dsmchr/rus_xviii_printed_geography

OCR Dataset of 18th-Century Russian Printed Texts

Dataset Overview

This dataset contains line-level image–text pairs extracted from historical Russian printed materials of the 18th century. It is designed for training and evaluating Optical Character Recognition (OCR) and Handwritten Text Recognition (HTR) systems on pre-reform Russian orthography.

The dataset is intended for research in computational linguistics, digital humanities, and historical text processing.

---

Dataset Structure

Each sample consists of:

• image: a cropped line-level image from scanned historical documents
• text: the corresponding transcription in historical Russian orthography

---

Format

The dataset is provided in JSONL format:

{"image": "images/00001.png", "text": "..." }

---

Dataset Statistics

• Total samples: 3214
• Average text length: ~38.9 characters
• Minimum text length: 1 character
• Maximum text length: 88 characters

The dataset contains a mixture of full-line and fragment-level text segments due to line segmentation in historical printed sources.

---

Linguistic Characteristics

The dataset preserves features of pre-reform Russian orthography, including:

• Ѣ ѣ
• Ъ ъ
• Ї ї (Note that the capital version has been annotated in two forms: İ and І)
• Ѳ ѳ
• Ѵ ѵ
• Ѕ ѕ
• Ѯ ѯ
• Ѱ ѱ

These features make the dataset suitable for studying historical spelling variation and OCR adaptation to archaic scripts.

---

Data Analysis Notes

Preliminary analysis shows a bimodal distribution of text lengths, reflecting:

• full line text segments (approx. 30–50 characters)
• short fragments due to segmentation artifacts (<5 characters)

This structure is typical for line-segmented OCR datasets derived from historical printed sources.

---

Intended Use

This dataset can be used for:

• OCR model training
• historical document digitization
• post-OCR correction models
• linguistic analysis of historical Russian orthography
• domain adaptation experiments in NLP

---

Limitations

• Line segmentation may produce incomplete or fragmented text units
• Some samples contain very short sequences due to layout structure
• No manual normalization of historical orthography has been applied

---

License

CC BY 4.0

---

Source

The dataset is based on the following historical printed edition:

• Varenius, B. (1718). Geographia generalis: Небесный и земноводный круги купно с их свойствы и действия в трех книгах описующая. Moscow: Moscow Print Yard.

---

Contact

For questions or collaboration, please contact the dataset author.