arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2605.30133cs.CL
收藏

CorPipe at CRAC 2026: 多语言共指消解中的空节点与跨语言迁移

CorPipe at CRAC 2026: Empty Nodes and Cross-Lingual Transfer in Multilingual Coreference Resolution

Charles University, Faculty of Mathematics and Physics Institute of Formal and Applied Linguistics(查理大学数学与物理系形式与应用语言学研究所)

Milan Straka

AI总结:

本文提出CorPipe 26系统,通过单一模型联合预测空节点、提及和共指链接,在CRAC 2026多语言共指消解共享任务中超越所有其他系统,并在LLM赛道和不受限赛道分别领先2.8和9.5个百分点。

摘要:

我们介绍CorPipe 26,这是我们在CRAC 2026多语言共指消解共享任务中的获胜提交。该共享任务的第五版主要关注生成式LLM与专用系统的比较;此外,还引入了5个更多数据集和2种新语言。CorPipe 26是CorPipe 25的改进版本,具有一种新变体,可在单个模型中同时预测空节点、提及和共指链接。我们的系统在LLM赛道中优于所有其他提交2.8个百分点,在不受限赛道中优于所有提交9.5个百分点。此外,我们进行了一系列消融实验,涉及不同模型大小、空节点预测方法以及跨语言零样本评估。源代码和训练好的模型可在https://github.com/ufal/crac2026-corpipe公开获取。

英文摘要:

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.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑