DDIS Research at EMNLP 2026 in Budapest
Researchers from the Dynamic and Distributed Information Systems (DDIS) group presented their work at EMNLP 2026 in Budapest, contributing to research on language models, text representations, and AI interpretability. A highlight was the Best Reproducibility Paper Award at the BlackboxNLP workshop.
The Conference on Empirical Methods in Natural Language Processing (EMNLP) 2026 brought researchers from around the world to Budapest to present advances in natural language processing and artificial intelligence. Researchers from the Dynamic and Distributed Information Systems (DDIS) group contributed to the conference's Findings track and associated workshops, presenting work on language models, text representations, and AI interpretability.
Selene Báez Santamaría presented “Holmes-Graph: A Framework for Probing the Linguistic Competence of Graph-Based Text Embeddings”, which appeared in Findings of EMNLP. The paper evaluates the linguistic capabilities of graph-based text embeddings. Together with Daan van der Weijden, she also presented “Reproducing and Evaluating the Generalizability of Subliminal Learning in Open-Weight Models” at the BlackboxNLP workshop. This research investigates whether subliminal learning effects can be reproduced and generalised across open-weight language models.
Noah Mamié and Laurin van den Bergh presented “Reproducing Transparent and Scrutable Recommendations: Exploring Open-Weight Models via Natural-Language User Profiles” at BlackboxNLP. Their work examines the reproducibility of recommendation approaches that use natural-language descriptions of user preferences to make recommendations more transparent and understandable.
Stefan F. Schouten presented a shared task report at the Conference on Machine Translation (WMT). “Probe Before you Answer! Answerability Probing for Abstention in Cross-Lingual Context-Based Question Answering” investigates how language models can identify questions they cannot reliably answer, particularly across languages.
A particular highlight was Stefan's paper “Probing Without Labels: Contrast-only Reproductions of Interpretability Results”. The paper received the Best Reproducibility Paper Award at the BlackboxNLP workshop, recognising its valuable contribution.
Together, these research activities reflect contributions in natural language processing, language model evaluation, and AI interpretability.
Further information: