shtosti/CATS

---
license: cc-by-4.0
language:
- en
size_categories:
- 10K<n<100K
---

# 🐾 Taming CATS — Dataset Collection

This repository serves as a **central landing page** for the datasets used in the study:

> **Taming CATS: Controllable Automatic Text Simplification through Instruction Fine-Tuning with Control Tokens**

🔗 **Code**: https://github.com/shtosti/taming-CATS
📄 **Paper**: TODO (add link once available)

---

## Overview

We provide a **multi-domain collection of datasets** for controllable automatic text simplification (CATS), covering:

- 🏥 medical text
- 🏛️ public administration
- 📚 encyclopedic text

All datasets have been:

- cleaned and filtered
- transformed into a **unified JSON schema**
- enriched with **precomputed control attributes**, including:
- readability metrics (FKGL, ARI, Dale–Chall)
- compression ratios (character and word level)

The unified format enables consistent **training, evaluation, and cross-domain comparison**.

---

## Available Datasets

The following datasets are publicly available on Hugging Face:

### 🏥 Med-EASi (Medical Domain)
- 🔗 https://huggingface.co/datasets/shtosti/Med-EASi
- 📌 `shtosti/Med-EASi`

### 🏛️ SimPA (Public Administration Domain)
- 🔗 https://huggingface.co/datasets/shtosti/SimPA
- 📌 `shtosti/SimPA`

### 📚 WikiLarge (Encyclopedic Domain)
- 🔗 https://huggingface.co/datasets/shtosti/WikiLarge_ori_splitwise
- 📌 `shtosti/WikiLarge_ori_splitwise`

Each dataset includes:

- standardized JSONL format
- precomputed readability and compression metrics
- train / validation / test splits
- preprocessing and filtering as described in the paper

## Newsela Dataset

The **Newsela** dataset was also used in this study but **cannot be redistributed** due to licensing restrictions.

Researchers can obtain access through the official Newsela data release.

---

## Data Format

All datasets follow a unified structure, including:

- `source_text`
- `simplification_text`
- `source_metrics`
- `target_metrics`
- metadata (domain, dataset, annotation type, etc.)

This schema enables **direct use for controllable generation tasks** without additional preprocessing.

---

## License

This repository serves as an **index of datasets** and does not contain the datasets themselves.

Each dataset is distributed under its **original license**.
Please refer to the individual dataset pages for detailed licensing information.

---

## Citation

If you use these datasets, please cite:
TODO add once available