Towards Generalized Offensive Language Identification

TL;DR

This paper presents GenOffense, the first generalized benchmark for offensive language detection, aggregating eight English datasets mapped to a unified OLID taxonomy to evaluate cross-dataset generalization. It systematically compares unsupervised approaches (public APIs and LLM prompting) and supervised models (LSTM and transformers) across diverse platform, language, task, and topic shifts. Key findings show that public APIs and domain-general supervised datasets (notably OLID, AHSD, and TCC) provide stronger cross-dataset generalization than many LLM prompts, with domain-specific datasets like HatE and OHS performing poorly in cross-domain settings. The work highlights the importance of generalization in real-world offensive content moderation and points to future directions, including adversarial testing and multilingual extensions, to build more robust systems.

Abstract

The prevalence of offensive content on the internet, encompassing hate speech and cyberbullying, is a pervasive issue worldwide. Consequently, it has garnered significant attention from the machine learning (ML) and natural language processing (NLP) communities. As a result, numerous systems have been developed to automatically identify potentially harmful content and mitigate its impact. These systems can follow two approaches; (1) Use publicly available models and application endpoints, including prompting large language models (LLMs) (2) Annotate datasets and train ML models on them. However, both approaches lack an understanding of how generalizable they are. Furthermore, the applicability of these systems is often questioned in off-domain and practical environments. This paper empirically evaluates the generalizability of offensive language detection models and datasets across a novel generalized benchmark. We answer three research questions on generalizability. Our findings will be useful in creating robust real-world offensive language detection systems.

Figures (1)

• Figure 1: Few-shot Learning Results for BERT