The paper proposes combining the Metadata Prior Dominance Score (MPDS) with an evidence-intervention statistic ΔEvi to audit whether weak-label benchmarks measure true evidence dependence rather than metadata shortcuts, illustrated with synthetic HotpotQA, SNLI, and FEVER examples.
Wrench: A comprehensive benchmark for weak supervision
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Metadata Predictability Is Not Evidence Dependence: An Intervention-Based Audit for Weak-Label Benchmarks
The paper proposes combining the Metadata Prior Dominance Score (MPDS) with an evidence-intervention statistic ΔEvi to audit whether weak-label benchmarks measure true evidence dependence rather than metadata shortcuts, illustrated with synthetic HotpotQA, SNLI, and FEVER examples.