README.md

metadata

dataset_info:
  - config_name: christian_to_jewish
    features:
      - name: text
        dtype: string
      - name: swapped
        dtype: string
      - name: masked
        dtype: string
      - name: source
        dtype: string
      - name: target
        dtype: string
      - name: source_id
        dtype: string
      - name: target_id
        dtype: string
    splits:
      - name: train
        num_bytes: 12230768
        num_examples: 7822
      - name: val
        num_bytes: 12230768
        num_examples: 7822
      - name: test
        num_bytes: 12230768
        num_examples: 7822
    download_size: 23423057
    dataset_size: 36692304
  - config_name: christian_to_muslim
    features:
      - name: text
        dtype: string
      - name: swapped
        dtype: string
      - name: masked
        dtype: string
      - name: source
        dtype: string
      - name: target
        dtype: string
      - name: source_id
        dtype: string
      - name: target_id
        dtype: string
    splits:
      - name: train
        num_bytes: 10232762
        num_examples: 6412
      - name: val
        num_bytes: 10232762
        num_examples: 6412
      - name: test
        num_bytes: 10232762
        num_examples: 6412
    download_size: 19573651
    dataset_size: 30698286
  - config_name: jewish_to_muslim
    features:
      - name: text
        dtype: string
      - name: swapped
        dtype: string
      - name: masked
        dtype: string
      - name: source
        dtype: string
      - name: target
        dtype: string
      - name: source_id
        dtype: string
      - name: target_id
        dtype: string
    splits:
      - name: train
        num_bytes: 10317346
        num_examples: 6412
      - name: val
        num_bytes: 10317346
        num_examples: 6412
      - name: test
        num_bytes: 10317346
        num_examples: 6412
    download_size: 19654828
    dataset_size: 30952038
configs:
  - config_name: christian_to_jewish
    data_files:
      - split: train
        path: christian_to_jewish/train-*
      - split: val
        path: christian_to_jewish/val-*
      - split: test
        path: christian_to_jewish/test-*
  - config_name: christian_to_muslim
    data_files:
      - split: train
        path: christian_to_muslim/train-*
      - split: val
        path: christian_to_muslim/val-*
      - split: test
        path: christian_to_muslim/test-*
  - config_name: jewish_to_muslim
    data_files:
      - split: train
        path: jewish_to_muslim/train-*
      - split: val
        path: jewish_to_muslim/val-*
      - split: test
        path: jewish_to_muslim/test-*
license: cc-by-sa-4.0
language:
  - en

GRADIEND Religion Data

This dataset consists of templated sentences with the masked word being sensitive to religion, e.g., Jewish.

See GENTER and GRADIEND Race Data for similar datasets.

Usage

genter = load_dataset('aieng-lab/gradiend_religion_data', pair_of_religions, trust_remote_code=True, split=split)

split can be either train, val, test, or all.

pair_of_religions can be christian_to_jewish, christian_to_muslim, or jewish_to_muslim.

Dataset Details

Dataset Description

This dataset is a filtered version of Wikipedia-10 containing only sentences that contain a religion bias sensitive word of the source_id religion. We used the same bias sensitive words as defined by Maede et al. (2021) (bias attribute words). Based on the masked term (source), an associated target is derived from a corresponding bias attribute pair, matching the casing of source (e.g., JEWISH gets to CHRISTIAN and not christian).

Dataset Sources

• Repository: github.com/aieng-lab/gradiend
• Paper:
• Original Data: Wikipedia-10 (a subset of English Wikipedia)

Dataset Structure

• text: the original entry of Wikipedia-10
• masked: the masked version of text, i.e., with template masks for the name ([NAME]) and the pronoun ([PRONOUN])
• swapped: like masked but with inserted target for [MASK]
• source: the word at the position of [MASK] in masked (e.g., Jewish)
• source_id: a normalized identifier for the source (e.g., jewish). All entries of the same pair_of_religions have the same source_id.
• target: the word inserted for [MASK] in swapped
• target_id: a normalized identifier for the target. All entries of the same pair_of_religionshave the same target_id.

Dataset Creation

Curation Rationale

For the training of a religion bias GRADIEND models, a diverse dataset is required to asses model gradients relevant to bias-sensitive information.

Source Data

The dataset is derived from Wikipedia-10 by filtering it and extracting the template structure. Whe Wikipedia-10 dump is derived from English Wikipedia by Maede et al. 2021.

Limitations

Note that the splitting is performed entirely random. Thus, the same masked text might occur in other splits (in combination with other target words). The same limitation holds across different pair_of_religions.

Citation

BibTeX:

@misc{drechsel2025gradiendfeaturelearning,
      title={{GRADIEND}: Feature Learning within Neural Networks Exemplified through Biases}, 
      author={Jonathan Drechsel and Steffen Herbold},
      year={2025},
      eprint={2502.01406},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2502.01406}, 
}

Dataset Card Authors

jdrechsel