harvesting the data our systems produce. Our AI was trained primarily without the help of pre-existing human-created data, combined with publicly available learning material that other AI companies also have access to. However, by providing mass translation capabilities, if this data is publicly shared, it will be impossible to guarantee it can never be used by another AI. As discussed in this section^[8], we believe the benefits of doing so outweigh the risks.

Thirdly, we do not believe Native American languages should be restricted only to Native American speakers. One article^[32] argues persuasively for the importance of passing these languages on more broadly. Furthermore, what defines a Native American in the first place? Blood quantum^[36] illustrates the deep problems with imposing a minimum blood percentage to determine identity. Given how small a percentage of the U.S. population is Native American, restricting the language to full-blood Native Americans is neither realistic nor ethical. Additionally, restricting a language to a single ethnicity constitutes a form of linguistic racism^[37].

We do offer paid subscriptions for access to our translator services, but we do so with the best of intentions. Free alternatives are available for anyone who cannot afford a premium subscription, and we offer generous discounts to schools, companies, and other organizations that provide access to this resource for learning-based purposes. In addition, paid tiers help restrict other AI systems from obtaining our data by limiting generative capabilities on the free tier.

The reason for this is simple: AI is expensive to train and operate. Premium subscriptions allow us to expand this translator to more Native American languages and to produce new learning resources at a much faster pace. Additionally, as discussed in this section^[22], we hope to expand our approach beyond translative AI. There are many dangers to AI, and we want to provide a safer alternative.