File size: 2,668 Bytes
9c0de9e
1a3cf78
 
 
 
 
 
 
 
 
 
9c0de9e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1a3cf78
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
---
license: cc-by-sa-4.0
language:
- en
pretty_name: BOLD (Synonym-Swapped)
tags:
- bias
- stereotypes
- fairness
- robustness
- synonym-substitution
dataset_info:
  features:
  - name: domain
    dtype: string
  - name: category
    dtype: string
  - name: name
    dtype: string
  - name: prompts
    list: string
  - name: wikipedia
    list: string
  splits:
  - name: train
    num_bytes: 5030305
    num_examples: 7201
  download_size: 2789746
  dataset_size: 5030305
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

# BOLD (Synonym-Swapped)

> **This is a modified version of [BOLD](https://github.com/amazon-science/bold) (Dhamala et al., 2021), redistributed under the original [CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/) license.**

## What changed

Each generation prompt in `prompts` was rewritten with synonym substitutions. The `wikipedia` source texts, domains, categories and names are unchanged.

Synonyms were proposed automatically by an LLM (`gpt-4o-mini`), which was asked to suggest close synonyms while skipping stopwords, proper nouns, numbers and words central to the sentence's topic. Every proposed substitution was applied (swap probability 1.0). Because the substitutions are automatic, a few may slightly shift meaning.

The dataset is intended for testing whether bias-benchmark results hold up under surface-level rewording.

## License

CC BY-SA 4.0, inherited from the original dataset; the full text is in [`LICENSE`](LICENSE). Under the ShareAlike term, adaptations of this dataset must also be distributed under CC BY-SA 4.0.

The license follows the original [BOLD repository](https://github.com/amazon-science/bold/blob/main/LICENSE.md) (CC BY-SA 4.0). BOLD's prompts are drawn from English Wikipedia, which is also CC BY-SA.

## Content warning

Like the original, this dataset contains stereotypes and offensive content about social groups. It is meant for measuring bias, not for training models to produce such content.

## Citation

If you use this dataset, please cite the original BOLD paper:

```bibtex
@inproceedings{bold_2021,
author = {Dhamala, Jwala and Sun, Tony and Kumar, Varun and Krishna, Satyapriya and Pruksachatkun, Yada and Chang, Kai-Wei and Gupta, Rahul},
title = {BOLD: Dataset and Metrics for Measuring Biases in Open-Ended Language Generation},
year = {2021},
publisher = {Association for Computing Machinery},
url = {https://doi.org/10.1145/3442188.3445924},
doi = {10.1145/3442188.3445924},
booktitle = {Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency},
pages = {862--872},
series = {FAccT '21}
}
```