Toxicity Classification in Ukrainian

• DOI: 10.48550/arxiv.2404.17841
• Publication Date: 2024-04-27
• Authors: Daryna Dementieva; Valeriia Khylenko; Nikolay Babakov; Georg Groh
• Keywords: Ukrainian

The task of toxicity detection is still a relevant task, especially in the context safe and fair LMs development. Nevertheless, labeled binary classification corpora are not available for all languages, which understandable given resource-intensive nature annotation process. Ukrainian, particular, among languages lacking such resources. To our knowledge, there has been no existing corpus Ukrainian. In this study, we aim to fill gap by investigating cross-lingual knowledge transfer techniques creating by: (i)~translating from an English corpus, (ii)~filtering toxic samples using keywords, (iii)~annotating with crowdsourcing. We compare LLMs prompting other approaches without fine-tuning offering insights into most robust efficient baselines.