How AI and Wikipedia sent vulnerable languages ​​into the spiral of destiny

Machine Learning


Iuara, who currently works as a professional translator between English and Igbo, said that the most damaging users are inexperienced users and that AI translations see Ibowikipedia as a way to quickly increase their profile. She often finds herself having to email the online editor she organizes, or to various error-prone editors, explaining that the outcome is the exact opposite and that it can push users aside.

These fears are reflected by Noah Haririo Solomon, assistant professor of Hawaiian language at the University of Hawaii. He reports that around 35% of the words on some pages of Wikipedia in Hawaii are not able to understand. “If this is a Hawaiian who exists online, it would do more harm than anything,” he says.

Hawaiians, who were on the verge of extinction decades ago, have been making recovery efforts led by indigenous activists and scholars. Seeing such poor Hawaiians on widely used platforms like Wikipedia is upset by Harririo Solomon.

“It's painful because it always reminds us that our culture and language is being allocated,” he says. “We've been fighting teeth and claws uphill to activate language. There's nothing easy about that. This can add an extra obstacle.

The results of all these Wikipedia errors may be immediately apparent. AI translators who have undoubtedly taken these pages in their training data are helping to produce books generated by AI-sparked with errors targeting learners of languages ​​as diverse as Inuktit and Cree, the Indigenous language spoken in Canada, Manx, and the small Celtic language spoken on the island of Man. Many of these are available on Amazon. “It was total nonsense,” says Richard Compton, a linguist at the University of Quebec in Montreal, of the volume he reviewed, claiming it was a phrasebook for Inuctitut's primer book.

Rather than making minority languages ​​more accessible, AI is now creating an ever-growing minefield for students and speakers in these languages ​​to navigate. “It's a slap in the face,” says Compton. He worries about the younger generation in Canada, hoping to learn the language in a community that is fighting discrimination to take over its legacy, and may simply exacerbate the problem by turning to online tools such as Amazon's ChatGpt and Fhasebook. “It's a scam,” he says.

Competition with time

According to UNESCO, the language is declared extinct every two weeks. However, whether the Wikimedia Foundation, which runs Wikipedia, has an obligation to the language used on its platform is an open question. When I spoke to the foundation's senior director, Runa Bhattacharjee, she said it's up to the individual community to decide what content they want to be present on Wikipedia. “Ultimately, the responsibility lies with the community and through machine translation and other means to ensure there are no vandalism or unwanted activities,” she said. Usually, Bhattacharjee added that it added editions for closure only if certain complaints were raised about them.

But if there is no active community, how can I fix the edition or even raise a complaint?



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *