We are happy to share that DANIS members presented three contributions at LREC-COLING 2026 and its co-located workshops in Palma, Mallorca.
At the main conference, Agnieszka Faleńska presented the poster for the paper "Prompt-Based Stance Control in German: An Evaluation of LLMs for Experimental Research on Attitude Change" co-authored with Florian Omiecienski and Cornelia Sindermann. The work is a collaboration with another IRIS3D group and investigates whether large language models can be reliably steered to adopt predefined political stances in German.
At the 19th Workshop on Building and Using Comparable Corpora (BUCC), Yassir El Attar presented the paper "Leveraging Comparable Toxicity Lexicons in Prompt Instructions for Multilingual Text Detoxification" co-authored with Esra Dönmez, Nina K. Ohlendorf, and Agnieszka Faleńska. The paper explores how comparable toxicity lexicons can be used in prompt instructions to support multilingual text detoxification.
Finally, at the Identity-Aware AI 2026 workshop, Agnieszka Faleńska presented the project "Misaligned Alignment in Socially Embedded LMs: A Review of Sociotechnical Alignment, Conceptual Slippage, and Theoretical Gaps" developed together with Esra Dönmez. The project examines how alignment is conceptualized in socially embedded language models and highlights theoretical gaps in current approaches.
Together, these presentations reflect DANIS' ongoing work on socially aware NLP, multilinguality, toxicity mitigation, political stance modeling, and the broader societal implications of language technologies.
| Contact |
|---|