Self-Explaining Hate Speech Detection with Moral Rationales

TL;DR

The paper tackles biases and interpretability gaps in hate speech detection by introducing SMRA, a self-explaining framework that aligns model attention with expert moral rationales drawn from Moral Foundations Theory. It coalesces a normative training objective with moral rationale supervision and evaluates on HateBRMoralXplain, a Brazilian Portuguese dataset enriched with moral categories and token-level rationales. Empirical results show SMRA improves explanation faithfulness and maintains strong classification performance, while large language models struggle with moral category detection unless guided by explicit rationales. The work also provides HateBRMoralXplain as a valuable resource for cross-cultural hate-speech research and highlights the importance of incorporating moral reasoning for robust, transparent moderation in multilingual contexts.

Abstract

Hate speech detection models rely on surface-level lexical features, increasing vulnerability to spurious correlations and limiting robustness, cultural contextualization, and interpretability. We propose Supervised Moral Rationale Attention (SMRA), the first self-explaining hate speech detection framework to incorporate moral rationales as direct supervision for attention alignment. Based on Moral Foundations Theory, SMRA aligns token-level attention with expert-annotated moral rationales, guiding models to attend to morally salient spans rather than spurious lexical patterns. Unlike prior rationale-supervised or post-hoc approaches, SMRA integrates moral rationale supervision directly into the training objective, producing inherently interpretable and contextualized explanations. To support our framework, we also introduce HateBRMoralXplain, a Brazilian Portuguese benchmark dataset annotated with hate labels, moral categories, token-level moral rationales, and socio-political metadata. Across binary hate speech detection and multi-label moral sentiment classification, SMRA consistently improves performance (e.g., +0.9 and +1.5 F1, respectively) while substantially enhancing explanation faithfulness, increasing IoU F1 (+7.4 pp) and Token F1 (+5.0 pp). Although explanations become more concise, sufficiency improves (+2.3 pp) and fairness remains stable, indicating more faithful rationales without performance or bias trade-offs

Figures (1)

• Figure 1: Supervised Moral Rationale Attention (SMRA) for Self-Explaining Hate Speech Detection.