Gibberish in, Gibberish out, makes us all want to shout!

Based on this article by Jacob Judah posted on Technology Review.

Many of Wikipedia’s smaller-language editions are being swamped with poor-quality, machine-translated articles created by non-speakers, a problem so severe it forced the closure of the Greenlandic version. This has created a "linguistic doom loop" where AI models train on this flawed Wikipedia content, which in turn degrades their translation accuracy for these vulnerable languages.

People then use these inaccurate AI tools to generate even more error-filled articles, creating a "garbage in, garbage out" cycle that pollutes the digital information ecosystem. Without dedicated communities to manually create high-quality content, this process threatens the digital survival of minority languages by embedding inaccuracies into the very AI systems meant to support them.

Check out the article.

Related Stories