Reproduce “Annotation Errors and NER: A Study with OntoNotes 5.0” 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
Gabriel Bernier-Colborne, Sowmya Vajjala
The authors revise OntoNotes 5.0 annotations by using data patterns to flag suspicious labels for manual review and correction. They change close to 10% of sentences across the train, development, and test splits. Micro-F1 rises for all four configurations, with spaCy-best increasing from 89.70 to 90.97; gains vary by entity type, and the comparison does not isolate changes to training versus test labels.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.