CCSG trains plausibility estimators on counterfactual statements made by word replacement and dropout, reporting 3.07% higher average accuracy than the previous state of the art VERA+T5.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Counterfactual Samples Constructing and Training for Commonsense Statements Estimation
CCSG trains plausibility estimators on counterfactual statements made by word replacement and dropout, reporting 3.07% higher average accuracy than the previous state of the art VERA+T5.