Fact-CoT: Explicit Fact-Enhanced Chain-of-Thought Prompting for Commonsense Reasoning

We improved the Chain-of-Thought prompting by simply adding a few sets of facts for every exemplar to tackle commonsense reasoning tasks. Experiments show this self-enhancement method can notably promote LMs' reasoning abilities. We explored the mechanism of this improvement and found that explicitly enumerating facts was more effective than implicitly mentioning them in the reasoning process.