HatePrototypes: Interpretable and Transferable Representations for Implicit and Explicit Hate Speech Detection

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Proskurina, Irina, Carpentier, Marc-Antoine, Velcin, Julien
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910104062787584
author Proskurina, Irina
Carpentier, Marc-Antoine
Velcin, Julien
author_facet Proskurina, Irina
Carpentier, Marc-Antoine
Velcin, Julien
contents Optimization of offensive content moderation models for different types of hateful messages is typically achieved through continued pre-training or fine-tuning on new hate speech benchmarks. However, existing benchmarks mainly address explicit hate toward protected groups and often overlook implicit or indirect hate, such as demeaning comparisons, calls for exclusion or violence, and subtle discriminatory language that still causes harm. While explicit hate can often be captured through surface features, implicit hate requires deeper, full-model semantic processing. In this work, we question the need for repeated fine-tuning and analyze the role of HatePrototypes, class-level vector representations derived from language models optimized for hate speech detection and safety moderation. We find that these prototypes, built from as few as 50 examples per class, enable cross-task transfer between explicit and implicit hate, with interchangeable prototypes across benchmarks. Moreover, we show that parameter-free early exiting with prototypes is effective for both hate types. We release the code, prototype resources, and evaluation scripts to support future research on efficient and transferable hate speech detection.
format Preprint
id arxiv_https___arxiv_org_abs_2511_06391
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle HatePrototypes: Interpretable and Transferable Representations for Implicit and Explicit Hate Speech Detection
Proskurina, Irina
Carpentier, Marc-Antoine
Velcin, Julien
Computation and Language
Artificial Intelligence
Optimization of offensive content moderation models for different types of hateful messages is typically achieved through continued pre-training or fine-tuning on new hate speech benchmarks. However, existing benchmarks mainly address explicit hate toward protected groups and often overlook implicit or indirect hate, such as demeaning comparisons, calls for exclusion or violence, and subtle discriminatory language that still causes harm. While explicit hate can often be captured through surface features, implicit hate requires deeper, full-model semantic processing. In this work, we question the need for repeated fine-tuning and analyze the role of HatePrototypes, class-level vector representations derived from language models optimized for hate speech detection and safety moderation. We find that these prototypes, built from as few as 50 examples per class, enable cross-task transfer between explicit and implicit hate, with interchangeable prototypes across benchmarks. Moreover, we show that parameter-free early exiting with prototypes is effective for both hate types. We release the code, prototype resources, and evaluation scripts to support future research on efficient and transferable hate speech detection.
title HatePrototypes: Interpretable and Transferable Representations for Implicit and Explicit Hate Speech Detection
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2511.06391