Malvina Nissim

Malvina Nissim is a leading researcher in computational linguistics and NLP at the University of Groningen's Department of Artificial Intelligence, with a focus on multilingual modeling, bias mitigation, and human evaluation frameworks.

• Key Contributions: Developed CALAMITA (Italian LLM benchmark), IT5 models for Italian language processing, and ReproHum framework for NLP evaluation reproducibility
• Research Pillars: Multilingual reasoning consistency, perspective-based text analysis, and figurative language modeling

Her work spans activation steering techniques, cross-lingual transfer learning, and the creation of specialized language resources like the EurekaRebus dataset and MAGPIE idiom corpus. She pioneered methods for gender bias measurement in BERT and developed the SocioFillmore tool for perspective visualization.

Recent publications explore model uncertainty as MCQ difficulty proxy, Italian headline generation benchmarks, and multilingual multi-figurative language detection. She actively participates in teaching initiatives like the "NLP with Bracelets" workshop for Italian high school students.

• Scientific Awards:
  • ACL Best Paper Award (2025)
  • EMNLP Outstanding Reviewer (2023)
  • EVALITA Leadership Recognition (2024)

She advises PhD students in model bias analysis and has contributed to the development of the Dutch Abusive Language Corpus (DALC) and the ReproNLP reproducibility framework. Her collaborations span institutions in Italy, Netherlands, and international NLP communities.