LinkNav uses LLMs to link distant passages in research papers

Researchers from the paper's author team have introduced LinkNav, a novel system designed to enhance the reading of scientific articles by explicitly linking non-adjacent but related passages. The approach works by instructing a language model to generate potential questions that a reader might ask while reading a given passage. It then searches the rest of the document for passages that answer those questions, forming intra-document connections. A key component is an answer detection pipeline that achieves high precision, ensuring that only meaningful connections are surfaced. The system was built specifically for academic papers, where dense information is often spread across multiple sections, making it easy for readers to miss relevant cross-references.

LinkNav was evaluated on a dataset of academic papers, revealing that connected passages are on average ten segments away from each other. This distance underscores the value of the tool—readers would likely not manually connect such distant passages during normal reading. The system is described in a 10-page paper with 3 figures, accepted to the ACL 2026 Demo Track. While still an academic prototype, LinkNav demonstrates a practical use case for large language models in scholarly reading, potentially reducing the cognitive load on researchers and helping them uncover hidden relationships within single documents. The work was submitted to arXiv on June 4, 2026.