CorPipe at CRAC 2025: Evaluating Multilingual Encoders for Multilingual Coreference Resolution

2025-09-22

AI Summary

This study addresses multilingual coreference resolution by proposing an efficient dual-mode system built upon multilingual pretrained encoders, supporting both LLM-guided and unconstrained inference paradigms. Methodologically, we perform a full-stack reconstruction from TensorFlow to PyTorch, integrate multilingual encoders, and adapt the CRAC 2025 reduced dataset. Crucially, we introduce a unified modeling framework that preserves lightweight design while enhancing cross-lingual generalization. In the CRAC 2025 shared task, our system achieves first place in both the LLM-guided and unconstrained tracks, with F1 scores exceeding the second-best systems by 8 percentage points—substantially outperforming all baselines. To foster reproducibility and extensibility, we fully open-source our code, models, and configurations, establishing a new, accessible benchmark for multilingual coreference resolution.

Technology Category

• Natural Language Processing: Machine Translation, Multilinguality, Cross-Lingual NLP
• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Multi-modal Vision

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

We present CorPipe 25, the winning entry to the CRAC 2025 Shared Task on Multilingual Coreference Resolution. This fourth iteration of the shared task introduces a new LLM track alongside the original unconstrained track, features reduced development and test sets to lower computational requirements, and includes additional datasets. CorPipe 25 represents a complete reimplementation of our previous systems, migrating from TensorFlow to PyTorch. Our system significantly outperforms all other submissions in both the LLM and unconstrained tracks by a substantial margin of 8 percentage points. The source code and trained models are publicly available at https://github.com/ufal/crac2025-corpipe.

Problem

• Evaluating multilingual encoders for coreference resolution across languages
• Developing systems for multilingual coreference resolution in shared tasks
• Comparing LLM-based and unconstrained approaches to coreference resolution

Innovation

• Migrated system from TensorFlow to PyTorch
• Evaluated multilingual encoders for coreference resolution
• Implemented complete reimplementation of previous systems