@inproceedings{sivanaiah-etal-2026-ragonauts,
    title = "{RAG}onauts at {S}em{E}val-2026 Task 8: {BM}25 Retrieval with Last-Turn Query Formulation for Multi-Turn {RAG} Conversations",
    author = "Sivanaiah, Rajalakshmi  and
      S, Angel Deborah  and
      C, Karthik Raja  and
      S, Rithika",
    editor = "Kochmar, Ekaterina  and
      Ghosh, Debanjan  and
      North, Kai  and
      Komachi, Mamoru",
    booktitle = "Proceedings of the 20th {I}nternational {W}orkshop on {S}emantic {E}valuation (2026)",
    month = jul,
    year = "2026",
    address = "San Diego, California, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl-workshops/2026.semeval-1.243/",
    pages = "1938--1943",
    ISBN = "979-8-89176-414-9",
    abstract = "This paper describes the submission to Task{\textasciitilde}A of SemEval-2026 Task{\textasciitilde}8: MTRAGEval, which evaluates passage retrieval for multi-turn Retrieval-Augmented Generation (RAG) conversations across multiple knowledge domains. The task requires retrieving relevant supporting passages given conversational history, where user queries often contain implicit references and incomplete contextual information. This paper proposes a lightweight and training-free retrieval framework based on BM25 ranking combined with conversational query formulation. Queries are derived from dialogue turns and retrieval is performed using domain-specific indices to preserve corpus relevance. Without neural retrievers or fine-tuning, our system achieves an nDCG@5 score of 0.2836 on the official evaluation set, ranking 33{\textbackslash}textsuperscript{\{}rd{\}} on the leaderboard. This result demonstrates that sparse lexical retrieval remains an efficient and reproducible baseline for conversational RAG systems."
}