Reproduce “Error Diversity Matters: An Error-Resistant Ensemble Method for Unsupervised Dependency Parsing” 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
Behzad Shayegh, Hobie H.-B. Lee, Xiaodan Zhu, Jackie Chi Kit Cheung, Lili Mou
Researchers at Amii, University of Alberta, Mila, and Queen's University developed an error-resistant ensemble method for unsupervised dependency parsing, explicitly leveraging "error diversity" to combine existing models effectively. Their approach, featuring a novel "society entropy" metric for ensemble selection, achieved 67.3% UAS, outperforming individual parsers and existing ensemble baselines while being significantly faster than other selection methods.
Key insights
- Effective ensembles for unsupervised dependency parsing require accounting for "error diversity" — how models err — in addition to "expertise diversity" (where models are correct).
- Naive ensemble aggregation can lead to error accumulation from weak or correlated components, thus reducing overall performance and making selection crucial.
- "Society entropy" is an efficient and effective metric that captures error diversity, providing a practical way to select robust ensemble components.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.