Reproduce “Do Lexical and Contextual Coreference Resolution Systems Degrade Differently under Mention Noise? An Empirical Study on Scientific Software Mentions” 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
Atilla Kaan Alkan, Felix Grezes, Jennifer Lynn Bartlett, Anna Kelbert, Kelly Lockhart, Alberto Accomazzi
Researchers from the Harvard-Smithsonian Center for Astrophysics empirically studied cross-document coreference resolution for scientific software mentions, comparing unsupervised lexical fuzzy matching and contextual embedding systems. They found both approaches achieved high performance in the SOMD 2026 shared task, with their Context Aware Representations system slightly outperforming fuzzy matching, and identified distinct robustness and efficiency trade-offs depending on input noise and corpus scale.
Key insights
- Simple, unsupervised methods like Fuzzy Matching and Context Aware Representations can achieve high performance for scientific software mention coreference due to the high lexical regularity of software names.
- The robustness of coreference systems to upstream noise is type-dependent: CAR is more robust to boundary errors, while FM degrades more gracefully under mention substitution errors.
- There is a scale-dependent precision-speed trade-off, with FM being faster for small corpora and CAR exhibiting more favorable linear scaling for large corpora.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.