FLAIR project at SFU harnesses AI to revive Indigenous languages

A new project hosted at Simon Fraser University (SFU) is helping Indigenous communities train a new generation of speakers and support them with technologies including artificial intelligence (AI).

First Languages AI Reality (FLAIR), housed at SFU’s Digital Democracies Institute (DDI) and Cedar Supercomputing Centre, is developing AI-powered learning modules for at-risk languages and training AI models on them to help more people learn languages that are in danger of disappearing.

The project, led by Indigenous scientists Caroline and Michael Running Wolf, is currently working with four Indigenous communities: the Gwich’in in Northwest Territories, Kwakwaka’wakw on Vancouver Island, the Makah Tribe in Washington State and the Yawanawá of the western Brazilian Amazon.

“There are some Indigenous languages here in Canada, and probably in a lot of Indigenous communities around the world, where some communities only have five speakers left, or no first language speakers at all,” says Caroline Running Wolf. “They don’t have speakers younger than 60, and so the number of speakers is declining and that gives this a bit of urgency.”

According to UNESCO, more than 40 per cent of the world’s languages are endangered and a language disappears every two weeks. The problem is exacerbated online, where the internet is predominately English. This leaves most Indigenous language speakers underserved and unable to take advantage of emerging digital services in their mother tongue.

For example, manually transcribing just one hour of audio recording can easily take a person up to 40 hours and requires someone to already be proficient in the language. Once trained, an AI-powered tool could transcribe a recording in a fraction of the time. While someone still needs to go through the transcript and verify its accuracy, the time it takes is greatly reduced and frees up an expert to focus on more nuanced work. This is a technology that could be leveraged for Indigenous languages in the future.

“There are huge access gaps. There are 7,000 spoken languages across the world and if there’s only a couple dozen that are well represented on the internet, that’s obviously creating a squeeze upon minority languages,” says Michael. “And as artificial intelligence becomes more and more critical to our day-to-day lives, work and how we interact, our languages are not going to have much space to live and co-exist.”

The need to support Indigenous languages with AI solutions is so great that FLAIR appears in the Government of Canada’s AI for All national artificial intelligence strategy under a priority to “support and amplify Indigenous-led AI initiatives that reinforce cultural expression and linguistic vitality in Canada.”

The first phase of FLAIR’s work in communities is to develop a language curriculum led by community members and Indigenous language experts. Next, they will collect a digital dataset of audio with corresponding text curated by the Indigenous community.

They will then use machine learning to train AI on the language, teach it the proper pronunciation and morphology and develop practical speech recognition, speech-to-text, and online learning tools.

Instead of the large language models we’re used to hearing about with artificial intelligence that require massive data centres, FLAIR is creating much smaller, more efficient tools that can be used by Indigenous communities on local devices like phones and computers, both online and offline.

And while the bulk of the computing and data crunching is done at the Cedar Supercomputing Centre, FLAIR has ensured that everything they co-develop with communities remains their property, ensuring the community retains full sovereignty over their own data.

“Because we’re working with communities where you only have a handful of fluent speakers, how do we make sure those skills are available and accessible to other community members?” says Michael. “AI offers that. It’s not going to be a complete solution but it’s a good beginning to get people practicing their language in the safety of their own home, whether they’re in the Northwest Territories or away from their community in British Columbia.”

“Our research goal is not to replicate, replace or push aside human education, culture and interactions. It’s to do the tedious things that AI is very good at where you just want to practice how to pronounce a sound and learn the basics of your own language so that the human teachers can focus on the complex morphology, culture and cultural knowledge.”

To achieve their goal, FLAIR has become a research stream at the DDI to solidify the collaboration between Caroline Running Wolf and DDI director Wendy Hui Kyong Chun.

They first worked together on the Data Fluencies Project. The DDI now provides administrative, logistical, financial and academic support to FLAIR.

“In terms of responsible AI, this means thinking what ‘responsible’ means at every step and allowing community members to determine and own what is produced. This is what Caroline does.

“Rather than some generic commitment to ‘AI for good’, FLAIR is committed to make and build meaningful technical social systems. It’s focused; it addresses questions of data ownership; it rebuilds systems and speaks to a real need in the community. I think that’s what’s so important about working with FLAIR.”

FLAIR says that while the AI learning tools it is developing are still in progress, communities have already seen immediate benefits in the first phases.

“To think that some of these languages are only spoken by a handful of people and could be lost forever is terribly sad,” says James Peltier, director of SFU’s Research Computing Group, which provides support for researchers and SFU partners with technical support, expertise and access to digital research infrastructure. “It’s so important to be helping an Indigenous-led organization in preserving these Indigenous languages and keeping them alive by hosting them here at the Cedar Supercomputing Centre.”

In addition to support from SFU, FLAIR has had help from corporate partners as well, such as Cisco donating the hardware required for the project’s server cluster.

If FLAIR’s four pilot projects are successful, the group believes the models can be rolled out in many more communities around to world to help preserve and revitalize Indigenous languages and create a new generation of speakers that can carry them forward.

“Our big dream is that all across the Americas, and the world, Indigenous languages are thriving,” says Caroline. “We wouldn’t have the need to preserve them because they are in people’s hearts and minds and in daily use and are alive and well. We’re just adding our little piece to that bigger dream.”