Three papers by TRAILS authors accepted to IJCNLP-AACL 2025

25. Nov 2025 1 min read

Three papers from researchers in the TRAILS project have been accepted at the 2025 International Joint Conference on Natural Language Processing & Asia-Pacific Chapter of the Association for Computational Linguistics (IJCNLP-AACL 2025). IJCNLP-AACL will take place December 20-24 in Mumbai, India. The first paper, titled “Language Arithmetics: Towards Systematic Language Neuron Identification and Manipulation”, analyzes language-specific neurons in LLMs to identify neurons that control language behaviour. The second paper presents a large-scale study of political orientation in modern open-source instruction-tuned LLMs across 14 languages. The third paper investigates an approach for extremely compressing multilingual encoder-only language models for low-resource languages.

Daniil Gurgurov, Katharina Trinley, Yusser Al Ghussin, Tanja Bäumel, Josef van Genabith, Simon Ostermann (2025). Language Arithmetics: Towards Systematic Language Neuron Identification and Manipulation. IJCNLP-AACL 2025.

PDF Cite URL

Daniil Gurgurov, Katharina Trinley, Ivan Vykopal, Josef van Genabith, Simon Ostermann, Roberto Zamparelli (2025). Multilingual Political Views of Large Language Models: Identification and Steering. IJCNLP-AACL 2025.

PDF Cite URL

Daniil Gurgurov, Michal Gregor, Josef van Genabith, Simon Ostermann (2025). On Multilingual Encoder Language Model Compression for Low-Resource Languages. IJCNLP-AACL 2025.

PDF Cite URL