
The author proposes SFAVEL, an unsupervised pretraining framework leveraging language model distillation to achieve high-quality claim-fact alignments without annotations. Results show state-of-the-art performance on fact verification datasets.


coremsg

unsupervised-pretraining-for-fact-verification-by-language-model-distillation-a-novel-approach


Unsupervised Pretraining for Fact Verification by Language Model Distillation: A Novel Approach


title_rewrite


Unsupervised pretraining framework SFAVEL achieves state-of-the-art results in fact verification by distilling language model features.


unsupervised-pretraining-for-fact-verification-by-language-model-distillation-at-iclr-2024


Unsupervised Pretraining for Fact Verification by Language Model Distillation at ICLR 2024
