---
configs:
- config_name: entailment
  data_files:
  - path:
    - entail.jsonl.zst
    split: train
  default: true
- config_name: neutral
  data_files:
  - path:
    - neutral.jsonl.zst
    split: train
- config_name: contradiction
  data_files:
  - path:
    - contra.jsonl.zst
    split: train
license: cc-by-nc-4.0
task_categories:
- text2text-generation
language:
- en
tags:
- natural-language-inference
- claim-extraction
---
# Facebook ANLI Claims Dataset

This dataset is a reformatted version of the [facebook/anli](https://huggingface.co/datasets/facebook/anli) dataset, reorganized by their original labels.

The original dataset's splits have been combined and separated into three distinct dataset configurations based on their labels: entailment, neutral, and contradiction.

For each configuration:
- **Input:** Premise text
- **Output:** A JSON list of hypotheses associated with the premise.