Compositional Generalisation for Explainable Hate Speech Detection
This addresses the challenge of online content moderation by improving model generalization, though it is incremental as it builds on existing methods with new data and benchmarks.
The paper tackles the problem of hate speech detection models struggling to generalize beyond training data due to biases and context-dependent label meanings, and shows that training on a synthetic dataset (U-PLEAD) combined with real data improves compositional generalization and achieves state-of-the-art performance on the PLEAD benchmark.
Hate speech detection is key to online content moderation, but current models struggle to generalise beyond their training data. This has been linked to dataset biases and the use of sentence-level labels, which fail to teach models the underlying structure of hate speech. In this work, we show that even when models are trained with more fine-grained, span-level annotations (e.g., "artists" is labeled as target and "are parasites" as dehumanising comparison), they struggle to disentangle the meaning of these labels from the surrounding context. As a result, combinations of expressions that deviate from those seen during training remain particularly difficult for models to detect. We investigate whether training on a dataset where expressions occur with equal frequency across all contexts can improve generalisation. To this end, we create U-PLEAD, a dataset of ~364,000 synthetic posts, along with a novel compositional generalisation benchmark of ~8,000 manually validated posts. Training on a combination of U-PLEAD and real data improves compositional generalisation while achieving state-of-the-art performance on the human-sourced PLEAD.