Lit3R system retrieves and reads science papers to answer questions
Lit3R: Retrieve-Relate-Read for Evidence-Grounded Question Answering over Scientific Literature
Computation and Language
Summary
Answering questions with evidence from many scientific papers is hard because you first need to find the right papers, then figure out the important parts, and finally give a clear answer. The authors created Lit3R, a system that looks up papers using different search methods, checks if they really match the question, and then picks out evidence from those papers. It combines information from multiple papers to give a final answer along with the evidence it found. This system was tested in a competition and ranked fourth overall.
What this means in practice
- •For legal research teams: Find relevant case law and supporting evidence across many legal documents to answer complex legal questions.
- •For medical information specialists: Combine evidence from multiple clinical research papers to provide comprehensive answers to medical queries.
Authors
Akira Ise, Kotaro Kumagai, Yuta Yamaguchi, Hisanori Ozaki, Yukio Uematsu, Ikuya Yamada
Abstract
We describe tus-nlp's Lit3R (Retrieve-Relate-Read) system for LitTraceQA, a shared task for literature-grounded question answering that requires systems to retrieve relevant papers, identify supporting evidence, and generate answers. Lit3R combines off-the-shelf retrieval, reranking, and large language model (LLM) components without task-specific training. The retriever iteratively combines BM25-based sparse and dense retrieval, cross-encoder reranking, and LLM-based verification, and complements retrieval based on the question with paper-to-paper expansion. The reader first identifies supporting evidence within individual papers and then synthesizes evidence across papers to produce the final answer and evidence trace. On the official test set, our system ranked 4th on the leaderboard. Our code is available at https://github.com/tus-ist-nlp/littraceqa.