Anja Belz

Anja Belz is a Professor at Dublin City University's School of Computing, specializing in Natural Language Processing research. She leads the Natural Language Processing Research Group and has established herself as a leading expert in human evaluation methodologies, reproducibility in NLP, and data-to-text generation systems. Her work bridges theoretical research and practical applications with significant contributions to medical text generation and evaluation standards.

Her research interests center on creating robust evaluation frameworks for NLP systems, with particular focus on human evaluation methodologies, reproducibility assessment, and quality criteria standardization. She has pioneered work on the Human Evaluation Data Sheet (HEDS) and the QCET (Quality Criteria for Evaluation Taxonomy), addressing critical gaps in evaluation comparability across NLP research. Her work on reproducibility spans multiple shared tasks (ReproNLP, ReproGen) that have become benchmarks in the field, examining how different experimental conditions affect evaluation outcomes.

Analysis of her recent publications reveals a clear research trajectory focused on making NLP evaluation more rigorous, transparent, and comparable. Her work increasingly incorporates large language models while maintaining critical scrutiny of their evaluation methodologies. She has made significant contributions to understanding when LLM-based evaluation correlates with human judgments, and has developed frameworks for assessing the reproducibility of NLP evaluation results in quantified terms.

Professor Belz has organized numerous workshops and shared tasks focused on human evaluation and reproducibility in NLP, including multiple ReproNLP shared tasks that have attracted international participation. Her research has been consistently published in top-tier NLP conferences including ACL, EMNLP, and INLG, with a strong emphasis on methodological rigor and practical applicability to real-world NLP evaluation challenges. She has also contributed significantly to NLP research in under-resourced languages, particularly Irish, Welsh, Breton, and Maltese.

Her work on medical text generation has practical implications for healthcare applications, particularly in automating clinical documentation and systematic reviews. She has developed methods for biomedical synthesis generation that could significantly reduce the time and cost of keeping medical practitioners updated with research. Her research on consultation checklists aims to standardize the human evaluation of medical note generation systems, addressing critical challenges in clinical safety and quality assessment.