README.md

metadata

license: cc-by-4.0
language:
  - en
task_categories:
  - text-classification
  - other
pretty_name: CondRelBench
size_categories:
  - 10K<n<100K
tags:
  - conditional-relations
  - relation-extraction
  - natural-language-inference

CondRelBench

CondRelBench is a mixed-domain benchmark for condition-scoped relation understanding. It contains extraction data and handcrafted P1–P5 probe families from Synthetic, PubMed, and Reddit domains.

Dataset record: https://doi.org/10.6084/m9.figshare.32948507

Code and evaluation repository: sydneyanuyah/ConditionalCausality

Dataset contents

Path	Contents
data/extraction/	Gold condition-scoped extraction data
data/probes/	Canonical P1–P5 probe tables and gold labels
data/splits/extraction/	Extraction train, validation, and test splits
data/splits/probes/	Probe train, validation, and test splits

The canonical probe files contain 52,112 rows grouped into 10,449 tuple families.

Probe structure

Each condition-scoped tuple has five intended probe claims:

Probe	Transformation	Gold label
P1	Contradict the relation	Not Supported
P2	Replace the condition	Not Enough Evidence
P3	Remove the condition	Not Enough Evidence
P4	Preserve the original condition-scoped relation	Supported
P5	Contradict the condition	Not Supported

Tuple_ID identifies the probe family, and ProbeID identifies an individual claim. Labels describe whether the complete claim is licensed by the supplied paragraph.

Included documentation

• data_dictionary.md
• split_protocol.md
• probe_construction.md
• annotation_and_boundary_rules.md
• CITATION.md
• LICENSE_DATA.md

Citation

Please cite CondRelBench: A Mixed-Domain Benchmark for Condition-Scoped Relation Understanding and the dataset DOI. See CITATION.md for the current citation record.

License

The dataset is distributed under Creative Commons Attribution 4.0 International (CC BY 4.0). See LICENSE_DATA.md.