Tag
The paper proposes a novel scope-conditioned generation framework that integrates structured stereotype characteristics into Large Language Model prompts for effective multilingual counterspeech, validated on a human-curated dataset with significant improvements in factuality and effectiveness.