Linguists at the University of Oslo are working with artificial intelligence experts to uncover the fake news language they call Fakespeak. The researchers say the main goal of the research is to improve existing confirmation tools.

How the research was done

In 2003, New York Times journalist Jason Blair was exposed for producing a series of news articles about American soldiers during the Iraq War. The journalist tried to present the war in a positive way and show the military heroes as much as possible. Scientists collected these fake texts and compared them to a series of real news written by Blair. The texts, which became only the beginning of a larger analysis, turned out to be really different in style.

Researchers found several major language differences:

• Fictional texts had a more informal style, while honest texts contained a higher information density.
• The use of nouns and substitute words is more common in authentic texts. On average, words are longer.
• Verbs are used more often in pseudotexts, especially in the present tense. Also, pronouns, adjectives, emotionally colored words and exclamations were more common.

He also uses fewer metaphors in his fake news articles than when he writes the truth.
– says project manager Silje Suzanne Alvestad.

However, Blair often uses linguistic elements that describe or attempt to evoke positive emotions, which is unusual for fake news that tends to scare. The researcher says it may be relevant.

Scientists go further

Since only 80 pages of Jason Blair’s texts were written in total, the machine learning experts decided to plot on a larger dataset. They added texts from various authors from verification services.

Now researchers are trying to identify the linguistic features of fake news in other languages. If it is possible to prove a significant similarity, they are confident that it will provide a powerful tool to fight counterfeit products on the Internet.