Datasets:

BGSynthMedic / README.md
svassileva's picture
Update README.md
573c5d8 verified
|
Raw
History Blame Contribute Delete
2.85 kB
---
language:
- en
- bg
license: cc-by-4.0
task_categories:
- translation
- text-generation
tags:
- medical
- clinical
- discharge-summaries
- parallel-corpus
- synthetic
- bulgarian
- english
- biomedical
pretty_name: BGSynthMedic
size_categories:
- 100<n<1K
---
# BGSynthMedic: Synthetic Parallel English–Bulgarian Discharge Summary Corpus
Parallel corpus of 450 synthetic discharge summaries in English and Bulgarian, covering 9 medical conditions across 4 specialties.
The English documents originate from [SynthMedic](https://huggingface.co/datasets/grazh/synth-medic) and were automatically translated into Bulgarian and validated. The dataset is intended for research in clinical NLP, machine translation, and cross-lingual entity linking for low-resource languages.
## Dataset Details
| Split | Documents |
|-------|----------:|
| train | 450 |
### Covered Specialties and Conditions
| Specialty | Condition |
|-----------|-----------|
| Cardiovascular | Arterial hypertension |
| Endocrinology | Hypothyroidism |
| Endocrinology | Type 1 diabetes |
| Endocrinology | Type 2 diabetes |
| Infectious diseases | Influenza |
| Neurology | Intracerebral hemorrhage |
| Neurology | Ischemic stroke |
| Traumatology | Ankle fractures |
| Traumatology | Distal humeral fractures |
50 documents per condition.
## Dataset Structure
```python
{
"document_id": "cardiovascular_arterial hypertension_0",
"discharge_summary_en": "...", # English discharge summary
"discharge_summary_bg": "...", # Bulgarian translation
}
```
### Fields
- **document_id** (`string`): Unique identifier encoding specialty, condition, and document index (e.g. `neurology_ischemic stroke_12`).
- **discharge_summary_en** (`string`): Synthetic English discharge summary generated by GPT-4.
- **discharge_summary_bg** (`string`): Bulgarian translation produced by BGGpt.
## Usage
```python
from datasets import load_dataset
ds = load_dataset("username/bgsynthmedic")
print(ds["train"][0])
```
## Source Data
English texts are derived from [SynthMedic](https://huggingface.co/datasets/grazh/synth-medic), a dataset of GPT-4-generated synthetic discharge summaries.
Bulgarian translations were produced using BGGpt and reviewed for quality.
## Citation
If you use this dataset, please cite:
```bibtex
@inproceedings{boytcheva2026bgsynthmedic,
title = {{BGSynthMedic}: Synthetic parallel {English}-{Bulgarian} discharge summary corpus with automatic annotations},
author = {Boytcheva, Svetla and Vassileva, Svetlana and Boytchev, Pavel},
booktitle = {Proceedings of the International Conference on Computational Linguistics in Bulgaria (CLIB 2026)},
year = {2026}
}
```
## License
[Creative Commons Attribution 4.0 International (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0/)