Listening to Affected Communities to Define Extreme Speech: Dataset and Experiments

TL;DR

This paper introduces XTREMESPEECH, a multilingual hate speech dataset created with direct input from affected communities, aiming to improve hate speech detection and understanding across diverse cultural contexts.

Contribution

It presents a new dataset involving affected communities in data collection, along with novel tasks, baselines, and analysis of cross-country transferability and model interpretability.

Findings

Cross-country training is generally ineffective due to cultural differences.

The dataset better captures community-specific hate speech.

Interpretability analysis sheds light on model decision processes.

Abstract

Building on current work on multilingual hate speech (e.g., Ousidhoum et al. (2019)) and hate speech reduction (e.g., Sap et al. (2020)), we present XTREMESPEECH, a new hate speech dataset containing 20,297 social media passages from Brazil, Germany, India and Kenya. The key novelty is that we directly involve the affected communities in collecting and annotating the data - as opposed to giving companies and governments control over defining and combatting hate speech. This inclusive approach results in datasets more representative of actually occurring online speech and is likely to facilitate the removal of the social media content that marginalized communities view as causing the most harm. Based on XTREMESPEECH, we establish novel tasks with accompanying baselines, provide evidence that cross-country training is generally not feasible due to cultural differences between countries and perform an interpretability analysis of BERT's predictions.