Community Discussion · Policy

Behind 50 languages: Who is secretly polluting the corpus?

Can't Finish Reading PapersAug 132026/08/13 210 views

Lately, when I'm tired of reading papers in the lab, I like browsing tech news to unwind. Today I saw a headline saying the browser extension Web Scraper used AI to translate itself into 50 languages in one go. At first glance, it sounds pretty cool—expanding a tool from English to half the globe's languages instantly lowers the barrier to entry. But right after that, I read another study claiming that Amazon scientists found over half of the sentences online are now translated into two or more languages, with a large portion being AI-translated. Looking at these two pieces of news together, something suddenly felt off.

Policy

2 replies

Ctrl + Enter to reply

He Ma Chu Lai DeAug 13

Lmao, I just tried this plugin for some minor languages a couple of days ago, and the translations looked like raw machine translation drafts... Some words were completely off. Community corrections? By the time someone notices, it'll be too late.

Feng sirAug 13

Disagree, this conclusion is too absolute. I ran Claude for a while on minor language translations, and the quality wasn't as bad as OP worried. AI translation can actually build the skeleton first, making manual fixes much faster than translating from scratch. Corpus contamination is an issue, but blaming it all on AI isn't fair.