Closing the Gap at CRAC 2026: Two-Stage Adaptation for LLM-Based Multilingual Coreference Resolution

• URL: http://arxiv.org/abs/2605.16984v2
• Date: Thu, 21 May 2026 14:20:53 GMT
• Title: Closing the Gap at CRAC 2026: Two-Stage Adaptation for LLM-Based Multilingual Coreference Resolution
• Abstract summary: We present our submission to the 2026 Computational Models of Reference, Anaphora and Coreference task.<n>With an average CoNLL F1 score of 74.32 on the official test set, our system ranked first in the LLM track, and third overall.
• Score: 2.708757385089282
• License: http://creativecommons.org/licenses/by-sa/4.0/
• Abstract: We present our submission to the LLM track of the 2026 Computational Models of Reference, Anaphora and Coreference (CRAC 2026) shared task. With an average CoNLL F1 score of 74.32 on the official test set, our system ranked first in the LLM track, and third overall. Our system is based on the Gemma-3-27b model, fine-tuned using a two-stage strategy with a multilingual base adapter followed by dataset-specific adapters. We represent mention spans by their headword using an XML-inspired format with local reindexing and annotate documents iteratively. These design choices proved effective across languages, document lengths, and annotation guidelines.

Related papers

• CorPipe at CRAC 2026: Empty Nodes and Cross-Lingual Transfer in Multilingual Coreference Resolution [0.6671865055671308]
  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.<n>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.
  arXiv  Detail & Related papers  (2026-05-28T16:01:03Z)
• OSGNet with MLLM Reranking @ Ego4D Episodic Memory Challenge 2026 [75.6762616797288]
  We present our champion solutions for the Natural Language Queries and GoalStep tracks of the Ego4D Episodic Memory Challenge at CVPR 2026.<n>We propose a reranking-based framework that effectively leverages the strong video-language reasoning capability of multimodal large language model (MLLM)<n>Our method achieved first place in both the Natural Language Queries and GoalStep tracks.
  arXiv  Detail & Related papers  (2026-05-20T07:14:11Z)
• PSK at SemEval-2026 Task 9: Multilingual Polarization Detection Using Ensemble Gemma Models with Synthetic Data Augmentation [0.0]
  We present our system for SemEval-2026 Task 9: Multilingual Polarization Detection, a binary classification task spanning 22 languages.<n>Our approach fine-tunes separate Gemma3 models per language using Low-Rank Adaptation (LoRA), augmented with synthetic data generated by a large language model (LLM)<n>Our final system achieves a mean macro-F1 of 0.811 across all 22 languages, ranking 2nd overall of the participating teams, with 1st place finishes in 3 languages and top-3 in 8 languages.
  arXiv  Detail & Related papers  (2026-05-06T17:29:14Z)
• The Chronicles of RiDiC: Generating Datasets with Controlled Popularity Distribution for Long-form Factuality Evaluation [56.22548620937169]
  The RiDiC dataset contains 3,000 entities from three domains -- rivers, natural disasters, and car models -- spanning different popularity tiers.<n>Each entity is accompanied by its geographical location, English and Chinese names, and relevant English and Chinese Wikipedia content.<n>These were then evaluated using a third-party factuality checker, which showed that entities from our dataset caused even frontier models to hallucinate.
  arXiv  Detail & Related papers  (2026-03-11T01:02:55Z)
• HPLT 3.0: Very Large-Scale Multilingual Resources for LLM and MT. Mono- and Bi-lingual Data, Multilingual Evaluation, and Pre-Trained Models [25.953042884928006]
  We present an initiative to provide open, very large, high-quality, and richly annotated textual datasets for almost 200 languages.<n>At 30 trillion tokens, this is likely the largest generally available multilingual collection of LLM pre-training data.<n>We train and evaluate a family of 57 monolingual encoder-decoder models, as well as a handful of monolingual GPT-like reference models.
  arXiv  Detail & Related papers  (2025-11-02T20:16:38Z)
• CorefInst: Leveraging LLMs for Multilingual Coreference Resolution [0.0]
  Coreference Resolution (CR) is a crucial yet challenging task in natural language understanding.<n>This study introduces the first multilingual CR methodology which leverages decoder-only LLMs to handle both overt and zero mentions.
  arXiv  Detail & Related papers  (2025-09-22T08:35:21Z)
• LexMatcher: Dictionary-centric Data Collection for LLM-based Machine Translation [67.24113079928668]
  We present LexMatcher, a method for data curation driven by the coverage of senses found in bilingual dictionaries. Our approach outperforms the established baselines on the WMT2022 test sets.
  arXiv  Detail & Related papers  (2024-06-03T15:30:36Z)
• TartuNLP @ SIGTYP 2024 Shared Task: Adapting XLM-RoBERTa for Ancient and Historical Languages [0.21485350418225246]
  We present our submission to the unconstrained subtask of the SIGTYP 2024 Shared Task on Word Embedding Evaluation for Ancient and Historical Languages.<n>We developed a simple, uniform, and computationally lightweight approach based on the adapters framework using parameter-efficient fine-tuning.<n>Our results show the feasibility of adapting language models pre-trained on modern languages to historical and ancient languages via adapter training.
  arXiv  Detail & Related papers  (2024-04-19T12:26:28Z)
• The Ups and Downs of Large Language Model Inference with Vocabulary Trimming by Language Heuristics [74.99898531299148]
  This research examines vocabulary trimming (VT) inspired by restricting embedding entries to the language of interest to bolster time and memory efficiency. We apply two languages to trim the full vocabulary - Unicode-based script filtering and corpus-based selection - to different language families and sizes. It is found that VT reduces the memory usage of small models by nearly 50% and has an upper bound of 25% improvement in generation speed.
  arXiv  Detail & Related papers  (2023-11-16T09:35:50Z)
• An Open Dataset and Model for Language Identification [84.15194457400253]
  We present a LID model which achieves a macro-average F1 score of 0.93 and a false positive rate of 0.033 across 201 languages. We make both the model and the dataset available to the research community.
  arXiv  Detail & Related papers  (2023-05-23T08:43:42Z)
• Zero-Shot Listwise Document Reranking with a Large Language Model [58.64141622176841]
  We propose Listwise Reranker with a Large Language Model (LRL), which achieves strong reranking effectiveness without using any task-specific training data. Experiments on three TREC web search datasets demonstrate that LRL not only outperforms zero-shot pointwise methods when reranking first-stage retrieval results, but can also act as a final-stage reranker.
  arXiv  Detail & Related papers  (2023-05-03T14:45:34Z)
• MuCGEC: a Multi-Reference Multi-Source Evaluation Dataset for Chinese Grammatical Error Correction [51.3754092853434]
  MuCGEC is a multi-reference evaluation dataset for Chinese Grammatical Error Correction (CGEC) It consists of 7,063 sentences collected from three different Chinese-as-a-Second-Language (CSL) learner sources. Each sentence has been corrected by three annotators, and their corrections are meticulously reviewed by an expert, resulting in 2.3 references per sentence.
  arXiv  Detail & Related papers  (2022-04-23T05:20:38Z)
• NEMO: Frequentist Inference Approach to Constrained Linguistic Typology Feature Prediction in SIGTYP 2020 Shared Task [83.43738174234053]
  We employ frequentist inference to represent correlations between typological features and use this representation to train simple multi-class estimators that predict individual features. Our best configuration achieved the micro-averaged accuracy score of 0.66 on 149 test languages.
  arXiv  Detail & Related papers  (2020-10-12T19:25:43Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.