The race to save indigenous languages using automatic speech recognition

Michael Running Wolf inactive has that aged TI-89 graphing calculator helium utilized successful precocious schoolhouse that helped propel his involvement successful technology.

"Back then, my teachers saw I was truly funny successful it," says Running Wolf, objective teacher of machine subject astatine Northeastern University. "Actually a mates of them printed retired hundreds of pages of instructions for maine connected however to code" the instrumentality truthful that it could play games.

What Running Wolf, who grew up successful a distant Cheyenne colony successful Birney, Montana, didn't recognize astatine the time, poring implicit the stack of printouts astatine location by the airy of kerosene lamps, was that helium was really teaching himself basal programming.

"I thought I was conscionable learning however to enactment machine games connected my calculator," Running Wolf says with a laugh.

But it hadn't been his archetypal brushwood with technology. Growing up successful the windy plains adjacent the Northern Cheyenne Indian Reservation, Running Wolf says that though his family—which is portion Cheyenne, portion Lakota—didn't person regular entree to moving h2o oregon electricity, sometimes, erstwhile the winds died down, the powerfulness would flicker on, and he'd plug successful his Atari console and play games with his sisters.

These aboriginal experiences would spur guardant a lifelong involvement successful computers, artificial intelligence, and software engineering that Running Wolf is present harnessing to assistance reawaken endangered indigenous languages successful North and South America, immoderate of which are truthful critically astatine hazard of extinction that their tallies of surviving native speakers person dwindled into the azygous digits.

Running Wolf's extremity is to make methods for documenting and maintaining these aboriginal languages done automatic code designation software, helping to support them "alive" and well-documented. It would beryllium a process, helium says, that tribal and indigenous communities could usage to supplement their ain language reclamation efforts, which person intensified successful caller years amid the threats facing languages.

"The grandiose plan, the far-off dream, is we tin make exertion to not lone preserve, but reclaim languages," says Running Wolf, who teaches machine subject astatine Northeastern's Vancouver campus. "Preservation isn't what we want. That's similar taking thing and embalming it and putting it successful a museum. Languages are surviving things."

The amended happening to accidental is that they've "gone to sleep," Running Wolf says.

And the threats to indigenous languages are real. Of the astir 6,700 languages spoken successful the world, astir 40 percent are successful information of atrophying retired of beingness forever, according to UNESCO Atlas of Languages successful Danger. The nonaccomplishment of these languages besides represents the nonaccomplishment of full systems of knowledge unsocial to a culture, and the quality to transmit that cognition crossed generations.

While the concern appears dire—and is, successful galore cases—Running Wolf says astir each Native American people is engaged successful connection reclamation efforts. In New England, 1 notable people doing truthful is the Mashpee Wampanoag Tribe, whose autochthonal lingua is present being taught successful nationalist schools connected Cape Cod, Massachusetts.

But the problem, helium says, is that successful the ever-evolving tract of computational linguistics, small probe has been devoted to Native American languages. This is partially owed to a deficiency of linguistic data, but it is besides due to the fact that galore autochthonal languages are "polysynthetic," meaning they incorporate words that comprise galore morphemes, which are the smallest units of meaning successful language, Running Wolf says.

Polysynthetic languages often person precise agelong words—words that tin mean an full sentence, oregon denote a sentence's worthy of meaning.

Further complicating the effort is the information that galore Native American languages don't person an orthography, oregon an alphabet, helium says. In presumption of what languages request to support them afloat, Running Wolf maintains that orthographies are not vital. Many indigenous languages person survived done a beardown oral contented successful lieu of a robust written one.

But for scholars looking to physique databases and transcription methods, similar Running Wolf, written texts are important to filling successful the gaps. What's holding researchers backmost from gathering automatic code designation for indigenous languages is precisely that determination is simply a deficiency of audio and textual information disposable to them.

Using hundreds of hours of audio from assorted tribes, Running Wolf has managed to nutrient immoderate rudimentary results. So far, the automatic code recognition bundle helium and his squad person developed tin admit single, elemental words from immoderate of the indigenous languages they person information for.

"Right now, we're gathering a corpus of audio and texts to commencement showing aboriginal results," Running Wolf says.

Importantly, helium says, "I deliberation we person an attack that's scientifically sound."

Eventually, Running Wolf says helium hopes to make a mode for tribes to supply their younker with tools to larn these past languages by mode of technological immersion—through things similar augmented oregon virtual reality, helium says.

Some of these technologies are already nether improvement by Running Wolf and his team, made up of a linguist, a information scientist, a instrumentality learning engineer, and his wife, who utilized to beryllium a programme manager, among others. All of the ongoing probe and improvement is being done successful consultation with galore tribal communities, Running Wolf says.

"It's each coming from the people," helium says. "They privation to enactment with us, and we're doing the champion to respect their cognition systems."

---

---