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

• Published in: arXiv.org 2024-12
• Main Authors: de Wynter, Adrian, Watts, Ishaan, Wongsangaroonsri, Tua, Zhang, Minghui, Farra, Noura, Nektar Ege Altıntoprak, Baur, Lena, Claudet, Samantha, Gajdusek, Pavel, Gören, Can, Gu, Qilong, Kaminska, Anna, Kaminski, Tomasz, Kuo, Ruby, Kyuba, Akiko, Lee, Jongho, Mathur, Kartik, Merok, Petter, Milovanović, Ivana, Paananen, Nani, Paananen, Vesa-Matti, Pavlenko, Anna, Bruno Pereira Vidal, Strika, Luciano, Tsao, Yueh, Turcato, Davide, Vakhno, Oleksandr, Velcsov, Judit, Vickers, Anna, Visser, Stéphanie, Widarmanto, Herdyan, Zaikin, Andrey, Si-Qing, Chen
• Format: Article
• Language: English
• Subjects: Large language models; Multilingualism; Safety; Toxicity
• ISSN: 2331-8422

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.