---
dataset_info:
  features:
  - name: sentence1
    dtype: string
  - name: sentence2
    dtype: string
  - name: labels
    dtype: string
  - name: metadata
    dtype: string
  splits:
  - name: train
    num_bytes: 1027913
    num_examples: 2939
  - name: validation
    num_bytes: 318416
    num_examples: 910
  - name: test
    num_bytes: 416493
    num_examples: 1204
  download_size: 485850
  dataset_size: 1762822
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
license: cc-by-sa-4.0
source_datasets: []
---

# perspectrum

PERSPECTRUM native claim/perspective stance with original claim partitions.

Joins the official perspective pool; uses the first original perspective in
each annotated cluster, without expanding paraphrases or inventing negatives.
Original distribution: https://github.com/CogComp/perspectrum/tree/master/data/dataset.
Dataset terms: CC BY-SA 4.0, recorded in the original data/dataset/license.txt.

Original data: [original distribution](https://github.com/CogComp/perspectrum). Repackaged as parquet for [tasksource](https://github.com/sileod/tasksource) by [scripts/repackage_dataset/stance.py](https://github.com/sileod/tasksource/blob/main/scripts/repackage_dataset/stance.py#L95).

[Conversion provenance](provenance.json): source revisions, hashes, configuration and exclusions.