Natural Language Inference¶
Natural Language Inference (NLI), also called Recognizing Textual Entailment (RTE) or textual entailment, is the task of determining whether one text (the "hypothesis") logically follows from another text (the "premise"). The relationship can be entailment (logically follows), contradiction (logically inconsistent), or neutral (neither follows nor contradicts). NLI is fundamental to fact verification and question answering tasks where evidence sentences must be reasoned over to determine claim validity.
Key papers¶
- Wang & Shu (2023) — Explainable Claim Verification via Knowledge-Grounded Reasoning with Large Language Models — uses NLI with knowledge-grounding and first-order logic decomposition for claim verification; LLMs perform reasoning over knowledge-grounded answers to make entailment-style predictions
- Thorne et al. (2018) — FEVER: A Large-Scale Dataset for Fact Extraction and VERification — dataset combining evidence retrieval from Wikipedia with textual entailment classification; 185,445 claims labeled as SUPPORTED/REFUTED/NOT ENOUGH INFO with evidence sentences; baseline uses decomposable attention model for NLI component
Related topics¶
- Claim Verification — NLI is a core component of automated fact verification
- Fact-checking and corrections — NLI supports determining veracity of claims against evidence
- Information Retrieval — often combined with evidence retrieval for end-to-end fact verification systems