Research on abusive content detection on social media has primarily focused on explicit forms of hate speech (HS), that are often identifiable by recognizing hateful words and expressions. Messages containing linguistically subtle and implicit forms of hate speech still constitute an open challenge for automatic hate speech detection. In this paper, we propose a new framework for generating adversarial implicit HS short-text messages using Auto-regressive Language Models. Moreover, we propose a strategy to group the generated implicit messages in complexity levels (EASY, MEDIUM, and HARD categories) characterizing how challenging these messages are for supervised classifiers. Finally, relying on (Dinan et al., 2019; Vidgen et al., 2021), we propose a “build it, break it, fix it”, training scheme using HARD messages showing how iteratively retraining on HARD messages substantially leverages SOTA models’ performances on implicit HS benchmarks.

Research on automatic hate speech (HS) detection has mainly focused on identifying explicit forms of hateful expressions on user-generated content. Recently, a few works have started to investigate methods to address more implicit and subtle abusive content. However, despite these efforts, automated systems still struggle to correctly recognize implicit and more veiled forms of HS. As these systems heavily rely on proper textual representations for classification, it is crucial to investigate the differences in embedding implicit and explicit messages. Our contribution to address this challenging task is fourfold. First, we present a comparative analysis of transformer-based models, evaluating their performance across five datasets containing implicit HS messages. Second, we examine the embedding representations of implicit messages across different targets, gaining insight into how veiled cases are encoded. Third, we compare and link explicit and implicit hateful messages across these datasets through their targets, enforcing the relation between explicitness and implicitness and obtaining more meaningful embedding representations. Lastly, we show how these newer representation maintains high performance on HS labels, while improving classification in borderline cases.

The research carried out so far in detecting abusive content in social media has primarily focused on overt forms of hate speech. While explicit hate speech (HS) is more easily identifiable by recognizing hateful words, messages containing linguistically subtle and implicit forms of HS (as circumlocution, metaphors and sarcasm) constitute a real challenge for automatic systems. While the sneaky and tricky nature of subtle messages might be perceived as less hurtful with respect to the same content expressed clearly, such abuse is at least as harmful as overt abuse. In this paper, we first provide an in-depth and systematic analysis of 7 standard benchmarks for HS detection, relying on a fine-grained and linguistically-grounded definition of implicit and subtle messages. Then, we experiment with state-of-the-art neural network architectures on two supervised tasks, namely implicit HS and subtle HS message classification. We show that while such models perform satisfactory on explicit messages, they fail to detect implicit and subtle content, highlighting the fact that HS detection is not a solved problem and deserves further investigation.