IIMAS-RAG at SemEval-2026 Task 8: Hybrid Sparse-Dense Retrieval and Answerability-Conditioned Generation for Multi-Turn RAG

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

Summary

The IIMAS-RAG system, developed for SemEval-2026 Task 8, addresses multi-turn retrieval-augmented generation challenges. Its architecture integrates LLM-based query rewriting, a hybrid sparse-dense retrieval mechanism combining SPLADE and Voyage-3-large via Reciprocal Rank Fusion, and answerability-conditioned generation utilizing GPT-4.1. The system achieved notable results, ranking 4th among 38 teams in Subtask A (Retrieval) and 13th among 29 teams in Subtask C (Full RAG). Key findings indicate that query rewriting significantly enhances retrieval performance, while the generation component faces difficulties in scenarios with limited or partially answerable context.

Key takeaway

For NLP Engineers building multi-turn RAG systems, prioritizing query rewriting is crucial for retrieval effectiveness. Your system's ability to accurately retrieve relevant documents will significantly improve by incorporating LLM-based query rewriting, as demonstrated by IIMAS-RAG's strong performance. However, be prepared for generation challenges in low-context or partially answerable queries, and consider answerability conditioning.

Key insights

LLM-based query rewriting significantly improves retrieval in multi-turn RAG systems.

Principles

• Query rewriting is the most impactful retrieval component.
• Generation remains challenging in low-context scenarios.
• Hybrid sparse-dense retrieval enhances RAG performance.

Method

IIMAS-RAG uses LLM-based query rewriting, hybrid SPLADE/Voyage-3-large retrieval fused by Reciprocal Rank Fusion, and GPT-4.1 for answerability-conditioned generation.

In practice

• Implement LLM-based query rewriting for multi-turn RAG.
• Combine sparse (SPLADE) and dense (Voyage-3-large) retrievers.
• Condition generation on answerability for better responses.

Topics

• Retrieval-Augmented Generation
• Multi-Turn RAG
• Query Rewriting
• Hybrid Retrieval
• Sparse-Dense Retrieval
• SemEval-2026 Task 8
• GPT-4.1

Best for: Research Scientist, AI Engineer, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• Beyond Semantic Search: The Complete Guide to Advanced RAG with Milvus | Sagar Patil — Advanced RAG requires multi-stage retrieval, combining hybrid search with techniques like query rewriting and reranking for production reliability.
• An Efficient Approach for Answering Not Readily Attainable Questions for RAG-based Applications — This paper proposes an efficient approach for RAG systems to answer questions not readily found in retrieved data.
• Combining Retrieval-Augmented Text Generation with LLMs for Reading Content Recommendations — RAG significantly enhances LLM-generated personalized reading content, improving relevance and groundedness.
• Benchmarking Retrieval Strategies for Biomedical Retrieval-Augmented Generation: A Controlled Empirical Study — Cross-Encoder Reranking significantly improves biomedical RAG performance by enhancing contextual precision.
• Uncertainty-Aware Hybrid Retrieval for Long-Document RAG — UMG-RAG adaptively fuses multi-granularity, multi-expert retrieval based on query-specific uncertainty to create compact, coherent RAG contexts.
• Scaling Retrieval-Augmented Reasoning with Parallel Search and Explicit Merging — MultiSearch uses parallel multi-query retrieval and explicit merging to enhance LLM reasoning and signal-to-noise ratio.

Open in AIssential →

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