OttomanNLP/ottoman-place-names-gazetteer

Ottoman Turkish Place Names Gazetteer (Transliteration Dataset)

Dataset Summary

This dataset serves as a specialized parallel corpus for Ottoman Turkish to Modern Turkish Latin script transliteration, focusing specifically on historical place names (toponyms). It is designed to enhance the performance of Large Language Models (LLMs) and OCR post-processing tools in recognizing and correctly transcribing historical geographical entities.

The dataset contains approximately 20,000+ unique pairs extracted from official Ottoman archival records. It is formatted for instruction tuning, making it ready for fine-tuning models like Qwen, Llama, or DeepSeek.

Dataset Structure

The dataset is provided in .jsonl format, following the conversational format (System, User, Assistant) suitable for instruction fine-tuning.

Data Instances

An example entry from the dataset:

{
  "messages": [
    {"role": "system", "content": "Sen Osmanlı Türkçesi (Matbu) metinleri Latin alfabesine transkribe eden uzman bir asistansın."},
    {"role": "user", "content": "آب كارون"},
    {"role": "assistant", "content": "Ab-ı Karun"}
  ]
}


Data Fields
system: The system prompt defining the task (Transliteration expert).

user: The source entity in Ottoman Turkish (Arabic script).

assistant: The ground truth transliteration in Modern Turkish (Latin script).

Source Data
The data is derived and processed from the "Kaynaklarıyla Osmanlı Coğrafyası Yer Adları Sözlüğü" (Dictionary of Ottoman Geography Place Names with Sources), published by the Presidency of State Archives of the Republic of Turkey (T.C. Cumhurbaşkanlığı Devlet Arşivleri Başkanlığı).

Original Source Type: Archival Gazetteer / Administrative Records

Processing: The raw data was cleaned, deduplicated based on unique script/transcription pairs, and converted into an NLP-ready format by Dr. Gökhan Usta (Istanbul Technical University).

Usage
This dataset is particularly useful for:

Fine-tuning LLMs for Ottoman Turkish OCR correction.

Named Entity Recognition (NER) tasks in historical documents.

Digital Humanities research focusing on Ottoman geography.

Loading the Dataset
Python

from datasets import load_dataset

dataset = load_dataset("ottoman-turkish-nlp-suite/ottoman-place-names-gazetteer", split="train")
print(dataset[0])
Citation
If you use this dataset in your research, please cite the original publication by the State Archives and this repository:

Kod snippet'i

@misc{usta2026ottomanplacenames,
  author = {Usta, Gökhan},
  title = {Ottoman Turkish Place Names Gazetteer for NLP},
  year = {2026},
  publisher = {Hugging Face},
  journal = {Hugging Face Hub},
  howpublished = {\url{[https://huggingface.co/datasets/ottoman-turkish-nlp-suite/ottoman-place-names-gazetteer](https://huggingface.co/datasets/ottoman-turkish-nlp-suite/ottoman-place-names-gazetteer)}}
}
Contact
For any questions or contributions, please contact Dr. Gökhan Usta or open an issue in the community tab.