Reproduce “LOREN: Logic-Regularized Reasoning for Interpretable Fact Verification” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jiangjie Chen, Qiaoben Bao, Changzhi Sun, Xinbo Zhang, Jiaze Chen, Hao Zhou, Yanghua Xiao, Lei Li
LOREN (LOgic-REgularized Neural latent model) offers an interpretable fact verification system that assesses overall claim veracity while simultaneously providing transparent, phrase-level explanations. The model achieves competitive performance on the FEVER 1.0 dataset and demonstrates high faithfulness in its phrase-level explanations, identifying which specific parts of a claim are supported, refuted, or unconfirmed.
Key insights
- Claim veracity can be compositionally determined from the truthfulness of its constituent phrases, enabling fine-grained interpretability.
- Symbolic logical rules can effectively weakly supervise a neural latent variable model to learn fine-grained phrase-level veracities without requiring explicit phrase-level annotations.
- Self-supervised machine reading comprehension (MRC) can generate high-quality 'local premises' from evidence, which are crucial for assessing the truthfulness of individual claim phrases.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.