Scope-conditioned multilingual counterspeech improves stereotype targeting
Deconstructing Stereotypes: Scope-Conditioned Generation for Effective Multilingual Counterspeech
Computation and Language
Summary
Online hate speech often relies on stereotypes, and automatically generated replies tend to be too generic to effectively respond. The authors propose a method that helps language models use specific details about stereotypes when creating counterspeech responses. They tested this method on a new dataset in English, Italian, and Spanish, showing it produces more factual, specific, and persuasive replies. This approach helps make counterspeech better at addressing both obvious and hidden stereotypes behind hateful messages.
What this means in practice
- •For social media platform teams: Create more effective automatic replies that directly address stereotypes in hate speech across multiple languages.
- •For moderation tool developers: Integrate stereotype-aware counterspeech generation to improve response quality in content moderation systems.
Authors
Greta Damo, Elias Urios Alacreu, Elena Cabrio, Paolo Rosso, Serena Villata
Abstract
Counterspeech (CS) - direct responses that counter online Hate Speech (HS) using reasoning and alternative viewpoints - has emerged as an alternative to content removal. Current automatic CS generation methods, however, frequently produce generic, ineffective replies that fail to target the implicit stereotypes behind HS. To bridge this gap, we propose a novel scope-conditioned generation framework that explicitly integrates structured stereotype characteristics into Large Language Models prompts. We validate our approach on a novel, human-curated dataset annotated in English, Italian, and Spanish. Extensive evaluations show that stereotype-conditioned prompting substantially outperforms generic baselines across all three languages, obtaining significant gains in factuality, specificity, cogency, and effectiveness for both explicit and implicit implied stereotypes.