A grammar-generated balanced synthetic dataset, U-PLEAD, plus a new TARGET benchmark shows that mixing synthetic examples into training improves hate speech models' recognition of unseen target-expression combinations while preserving in-domain performance for explainable slot-filling models.
Bos, Bj \" o rn Ross, Mirella Lapata, and Francesco Barbieri
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Compositional Generalisation for Explainable Hate Speech Detection
A grammar-generated balanced synthetic dataset, U-PLEAD, plus a new TARGET benchmark shows that mixing synthetic examples into training improves hate speech models' recognition of unseen target-expression combinations while preserving in-domain performance for explainable slot-filling models.