Howard University-AI4PC at SemEval-2026 Task 8: Query Reformulation and Dense-Lexical Retrieval Fusion for Multi-Turn Retrieval-Augmented Generation
Year: 2026
Type: conference-paper
Abstract: We present a training-free hybrid retrievethen-rerank system for multi-turn retrievalaugmented generation, submitted to all three subtasks of SemEval-2026 Task 8 (MTRAGEval): passage retrieval (Task A), generation with reference passages (Task B), and end-to-end RAG (Task C).Our system addresses the core multi-turn challenges-nonstandalone questions, unanswerable queries, and shifting passage relevance-across four domain-specific corpora: ClapNQ, Cloud, FiQA, and Govt.Queries are reformulated through LLM-driven rewriting, decomposition into sub-queries, and Hypothetical Document Embeddings (HyDE).Retrieved candidates from dense vector search (BGE-base-en-v1.5)and BM25 lexical matching are fused via Reciprocal Rank Fusion and reranked by a crossencoder (BGE-reranker-large). Llama-3.3-70B-Instructgenerates extractive, context-grounded responses with built-in abstention for unanswerable queries.Using only open-source models without fine-tuning, the system achieves nDCG@5 of 0.4098 on Task A (22nd/38), a harmonic mean of 0.7462 on Task B (9th/26), and 0.5796 on Task C (2nd/29), coming within 1.1% of the top submission.We attribute the strong Task C result to the synergy between multi-signal query reformulation and faithful extractive generation.
Authors: Sijan Shrestha, Saurav Aryal
Institution: Howard University
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7166809938