One of those initiatives focuses on building text-to-speech software for learners of three Indigenous languages: SENĆOŦEN, Kanyen’kéha, and Plains Cree.
The project originated on the Six Nations reserve in southern Ontario, where technical lead Aidan Pine was helping to build verb conjugators for the Kanyen’kéha language after educator Brian Maracle suggested that technology would be helpful.
The language has hundreds of thousands of conjugations, which software can help students to navigate.
However, Pine heard from teachers and students who also wanted to hearwhat the conjugations sounded like. The idea for a specialist form of text-to-speech software was born – one that could generate the sounds without relying on humans to provide them.
Because fluent speakers are in high demand, having someone physically sit down and record was out of the question.
“We had hundreds of thousands of different conjugations. It was completely unreasonable to ask somebody to sit down and record each one of those,” said Pine.
Instead, the initiative’s goal is to generate audio directly from text that is good enough for classroom use. The work is in its first year, with two more to go.
‘Quite incredible’ results
The W̱SÁNEĆ school board, in southern BC’s Brentwood Bay, is one of the institutions working with Pine to help learners of SENĆOŦEN, the language of the Saanich people.
“It’s really quite groundbreaking, as far as having that ability of technology to read some SENĆOŦEN,” said Tye Swallow, the W̱SÁNEĆ school board’s language revitalization co-ordinator. Swallow said there are “less than a handful” of native speakers.
The school board has run a SENĆOŦEN immersion program for more than a decade, and will next year expand its program to cover preschool to Grade 10.
Champions of the SENĆOŦEN language have recorded and documented the language, leaving archival material behind, Swallow said. But text-to-speech technology can rapidly expand what’s available.
Swallow said a teacher at the school, PENÁĆ (David Underwood), sat down with Pine to speak SENĆOŦEN and provide samples of SENĆOŦEN text to what Swallow called “a supercomputer.”
“And out came this speech that was really quite incredible,” Swallow said.
“We kind-of joke – we call it cyber PENÁĆ, or lobot PENÁĆ,” he continued, noting that SENĆOŦEN does not have an R sound, so robot becomes lobot, “because there was still some sort of twang. You could hear a slight metallic twang.”
The program was able to produce audio of SENĆOŦEN that PENÁĆ hadn’t recorded, Swallow said, and did a pretty convincing job.
He hopes to fine-tune the program over the next few years to make the model more adept. The more data it receives in the form of recordings, the more accurate the audio synthesized by the model will be.
However, Swallow said the technology could be abused. For example, someone could manipulate a voice used to power the recordings, like that of an Elder, into saying something nasty.
The model is not yet being used in classrooms at the school board. Instead, its potential is part of a broader conversation being had as the community begins to work with this kind of technology.
The goal of the project is to create a “recipe,” said Pine, that lets Indigenous communities build their own text-to-speech models.
Communities “maintain ownership of their data and maintain ownership of their technology,” Pine said. “It’s an important part of developing technology for language revitalization.”
Still, Pine recommends that those interested in using technology for language be familiar with the risks, as outlined in a guide made by the First Peoples’ Cultural Council, a BC language, arts and culture group.
“The natural reaction for a lot of us when we see a new technology is: ‘It’s shiny, it’s new, it looks cool, let’s just adopt it.’ And sometimes, that can have bad consequences,” Pine said.
FirstVoices, developed by the First Peoples’ Cultural Council, is another example of digital technology being used to maintain Indigenous languages.
The website, which is free, allows Indigenous communities to share their language, oral culture and linguistic history.
Communities can choose whether their site is public or for members only. The platform also provides a downloadable digital keyboard app for various languages.
The Gwich’in Tribal Council’s Department of Early Childhood and Language has made digital language resources available on its own website.
Michelle Wright, the manager of early learning and language at the Gwich’in Tribal Council in Inuvik, says an urgency surrounds Gwich’in revitalization efforts.
There are 550 speakers across Canada, with 350 located in the Northwest Territories. According to the Endangered Languages Project, Gwich’in is severely endangered.
Wright said recordings of Gwich’in dating back to the mid-20th century have been translated into written documents and digitized. Elders can access them through radio programs or CDs, while youth can turn to Gwich’in resources available through the internet and social media.
But artificial intelligence and similar computer models could be a useful resource for Gwich’in revitalization, Wright added.
“We need to do as much as we can and try different avenues,” she said, expressing the view that technology alone cannot replace the act of learning a language by being immersed in that language’s culture and on the land.
“Language is definitely tied to the land, our culture, our heritage. I think it would be best and ideal if we were learning with our Elders,” said Wright.
“When you’re learning the language with Elders, you’re also learning history. You’re learning stories when you’re having those conversations with them.”
For people living outside the Gwich’in Settlement Area who do not readily have a connection to Elders, Wright thinks computer models like text-to-speech could be helpful.
“I don’t think we’re in dire straits yet, but we’re just hopeful in revitalizing our language,” says Wright. “And we are fortunate that we still have some speakers.”