Targeted counterspeech generation

Abstract

Counterspeech is regarded as an effective method to counter or negate hate speech online. Various vanilla and controlled counterspeech generation strategies have been investigated in recent years, but often a single counterspeech for a given hostile instance is insufficient. Hate speech can be a complicated matter to handle, and different forms of counter-narratives might prove to be helpful in different situations. We seek to begin research into the efficacy of this technique by curating the first diversified intent-specific counterspeech dataset and developing a novel methodology for targeted counterspeech generation.