ACCI combines trigger-masked counterfactual inputs with argument-only representations to debias event coreference models, reporting 88.4 CoNLL F1 on ECB+ and 85.2 on GVC, but the SOTA claim and causal interpretation are overstated.
Pairwise representa- tion learning for event coreference
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
baseline 1
citation-polarity summary
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1roles
baseline 1polarities
baseline 1representative citing papers
citing papers explorer
-
Argument-Centric Causal Intervention Method for Mitigating Bias in Cross-Document Event Coreference Resolution
ACCI combines trigger-masked counterfactual inputs with argument-only representations to debias event coreference models, reporting 88.4 CoNLL F1 on ECB+ and 85.2 on GVC, but the SOTA claim and causal interpretation are overstated.