Parsed from the Common Crawl. The corpus contains 6 million pairs of questions and answers in 21 different languages.