New Publication in Computational Communication Research

Emese Domahidi from the CCS group has published a new open-access article together with an international team of authors titled “Gender Representation in Large Language Models: A Cross-Linguistic and Cross-Model Analysis” in the journal Computational Communication Research.

Here's what it's all about: 

The representation of gender in large language models (LLMs) can reflect and reinforce existing sociocultural inequalities. However, the nature of these biases may vary significantly across languages, influenced by linguistic features as well as the models’ training data. In this study, the team examined gender representation in 24 open-weight LLMs across six linguistically diverse languages: English, German, Russian, Czech, Albanian, and Serbian. Going beyond binary frameworks, the study also incorporates nonbinary individuals. The authors analyze associations between different gender categories and psychometrically validated stereotype dimensions, including agency, communality, dominance, weakness, and giftedness. The results show that traditional gender stereotypes persist, albeit with varying strength across languages and models, while associations involving nonbinary identities display substantial cross-linguistic variation. Additional temperature analyses indicate that these patterns are deeply embedded in the models’ parameters rather than being mere artifacts of sampling procedures. Overall, the findings suggest that both contextual (e.g., language) and technical factors shape the identification and potential mitigation of gender bias in LLMs.

More details about the paper: https://doi.org/10.5117/CCR2026.2.11.URMA