Study: Sliced Evaluation Gaps in NLP Can Shift Without Any Model Change
A new arXiv paper argues that sliced performance differences in NLP evaluation, such as the intra-inter gap in document-level relation extraction, can move even when models themselves are unchanged.
Read the full story at arXiv cs.CL ↗
Timeline · 1 report
- 2026-11-04 05:00 · arXiv cs.CL
Evaluation Contracts and Evaluation Leverage: Sliced-Gap Readings in Structured Prediction Can Move Without Model Change