Landcore: Coreference Resolution with Language-Specific LLM-Enhanced Prompts and XML-Inspired Annotation Scheme

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, medium

Summary

Landcore (LANguage Dependent COference REsolution) is a system submitted to the LLM Track of the CRAC 2026 Shared Task on Multilingual Coreference Resolution. It investigates the effectiveness of Large Language Models in resolving coreference across diverse languages and domains, employing a few-shot prompting methodology. The system utilizes a detailed prompt, which is further refined by an LLM to generate language-specific instructions and examples. Landcore also introduces an XML-inspired annotation scheme, designed to be more compatible with LLMs than existing formats. While not achieving the highest performance in the task, the underlying concepts and techniques demonstrated by Landcore consistently enhance performance across different experimental configurations.

Key takeaway

For NLP Engineers developing multilingual coreference resolution systems, consider integrating LLM-enhanced, language-specific prompts. Your systems can benefit from an XML-inspired annotation scheme, which improves LLM compatibility and performance across diverse settings. Experiment with few-shot prompting to adapt models efficiently to new languages and domains, even if it doesn't yield top-tier benchmark results initially, as these techniques consistently show performance improvements.

Key insights

LLM-enhanced, language-specific prompts and XML-inspired annotation improve multilingual coreference resolution.

Principles

• LLMs can generate language-specific prompts.
• XML-inspired annotation schemes suit LLMs better.
• Few-shot prompting is effective for multilingual tasks.

Method

Design a comprehensive prompt with instructions and examples, then use an LLM to enhance it for language-specific contexts. Apply an XML-inspired annotation scheme for LLM compatibility.

In practice

• Adapt prompts using LLMs for specific languages.
• Structure coreference data with XML-like formats.
• Employ few-shot learning for cross-lingual NLP.

Topics

• Coreference Resolution
• Large Language Models
• Multilingual NLP
• Prompt Engineering
• XML Annotation
• Few-shot Learning

Best for: Research Scientist, AI Scientist, NLP Engineer, Prompt Engineer

Related on AIssential

• A Larger Annotated Corpus of Portuguese Coreference — LLMs can annotate Portuguese coreference, but a new LLM-based system did not outperform existing rule-based methods.
• Supervision versus Demonstration-Based In-Context Learning for Multiword Expression Classification — Prompt sensitivity and demonstration quality critically impact LLM performance in metalinguistic classification tasks.
• CorPipe at CRAC 2026: Empty Nodes and Cross-Lingual Transfer in Multilingual Coreference Resolution — CorPipe 26 integrates empty node prediction with mention and coreference resolution for improved multilingual performance.
• Coreference Resolution in spaCy — spaCy's experimental "coref" component enables robust coreference resolution for NLP applications.
• Closing the Gap at CRAC 2026: Two-Stage Adaptation for LLM-Based Multilingual Coreference Resolution — Two-stage fine-tuning and optimized input formats enable LLMs to achieve state-of-the-art multilingual coreference resolution.
• TalkTag: Fine-Grained Morphosyntactic Error Annotation for Transcribed Speech — TalkTag automates fine-grained morphosyntactic error annotation in spoken transcripts using an LLM, even with scarce data.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.