---
license: openrail
language:
- is
tags:
- bias-detection
- icelandic
- ner
- socially-responsible-ai
- prejudice-detection
- huggingface
- dataset
---

# Icelandic Bias-Aware NER Dataset

**Trigger warning:** This dataset contains biased, offensive, or harmful language. Examples are included solely for research purposes.

## Dataset Description

This dataset contains Icelandic sentences annotated for biased and potentially harmful expressions across 14 categories. It was developed to support research in fairness-oriented NLP, especially for low-resource languages.

### Classes
- **B-ADDICTION, I-ADDICTION**
- **B-DISABILITY, I-DISABILITY**
- **B-ORIGIN, I-ORIGIN**
- **B-GENERAL, I-GENERAL**
- **B-LGBTQIA, I-LGBTQIA**
- **B-LOOKS, I-LOOKS**
- **B-PERSONAL, I-PERSONAL**
- **B-PROFANITY, I-PROFANITY**
- **B-RELIGION, I-RELIGION**
- **B-SEXUAL, I-SEXUAL**
- **B-SOCIAL_STATUS, I-SOCIAL_STATUS**
- **B-STUPIDITY, I-STUPIDITY**
- **B-VULGAR, I-VULGAR**
- **B-WOMEN, I-WOMEN**

Annotations follow the BIO scheme (e.g., `B-WOMEN`, `I-WOMEN`, `O`).

### Contents
- **all_balanced.txt**: 184,580 sentence examples created via weak supervision (lemmatized string matching against curated bias lexicon)
-**dev.txt**: A development dataset sampled from all_balanced.txt. It includes 15,381 sentence examples.
-**test.txt**: A testing dataset sampled from all_balanced.txt. It includes 15,383 sentence examples. 
-**train.txt**: A training dataset sampled from all_balanced.txt. It includes 153,816 sentence examples
-**gold.txt**

**Automatically annotated set**: 15,383 sentences created via weak supervision (lemmatized string matching against curated bias lexicon)
- **Gold set**: 190 manually reviewed sentence examples from sources not included in the training, testing or development sets.

## Intended Uses & Limitations

### Intended Use
- Research on bias detection in Icelandic text
- Training and evaluation of bias-aware NLP models
- Educational purposes for raising awareness of bias in language

### Limitations
- Automatically annotated examples may include false positives and false negatives
- Vocabulary-based matching may miss subtle, euphemistic, or emerging forms of bias
- Gold set is relatively small

⚠ **Not intended for punitive monitoring or censorship.** Outputs are prompts for reflection, not judgments.

## Ethical Considerations

This dataset is released under the **[BigScience OpenRAIL-D License](https://www.licenses.ai/ai-licenses)**, which allows free use with responsible-use restrictions.  
Prohibited uses include:
- Harassment or discrimination
- Generating disinformation or hateful content
- Surveillance targeting individuals or groups

The dataset includes harmful language and should be handled with care. Trigger warnings are recommended for any public deployment.

## Citation

Will be added.