More Than Sum of Its Parts: Deciphering Intent Shifts in Multimodal Hate Speech Detection

TL;DR

This paper introduces a new benchmark and a reasoning framework to improve multimodal hate speech detection by understanding complex intent shifts between visual and textual content.

Contribution

It presents the H-VLI benchmark for nuanced intent detection and the ARCADE framework that uses simulated debate to enhance model reasoning capabilities.

Findings

ARCADE outperforms state-of-the-art baselines on H-VLI

The approach improves detection of implicit hate speech cases

Code and data are publicly available at the provided GitHub link

Abstract

Combating hate speech on social media is critical for securing cyberspace, yet relies heavily on the efficacy of automated detection systems. As content formats evolve, hate speech is transitioning from solely plain text to complex multimodal expressions, making implicit attacks harder to spot. Current systems, however, often falter on these subtle cases, as they struggle with multimodal content where the emergent meaning transcends the aggregation of individual modalities. To bridge this gap, we move beyond binary classification to characterize semantic intent shifts where modalities interact to construct implicit hate from benign cues or neutralize toxicity through semantic inversion. Guided by this fine-grained formulation, we curate the Hate via Vision-Language Interplay (H-VLI) benchmark where the true intent hinges on the intricate interplay of modalities rather than overt visual or textual slurs. To effectively decipher these complex cues, we further propose the Asymmetric Reasoning via Courtroom Agent DEbate (ARCADE) framework. By simulating a judicial process where agents actively argue for accusation and defense, ARCADE forces the model to scrutinize deep semantic cues before reaching a verdict. Extensive experiments demonstrate that ARCADE significantly outperforms state-of-the-art baselines on H-VLI, particularly for challenging implicit cases, while maintaining competitive performance on established benchmarks. Our code and data are available at: https://github.com/Sayur1n/H-VLI