CorPipe at CRAC 2026: Empty Nodes and Cross-Lingual Transfer in Multilingual Coreference Resolution
For researchers in multilingual coreference resolution, this work provides a state-of-the-art specialized system that surpasses generative LLMs, though it is an incremental improvement over the previous CorPipe 25.
CorPipe 26 won the CRAC 2026 Shared Task on Multilingual Coreference Resolution, outperforming LLM track submissions by 2.8% and unconstrained track submissions by 9.5% through an improved single-model architecture that jointly predicts empty nodes, mentions, and coreference links.
We introduce CorPipe 26, our winning submission to the CRAC 2026 Shared Task on Multilingual Coreference Resolution. The fifth edition of this shared task focuses mainly on the comparison of generative LLMs and specialized systems; additionally, 5 more datasets and 2 new languages are introduced. CorPipe 26 is an improved version of CorPipe 25, with a new variant predicting empty nodes together with mentions and coreference links in a single model. Our system outperforms all other submissions in the LLM track by 2.8 percent points and all submissions in the unconstrained track by 9.5 percent points. Furthermore, we perform a series of ablation experiments with different model sizes, empty node prediction methods, and cross-lingual zero-shot evaluation. The source code and the trained models are publicly available at https://github.com/ufal/crac2026-corpipe.