PFW Task 8 at SemEval-2026 Task 8: Lightweight Tri-Fusion Retrieval with Prompt-Engineered Faithful Generation for Multi-Turn RAG

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

Summary

PFW Task 8's system for SemEval 2026 Task 8 (MTRAGEval) addresses multi-turn retrieval-augmented generation across four English corpora. The system integrates BM25, SPLADE-v3, and Jina Embeddings v4 using weighted reciprocal rank fusion for retrieval, paired with zero-shot GPT-4o or GPT-4o-mini prompting for generation. It officially ranked 6th of 26 on Task B (H = 0.756), 14th of 29 on Task C (H = 0.533), and 20th of 38 on Task A (nDCG@5 = 0.433). Analysis revealed that explicit citation-format instructions increased citation use from 4% to 93% on a development sample. Furthermore, GPT-4o improved Task C H-score from 0.463 to 0.523 compared to GPT-4o-mini. The system demonstrated near-perfect UNANSWERABLE handling (H = 0.990) but struggled with UNDERSPECIFIED turns.

Key takeaway

For NLP Engineers developing multi-turn RAG systems, this work highlights the value of a tri-fusion retrieval approach combined with precise prompt engineering. You should integrate explicit citation-format instructions into your prompts to drastically improve citation use, as shown by the 4% to 93% increase. Also, carefully evaluate the performance impact of different LLMs, like GPT-4o versus GPT-4o-mini, for specific tasks.

Key insights

Tri-fusion retrieval and prompt-engineered generation enhance multi-turn RAG, showing specific prompting impacts.

Principles

• Weighted reciprocal rank fusion effectively combines diverse retrieval methods.
• Explicit citation instructions significantly improve RAG citation adherence.
• LLM choice impacts performance, with GPT-4o outperforming GPT-4o-mini.

Method

The system combines BM25, SPLADE-v3, and Jina Embeddings v4 via weighted reciprocal rank fusion, then uses zero-shot GPT-4o/GPT-4o-mini prompting for generation.

In practice

• Implement tri-fusion retrieval for robust RAG.
• Add explicit citation instructions to RAG prompts.
• Evaluate LLM variants like GPT-4o vs. GPT-4o-mini.

Topics

• Multi-Turn RAG
• Retrieval-Augmented Generation
• Prompt Engineering
• Rank Fusion
• GPT-4o
• SemEval 2026

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

Related on AIssential

• How Retrieval Systems Power Modern RAG Applications — RAG combines external retrieval with LLMs to overcome knowledge cutoffs and hallucinations, providing grounded, up-to-date responses.
• Build a Hybrid RAG System with FAISS, BM25, LangGraph and Claude Sonnet Model — Hybrid RAG combines semantic and keyword search via Reciprocal Rank Fusion to enhance retrieval accuracy for diverse query types.
• Multi-Field Hybrid Retrieval-Augmented Generation for Maritime Accident Root Cause Analysis — Field-aware hybrid RAG significantly improves maritime accident root cause analysis by structuring reports and enhancing retrieval and generation.
• Implementing Hybrid Semantic-Lexical Search in RAG — Hybrid search, combining lexical and semantic methods via Reciprocal Rank Fusion, enhances RAG system retrieval accuracy.
• 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.
• Vector Search Using Ollama for Retrieval-Augmented Generation (RAG) — RAG combines vector search with LLMs to provide context-aware, fact-grounded responses from external data.

Open in AIssential →

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