This paper contains our system description for the second Fact Extraction and VERification (FEVER) challenge. We propose a two-staged sentence selection strategy to account for examples in the dataset where evidence is not only conditioned on the claim, but also on previously retrieved evidence. We use a publicly available document retrieval module and have fine-tuned BERT checkpoints for sentence selection and as the entailment classifier. We report a FEVER score of 68.46% on the blind test set.

In Automated Claim Verification, we retrieve evidence from a knowledge base to determine the veracity of a claim. Intuitively, the retrieval of the correct evidence plays a crucial role in this process. Often, evidence selection is tackled as a pairwise sentence classification task, i.e., we train a model to predict for each sentence individually whether it is evidence for a claim. In this work, we fine-tune document level transformers to extract all evidence from a Wikipedia document at once. We show that this approach performs better than a comparable model classifying sentences individually on all relevant evidence selection metrics in FEVER. Our complete pipeline building on this evidence selection procedure produces a new state-of-the-art result on FEVER, a popular claim verification benchmark.

Automated claim checking is the task of determining the veracity of a claim given evidence retrieved from a textual knowledge base of trustworthy facts. While previous work has taken the knowledge base as given and optimized the claim-checking pipeline, we take the opposite approach – taking the pipeline as given, we explore the choice of the knowledge base. Our first insight is that a claim-checking pipeline can be transferred to a new domain of claims with access to a knowledge base from the new domain. Second, we do not find a ”universally best” knowledge base – higher domain overlap of a task dataset and a knowledge base tends to produce better label accuracy. Third, combining multiple knowledge bases does not tend to improve performance beyond using the closest-domain knowledge base. Finally, we show that the claim-checking pipeline’s confidence score for selecting evidence can be used to assess whether a knowledge base will perform well for a new set of claims, even in the absence of ground-truth labels.

No abstract

No abstract