Fine-tuning an NLI model on synthetic data does not consistently reduce HANS blindspot accuracy, contrary to the hypothesis that synthetic data reinforces heuristics.
Title resolution pending
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
-
Does Training on Synthetic Data Make Models Less Robust?
Fine-tuning an NLI model on synthetic data does not consistently reduce HANS blindspot accuracy, contrary to the hypothesis that synthetic data reinforces heuristics.