Vicarious Offense and Noise Audit of Offensive Speech Classifiers: Unifying Human and Machine Disagreement on What is Offensive

TL;DR

This paper explores the extensive disagreement between human and machine moderators on offensive speech, revealing political and subjective influences, and introduces a novel dataset of vicarious offense to better understand these disagreements.

Contribution

It provides the first large-scale noise audit of offensive speech classifiers and introduces a dataset capturing vicarious offense influenced by political leanings.

Findings

Moderation outcomes vary widely across different machine classifiers.

Political leanings and sensitive issues significantly influence offense perception.

Humans and models struggle to predict each other's responses on offensiveness.

Abstract

Offensive speech detection is a key component of content moderation. However, what is offensive can be highly subjective. This paper investigates how machine and human moderators disagree on what is offensive when it comes to real-world social web political discourse. We show that (1) there is extensive disagreement among the moderators (humans and machines); and (2) human and large-language-model classifiers are unable to predict how other human raters will respond, based on their political leanings. For (1), we conduct a noise audit at an unprecedented scale that combines both machine and human responses. For (2), we introduce a first-of-its-kind dataset of vicarious offense. Our noise audit reveals that moderation outcomes vary wildly across different machine moderators. Our experiments with human moderators suggest that political leanings combined with sensitive issues affect both first-person and vicarious offense. The dataset is available through https://github.com/Homan-Lab/voiced.