Reproduce “Decontextualization: Making Sentences Stand-Alone” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Eunsol Choi, Jennimaria Palomaki, Matthew Lamm, Tom Kwiatkowski, Dipanjan Das, Michael Collins
This research defines and addresses sentence decontextualization, the task of rewriting context-dependent sentences to be self-contained. It introduces a new dataset and demonstrates that large generative models can successfully perform this task, improving user experience in question answering and enhancing retrieval efficiency for open-domain QA systems.
Key insights
- A significant portion (88%) of Wikipedia sentences can be decontextualized, with 68% requiring explicit edits to become stand-alone.
- Large language models like T5 can effectively perform decontextualization and feasibility classification, often outperforming human agreement on classification.
- While models generate fluent outputs, human annotations still exhibit greater thoroughness and accuracy in decontextualization compared to current state-of-the-art models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.