Who Owns a Language's Data?
Michael Running Wolf, a citizen of the Northern Cheyenne Tribe, grew up in Birney, a remote village in southeastern Montana where running water and electricity were intermittent. He was raised in part by a grandmother who spoke Cheyenne. He earned bachelor's and master's degrees in computer science from Montana State University, worked as a software engineer on Amazon's Alexa, taught at Northeastern University, and went on to pursue a PhD in computer science at McGill University.
Modern automatic speech recognition (ASR) is built on machine learning: a model is trained on large quantities of transcribed audio until it can generalize to speakers it has never heard. Performance depends heavily on the volume and variety of the training data. For English and other widely spoken languages, that data is abundant. For many endangered Indigenous languages it is scarce — Michael has noted that for many Montana tribes the available recordings may total around 100 hours, and some tribes have essentially none.
The languages themselves pose a second challenge. Many Indigenous languages are polysynthetic: words are built from many meaningful parts, so one word can express what English needs a full sentence to say. Systems designed around vocabularies of shorter, separate words handle this poorly, and Michael has described it as an unsolved problem in AI. Inspired in part by Māori researchers in New Zealand who applied AI to their own language, he co-founded First Languages AI Reality (FLAIR), an initiative developing adaptable AI tools for Indigenous language revitalization.
Running through his work is the principle of data sovereignty — the idea that a community should own and control data about itself, including recordings of its language, stories and ceremonies. Michael argues that tribes need their own engineers and computer scientists and must set the terms for how their data is collected and used; otherwise the same technology meant to help a language could end up owned by someone else.
