Improving the fact-checking performance of language models by relying on their entailment ability

Automated fact-checking is a crucial task in this digital age. The NLP community has been trying various strategies to build robust fact-checking systems. However, we have not been very successful yet. One main reason behind this is that fact verification is a complex process. Language models have to parse through multiple pieces of evidence, often contradicting each other, to predict a claim's veracity. In this paper, we proposed a simple yet effective strategy, where we relied on the entailment ability of language models to improve the fact-checking performance. Apart from that, we did a comparison of different prompting and fine-tuning strategies, as it is currently lacking in the literature. Some of our observations are: (i) training language models with raw evidence sentences (TBE-1) and overall claim-evidence understanding (TBE-2) resulted in an improvement up to 8.20% and 16.39% in macro-F1 for RAW-FC dataset, and (ii) training language models with entailed justifications (TBE-3) outperformed the baselines by a huge margin (up to 28.57% and 44.26% for LIAR-RAW and RAW-FC, respectively). We have shared our code repository to reproduce the results.