RTP-LX: Can LLMs Evaluate Toxicity in Multilingual Scenarios?

TL;DR

This paper introduces RTP-LX, a multilingual, human-annotated dataset for evaluating LLMs' ability to detect toxicity across 28 languages, highlighting current models' limitations in cultural sensitivity and context understanding.

Contribution

The paper presents RTP-LX, a novel multilingual toxicity evaluation dataset with cultural considerations, and assesses LLMs' performance, revealing gaps in holistic and context-aware toxicity detection.

Findings

Models have acceptable accuracy but low agreement with human judgments.

Difficulty in detecting subtle, context-dependent harm such as microaggressions.

Dataset aims to improve multilingual safety evaluations of LLMs.

Abstract

Large language models (LLMs) and small language models (SLMs) are being adopted at remarkable speed, although their safety still remains a serious concern. With the advent of multilingual S/LLMs, the question now becomes a matter of scale: can we expand multilingual safety evaluations of these models with the same velocity at which they are deployed? To this end, we introduce RTP-LX, a human-transcreated and human-annotated corpus of toxic prompts and outputs in 28 languages. RTP-LX follows participatory design practices, and a portion of the corpus is especially designed to detect culturally-specific toxic language. We evaluate 10 S/LLMs on their ability to detect toxic content in a culturally-sensitive, multilingual scenario. We find that, although they typically score acceptably in terms of accuracy, they have low agreement with human judges when scoring holistically the toxicity of a prompt; and have difficulty discerning harm in context-dependent scenarios, particularly with subtle-yet-harmful content (e.g. microaggressions, bias). We release this dataset to contribute to further reduce harmful uses of these models and improve their safe deployment.