Digital history project

Beyond the published census.

Tamuz uses handwritten text recognition and computational methods to return to the original household schedules of the 1897 Census of the Russian Empire. This makes it possible to reconsider published census data and analyse information that disappeared from the official statistical tables.

Learn about the project
01 — About

Returning from aggregated tables to individual records.

The published volumes of the 1897 census remain an essential source for historical research. Yet publication transformed millions of individual records into aggregated categories and statistical tables.

Tamuz works with the surviving handwritten census sheets themselves. By recognising, correcting, and structuring these records, we can test published figures, reconstruct how categories were produced, and study combinations of variables that were never included in the printed census.

01

Recover

Return to household-level records rather than relying only on published aggregates.

02

Connect

Analyse age, occupation, birthplace, language, religion, kinship, and residence together.

03

Reconsider

Compare original records with published statistics and question established interpretations.

02 — The model

Training handwriting recognition for historical census forms.

The project trained and evaluated a handwritten text recognition model for Russian-language census schedules in Transkribus. The workflow combines model training with manual correction, layout-aware transcription, and Python-based data processing.

32.92%
Character error rate before specialised model training.
9.75%
Character error rate after training the project-specific model.
  • 01
    Select and prepare representative census pages.
  • 02
    Create corrected ground-truth transcriptions.
  • 03
    Train and evaluate a specialised HTR model in Transkribus.
  • 04
    Transform recognised text into structured records with Python.
  • 05
    Validate the resulting data against the original documents.
03 — Research

Questions already explored with the data.

  • 01
    Migration to Demiivka, a rapidly changing Kyiv suburb Place of birth, mobility, gendered migration, residence, and the social composition of an industrialising urban periphery. Presented at the Computational Humanities Research Conference in Luxembourg.
  • 02
    Household and family structures Household size, generational composition, co-residence, marital status, female-headed households, occupational combinations, and relationships between household members.
  • 03
    Jewish households as one focused case study A conference-specific analysis of migration and household structures, produced by filtering the wider dataset by religion rather than defining the project as a whole.
  • 04
    Published statistics and original schedules Comparison of printed categories with individual-level records, including parameters and combinations of variables unavailable in the published census.
04 — Why it matters

The original forms allow different questions to be asked.

Published census tables answer questions defined by the statistical authorities of the late Russian Empire. The household schedules allow researchers to formulate new questions and combine variables at the level of people, families, and places.

Tamuz therefore treats digitisation not simply as access work, but as a method of historical reinterpretation.

05 — Contact

Get in touch.

Contact us about research collaboration, census collections, handwritten text recognition, data methods, presentations, or supporting the project.