Multilingual moderation systems often need to account for the possibility that online incivility is expressed differently across languages, beyond differences in vocabulary or other surface-level linguistic features. To examine this issue, we analyze a manually annotated bilingual corpus of 800 YouTube comments (400 English and 400 Russian) labeled across four dimensions: incivility type, severity, target, and clarity of expression. While severity does not differ significantly between the two corpora (Cramér’s V = .06), significant differences emerge for type, target, and clarity (V = .13, .22, and .10, respectively). English comments contain more abusive and explicit forms of incivility, whereas Russian comments more frequently target individuals and employ implicit expression, including sarcasm and irony. User engagement analyzes further show that the forms of incivility associated with greater visibility differ across the two corpora. Together, these findings suggest that, within the corpora examined, severity is comparatively stable across languages, whereas communicative dimensions exhibit greater cross-linguistic variation. The results suggest that multilingual moderation may benefit from distinguishing comparatively stable dimensions from language-sensitive communicative dimensions when developing annotation guidelines and moderation resources.

Online objectification of children can harm their self-image and influence how others perceive them. Objectifying comments may start with a focus on appearance but also include language that treats children as passive, decorative, or lacking agency. On TikTok, algorithm-driven visibility amplifies this focus on looks. Drawing on objectification theory, we introduce a Child Objectification Language Typology to automatically classify objectifying comments. Our dataset consists of 562,508 comments from 9,090 videos across 482 TikTok accounts. We compare language models of different complexity, including an n-gram-based model, RoBERTa, GPT-4, LlaMA, and Mistral. On our training dataset of 6,000 manually labeled comments, we found that RoBERTa performed best overall in detecting appearance- and objectification-related language. 10.35% of comments contained appearance-related language, while 2.90% included objectifying language. Videos with school-aged girls received more appearance-related comments compared to boys in that age group, while videos with toddlers show a slight increase in objectification-related comments compared to other age groups. Neither gender alone nor engagement metrics showed significant effects.The findings raise concerns about children’s digital exposure, emphasizing the need for stricter policies to protect minors.

This paper introduces GENTRAC, an open-access web-based tool built to interactively detect and analyze potentially traumatic content in witness statements of genocide and mass atrocity trials. Harnessing recent developments in natural language processing (NLP) to detect trauma, GENTRAC processes and formats court transcripts for NLP analysis through a sophisticated parsing algorithm and detects the likelihood of traumatic content for each speaker segment. The tool visualizes the density of such content throughout a trial day and provides statistics on the overall amount of traumatic content and speaker distribution. Capable of processing transcripts from four prominent international criminal courts, including the International Criminal Court (ICC), GENTRAC’s reach is vast, tailored to handle millions of pages of documents from past and future trials. Detecting potentially re-traumatizing examination methods can enhance the development of trauma-informed legal procedures. GENTRAC also serves as a reliable resource for legal, human rights, and other professionals, aiding their comprehension of mass atrocities’ emotional toll on survivors.