README.md

---
license: cc-by-4.0
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
dataset_info:
  features:
  - name: MISTAKE
    dtype: string
  - name: CORRECT
    dtype: string
  - name: ERROR_TYPE
    dtype: string
  - name: Operation
    dtype: string
  - name: TagsBIO
    dtype: string
  - name: LANGUAGE
    dtype: string
  splits:
  - name: train
    num_bytes: 81855002
    num_examples: 796889
  download_size: 13850286
  dataset_size: 81855002
task_categories:
- token-classification
language:
- az
tags:
- spell
- spelling
size_categories:
- 100K<n<1M
---

# Azerbaijani Spelling Correction Dataset

This dataset contains synthetically generated spelling errors in Azerbaijani text, paired with their corrections. It's designed to help train and evaluate models for Azerbaijani spelling correction tasks.

Training code: https://github.com/LocalDoc-Azerbaijan/azerbaijani_spell_correction_transformer_model
## Dataset Overview

The dataset includes four main error types:

1. **TRANSLITERATION**: Common substitutions in informal Azerbaijani text (particularly in social media) where Latin characters with diacritics are replaced with their basic Latin equivalents:
   - 'ş' → 's'
   - 'ç' → 'c'
   - 'ğ' → 'g'
   - 'ö' → 'o'
   - 'ü' → 'u'
   - 'ı' → 'i'
   - 'ə' → 'e'

   Example: "çöldə" → "colde" or "qəşəng" → "qeseng"

2. **KEYBOARD**: Accidental keypress errors where adjacent keys on the keyboard are mistakenly pressed
   
3. **DELETION**: Errors where characters are missing from the correct word
   
4. **INSERTION**: Errors where additional characters are inserted into the correct word

Note: While TRANSLITERATION and KEYBOARD are technically similar types of character substitution errors, TRANSLITERATION represents a specific pattern of substitutions common in informal Azerbaijani writing.

## Data Format

Each line in the dataset contains the following fields:

```
MISTAKE,CORRECT,ERROR_TYPE,Operation,TagsBIO
```

Where:
- **MISTAKE**: The word containing the spelling error
- **CORRECT**: The corrected word
- **ERROR_TYPE**: One of the four error types (TRANSLITERATION, KEYBOARD, DELETION, INSERTION)
- **Operation**: Character-level edit operations represented as:
  - "C": Correct (unchanged) character
  - "S": Substitution (character is replaced)
  - "D": Deletion (character is missing)
  - "I": Insertion (extra character is present)
- **TagsBIO**: BIO tagging scheme at the character level:
  - "O": Outside (character not part of an error)
  - "B": Error span

### Example

```
abslu,absolu,DELETION,"C,C,C,D,C","O,O,O,B,O"
```

In this example:
- "abslu" is the misspelled word
- "absolu" is the correct word
- The error type is DELETION
- The operation sequence "C,C,C,D,C" indicates that:
  - The first three characters "abs" are correct
  - The fourth character "o" is deleted
  - The last character "u" is correct
- The BIO tags "O,O,O,B,O" indicate that:
  - The first three characters are outside an error span
  - The fourth position marks the beginning of an error
  - The last character is outside an error span


## Citation

If you use this dataset in your research, please cite:

```
@misc {localdoc_2025,
	author       = { {LocalDoc} },
	title        = { spell_mistake_correct_pairs_azerbaijani (Revision 33c8636) },
	year         = 2025,
	url          = { https://huggingface.co/datasets/LocalDoc/spell_mistake_correct_pairs_azerbaijani },
	doi          = { 10.57967/hf/5010 },
	publisher    = { Hugging Face }
}
```

## License

This dataset is licensed under the [Creative Commons Attribution 4.0 International License (CC-BY-4.0)](https://creativecommons.org/licenses/by/4.0/).

You are free to:
- **Share** — copy and redistribute the material in any medium or format
- **Adapt** — remix, transform, and build upon the material for any purpose, even commercially

Under the following terms:
- **Attribution** — You must give appropriate credit, provide a link to the license, and indicate if changes were made.


## Contact

For more information, questions, or issues, please contact LocalDoc at [v.resad.89@gmail.com].