OttomanNLP/CHURRO-Ottoman-Turkish-Subset

CHURRO Ottoman Turkish Subset

This repository contains the Ottoman Turkish historical document subset extracted from the CHURRO-DS dataset published at EMNLP 2025.

CHURRO is a 3B-parameter open-weight Large Vision-Language Model (VLM) specialized for high-accuracy, low-cost historical text recognition across diverse scripts and historical variants.

---

📊 Dataset Overview

• Total Samples: 237 historical manuscript page images with page-level transcriptions.
• Format: Native Hugging Face Image features enabled for clean side-by-side rendering in Hugging Face Dataset Viewer.
• Primary Source Subsets:
  1. Osman Agha Dataset (osman) – 236 pages:
     • Source: 18th-century Ottoman manuscript (The Memoirs of Osman Agha of Timişoara / Tımışvarlı Osman Ağa'nın Anıları), completed on May 18, 1724.
     • Archive: British Library (MS. Or. 3213).
     • Transcription: Page-aligned Arabic-script transcription curated by the authors of the CHURRO paper (Merve Tekgürler et al.).
  2. Muharaf Dataset (muharaf) – 1 page:
     • Source: Manuscripts of Handwritten Arabic dataset (Muharaf) for cursive text recognition (Oussani2018-0282a).

---

⚙️ Dataset Structure & Columns

Column Name	Data Type	Description
image	Image	High-resolution page scan (rendered directly in Hugging Face DataViewer).
original_transcription	string	Ottoman Arabic script transcription text.

---

💻 Usage Example

from datasets import load_dataset

# Load the dataset directly from Hugging Face Hub
dataset = load_dataset("OttomanNLP/CHURRO-Ottoman-Turkish-Subset")

# Inspect the first sample
sample = dataset["train"][0]
print("Transcribed Text:", sample["original_transcription"][:200])

# Access the image directly as a PIL Image object
image = sample["image"]
image.show()

---

📜 Citation

If you use this dataset in your work or research, please cite the original CHURRO EMNLP 2025 paper:

@inproceedings{semnani-etal-2025-churro,
    title = "{CHURRO}: Making History Readable with an Open-Weight Large Vision-Language Model for High-Accuracy, Low-Cost Historical Text Recognition",
    author = "Semnani, Sina  and
      Zhang, Han  and
      He, Xinyan  and
      Tekgurler, Merve  and
      Lam, Monica",
    editor = "Christodoulopoulos, Christos  and
      Chakraborty, Tanmoy  and
      Rose, Carolyn  and
      Peng, Violet",
    booktitle = "Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing",
    month = nov,
    year = "2025",
    address = "Suzhou, China",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.emnlp-main.1763/",
    doi = "10.18653/v1/2025.emnlp-main.1763",
    pages = "34777--34824",
    ISBN = "979-8-89176-332-6",
    abstract = "Accurate text recognition for historical documents can greatly advance the study and preservation of cultural heritage. Existing vision-language models (VLMs), however, are designed for modern, standardized texts and are not equipped to read the diverse languages and scripts, irregular layouts, and frequent degradation found in historical materials.This paper presents CHURRO, a 3B-parameter open-weight VLM specialized for historical text recognition. The model is trained on CHURRO-DS, the largest historical text recognition dataset to date. CHURRO-DS unifies 155 historical corpora comprising 99,491 pages, spanning 22 centuries of textual heritage across 46 language clusters, including historical variants and dead languages.We evaluate several open-weight and closed VLMs and optical character recognition (OCR) systems on CHURRO-DS and find that CHURRO outperforms all other VLMs. On the CHURRO-DS test set, CHURRO achieves 82.3{\%} (printed) and 70.1{\%} (handwritten) normalized Levenshtein similarity, surpassing the second-best model, Gemini 2.5 Pro, by 1.4{\%} and 6.5{\%}, respectively, while being 15.5 times more cost-effective.By releasing the model and dataset, we aim to enable community-driven research to improve the readability of historical texts and accelerate scholarship."
}