GenAIus at RAG4Reports 2026: Citation-Aware Compression for Multilingual Report Generation

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

Summary

The GenAIus system participated in the RAG4Reports 2026 Multilingual Report Generation Task, achieving second place overall with its genaius-cluster-gpt4 run, scoring an F1 of 0.5456. This submission focused on a novel citation-aware compression strategy for the final generation stage. Instead of independently creating long and short reports, the system first generates a comprehensive long report from clustered evidence nuggets and then derives the shorter version from it. The GenAIus pipeline reuses evidence preparation stages from its earlier TREC RAGTIME system, including question generation, multilingual retrieval, nugget generation, and nugget clustering. Its baseline run, following a traditional TREC-style setup, secured third place. These results indicate that citation-aware compression is a promising approach for generating length-constrained, citation-grounded reports.

Key takeaway

For NLP Engineers developing RAG systems for multilingual report generation, consider implementing a citation-aware compression strategy. Generating a full-length report first and then deriving shorter versions from it can improve performance, as demonstrated by the GenAIus system's F1 score of 0.5456. This approach offers a better balance between nugget coverage and sentence support for length-constrained, citation-grounded outputs. Evaluate this two-stage generation against independent report generation methods.

Key insights

Citation-aware compression, generating short reports from long ones, improves multilingual report generation.

Principles

• Derive short reports from long.
• Cluster evidence nuggets first.
• Prioritize nugget coverage.

Method

Generate a long report from clustered evidence nuggets, then compress it to derive the short report, rather than generating both independently. This builds on a TREC RAGTIME pipeline for evidence preparation.

In practice

• Integrate nugget clustering.
• Test two-stage generation.
• Focus on citation grounding.

Topics

• Retrieval-Augmented Generation
• Multilingual NLP
• Report Generation
• Text Compression
• Citation Grounding
• GPT-4

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• Retrieval-Augmented Generation as a Bridge Between Classical NLP Pipelines and Modern LLM Systems — Retrieval-augmented generation (RAG) addresses the limitation of static knowledge in large language models (LLMs) by connecting them to external, up-to-date knowledge sources.
• X-MADAM-RAG: Diagnosing and Handling Chinese-English Evidence Conflict in Retrieval-Augmented Generation — Multilingual RAG systems face significant challenges with contradictory evidence, requiring specialized diagnostic tools and pipelines like X-MADAM-RAG.
• Evaluation of Chunking Strategies for Effective Text Embedding in Low-Resource Language on Agricultural Documents — Effective text chunking significantly improves RAG performance in low-resource, morphologically complex languages.
• Context Windows Aren't Enough: Why RAG Matters for High-Stakes AI — RAG is indispensable for high-stakes, citation-sensitive domains, even with large context windows, due to its accuracy and explainability.
• Uncertainty-Aware Hybrid Retrieval for Long-Document RAG — UMG-RAG improves RAG by dynamically fusing multi-granularity retrieval based on query-specific uncertainty.
• Disco-RAG: Discourse-Aware Retrieval-Augmented Generation — Explicitly modeling discourse structure and planning significantly improves RAG system performance, coherence, and factual consistency.

Open in AIssential →

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