Data from the paper "Fixing FOLIO and MALLS: Verified Annotations and an LLM-assisted Framework to Focus Human Relabeling" (
https://arxiv.org/pdf/2606.02837)
Curated FOLIO instances (validation split) for evaluating LLM-based first-order-logic (FOL) formula curation.
Each instance pairs a natural-language sentence with the original FOL formalization (FOL_sentence_old) and a curated reference (FOL_sentence), grouped by story;
NLI labels… See the full description on the dataset page:
https://huggingface.co/datasets/DSAVlab-UNIUD/FOLIO_validation-curated.
