FALLACIES

This is the data for A Closer Look at the Self-Verification Abilities of Large Language Models in Logical Reasoning

• step_fallacy.test.jsonl: The FALLACIES Dataset. Each row is a sample containing
  • the reasoning step ("step"),
  • the label ("label", 0 for correct step and 1 for fallcious step),
  • the corresponding fallacy type ("fallacy"), and
  • the corresponding topic ("entity").

    {
        "step": "Since [I am wearing a necklace made of hemalyke] and [I either wear jewelry made of gold or hemalyke], therefore, [I do not wear jewelry made of gold].", 
        "entity": "hemalyke", 
        "fallacy": "Affirming a Disjunct", 
        "label": 1
    }

• fallacies.json: The names of 232 fallacies as well as their definitions.

• fallacy_taxonomy.json: The hierarchical taxonomy of fallacies.