Language similarity helps hate speech detection cross languages effectively
When Does Hate Transfer? A Large-Scale Study of Cross-Lingual Hate Speech Detection
Computation and LanguageArtificial Intelligence
Summary
Hate speech detection varies across different languages, which can make it hard to keep online spaces safe worldwide. The authors studied 14 languages to see if training a hate speech detector in one language works well in another. They found that the closer two languages are, the better this cross-language transfer works. Other factors like timing and dataset size help a bit, but cultural similarities may also play a role. This study helps decide when it’s practical to use hate speech data from one language to improve detection in another, especially for less common languages.
What this means in practice
- •For content moderation teams: Improve hate speech filters for less common languages using data from similar languages to enhance detection accuracy.
- •For social media platform developers: Design better multilingual hate speech detection systems informed by language and cultural similarities for global user safety.
Authors
Zahra Bokaei, Walid Magdy, Bonnie Webber
Abstract
Effective detection of hate content is essential for maintaining healthy online communities, yet performance can vary substantially across languages. This work investigates a core practical question: Can training data from one language support hate-speech detection in another language with comparable performance? To answer this question, we conduct a large-scale study across 14 languages. \change{We observe consistent patterns in which certain language pairs transfer more effectively than others. Among the factors we measure, linguistic similarity is the strongest predictor of these patterns, with temporal overlap and dataset-size difference playing smaller roles. These measurable factors do not fully explain the observed patterns; a complementary qualitative analysis suggests that shared cultural cues and communicative norms may be associated with some of the remaining cases, particularly within several clusters. We present this cultural interpretation as a hypothesis from qualitative inspection rather than a measured cause.}{} Our analysis includes low-resource languages, offering practical insights into when cross-lingual transfer can support them and where its limitations emerge. These findings inform the design of more reliable multilingual hate-detection systems.