ParsCN: A Persian Dataset for Counter-Narrative Generation to Combat Online Hate Speech

We introduce ParsCN, the first and most comprehensive Persian counter-narrative dataset. Consisting of 1,100 hate speech and counter-narrative pairs, it provides fine-grained annotations across six target groups and six countering strategies, tailored to the socio-cultural context of Persian online discourse. Our dataset serves as a foundational benchmark to advance research on Persian counter-narrative generation and foster safer, more inclusive digital spaces. Our hate speech target categories are as follows: i) Religious: Statements targeting religious groups, such as Islam or the Jewish religionii) Racial: Speech discriminating against ethnic/racial groups like Azeris, Kurds, or Black Africans; iii) Gender: Statements derogatory to men, women, and including activistsiv) Political: Expressions targeting the government, politicians, or country lawsv) Occupational: Statements targeting professions, such as police or teachersvi) National: Speech targeting national or immigrant groups like Afghans, Chinese, or Arabs. Details for countering strategies are as follows: Positive Response: These counter-narratives counter hostility with inclusive, supportive statements that encourage empathy and unity. For example, in response to gender-based hate, they might advocate mutual respect in relationships to promote harmony. This approach seeks to reframe divisive rhetoric into constructive dialogue. Counter questions: This type uses probing questions to challenge the biases or assumptions in hate speech, urging reflection. For instance, questioning claims about Sunni minorities’ ambitions can prompt reconsideration of stereotypes, fostering critical thinking. Denouncing: These statements firmly reject harmful rhetoric, condemning its ethical or social damage. For example, denouncing slurs against religious groups as divisive aims to curb their spread by highlighting their harm. Fact-based: This approach refutes misinformation with credible evidence, enhancing discourse accuracy. For instance, citing statistics to debunk myths about women’s driving in Iran corrects false narratives and builds trust in factual dialogue. Warning of consequences: These counter-narratives caution against the adverse effects of hate speech, such as social division from targeting Afghan immigrants. By emphasizing risks, they encourage more responsible perspectives. Contradiction: This type exposes inconsistencies in hate speech, such as criticizing women’s autonomy while ignoring men’s. By highlighting logical flaws, it undermines the credibility of harmful claims.