ProofWriter — structured

A cleaned, structured build of AI2's ProofWriter for logical entailment with reasoning-graph supervision. Each row is one theory (facts + Horn-clause rules) with the questions posed against it; the zip's formal string reps are parsed into typed atoms (subject, relation, object, polarity), and every question keeps its gold answer and gold derivation as a structured proof graph — so no natural-language reverse-engineering is needed downstream.

Subsets & splits

• OWA — open-world assumption; answers are True / False / Unknown.
• CWA — closed-world assumption; answers are True / False (unprovable = false).

Each subset has train / validation / test splits. Core reasoning depths {0, 1, 2, 3, 5} are pooled within a subset (see the depth column); the theory family (AttNeg / AttNoneg / RelNeg / RelNoneg) is in the family column.

Row schema

column	type	notes
id / depth / family	str / int / str	per-theory metadata
theory_text	str	the theory in natural language
facts	list	atoms {id, text, subject, relation, object, polarity}
rules	list	{id, text, premises:[atom], conclusion:atom} (NAF ~ in CWA)
questions	list	{id, text, query, answer, qdep, qlen, strategy, proofs, proof_graph}
proof_graph	list	{tree, intermediates:[atom]} — gold derivation (qlen -1 if none)

Provenance & citation

Built with graph-reasoner (gr data proofwriter) from the official ProofWriter release (V2020.12.3), which extends RuleTaker. Please cite the upstream work; confirm the upstream data license before any downstream redistribution.

@inproceedings{tafjord2021proofwriter,
  title={ProofWriter: Generating Implications, Proofs, and Abductive Statements over
    Natural Language},
  author={Tafjord, Oyvind and Dalvi, Bhavana and Clark, Peter},
  booktitle={Findings of ACL}, year={2021}
}
@inproceedings{clark2020ruletaker,
  title={Transformers as Soft Reasoners over Language},
  author={Clark, Peter and Tafjord, Oyvind and Richardson, Kyle},
  booktitle={IJCAI}, year={2020}
}