ProofWriter

这些数据集伴随着论文“ProofWriter：在自然语言上生成含义、证明和溯因陈述”。它们包含更新的 RuleTaker 风格的数据集，其中包含 50 万个问题、答案和基于自然语言规则库的证明，用于表明 Transformer 可以模拟对用语言表达的规则进行推理，包括证明生成。它包括使用封闭和开放世界语义的变体。证明包括中间结论。额外的注释提供数据来训练迭代的 ProofWriter 模型以及溯因推理来确定不确定的陈述。

These datasets accompany the paper titled "ProofWriter: Generating Implications, Proofs, and Abductive Statements over Natural Language". They include updated RuleTaker-style datasets containing 500,000 questions, answers, and proofs grounded in natural language rule bases, designed to demonstrate that Transformers can simulate reasoning over linguistically expressed rules, including proof generation. These datasets feature variants utilizing closed-world and open-world semantics, where the proofs incorporate intermediate conclusions. Additional annotations provide data for training iterative ProofWriter models and facilitating abductive reasoning to determine uncertain statements.