(New Scientist) The internet has a personality problem. Abusive behaviour is a scourge in most corners of the internet, from website comments sections to social media to chat sessions in online games. Attempts to clamp down have yielded frustrating results – the concept is just too slippery. But a new way of automatically identifying the subtle linguistic fingerprints of true hate speech – and separating it from more benign uses of similar words – could finally help crack down on the worst offenders.
“Hate speech is notoriously difficult to detect,” says Dana Warmsley at Cornell University in New York. Simply using offensive language does not make someone abusive. People swear for all sorts of reasons. Friends call each other names for fun.