DeSePtion: Dual Sequence Prediction and Adversarial Examples for Improved Fact-Checking

The increased focus on misinformation has spurred development of data and\nsystems for detecting the veracity of a claim as well as retrieving\nauthoritative evidence. The Fact Extraction and VERification (FEVER) dataset\nprovides such a resource for evaluating end-to-end fact-checking, requiring\nretrieval of evidence from Wikipedia to validate a veracity prediction. We show\nthat current systems for FEVER are vulnerable to three categories of realistic\nchallenges for fact-checking -- multiple propositions, temporal reasoning, and\nambiguity and lexical variation -- and introduce a resource with these types of\nclaims. Then we present a system designed to be resilient to these "attacks"\nusing multiple pointer networks for document selection and jointly modeling a\nsequence of evidence sentences and veracity relation predictions. We find that\nin handling these attacks we obtain state-of-the-art results on FEVER, largely\ndue to improved evidence retrieval.\n