In Moscow, neural networks were taught to read historical documents

In January 2023, a joint platform of the capital’s Main Archive and Yandex “Search in Archives” was launched in Moscow. In it, with the help of neural networks, archival records with complex spelling of the 18th – early 20th centuries are deciphered. Anastasia Rakova, Deputy Mayor of Moscow for Social Development, spoke about the development of the project for the introduction of computer vision technologies in the field of archiving on the International Day of Archives.

Artificial intelligence has learned to find references to surnames, first names, toponyms and any other words in the most complex historical documents in seconds, which makes searching for information about the ancestors of Muscovites easier and faster, and also helps in compiling a family tree.

The deputy mayor noted that the new service has become an effective assistant for researchers and archivists. Since its launch, users have found and studied about five million images with text transcripts – more than half of all materials presented on the platform. The service for reading ancient manuscripts is available absolutely free of charge. In it, you can search for the materials most popular with Muscovites of two centuries. Search is available in the catalog or through the search bar. In addition, there are filters by cities, funds and inventories. Next to the scan of each page, a line-by-line decoding made by neural networks is displayed. When you hover over the desired fragment, it is also highlighted on the digital copy.

Moscow, Orenburg, Novgorod regions, as well as the Republic of Mordovia have already joined the experiment on the introduction of artificial intelligence into archives. Neural networks developed in Moscow analyzed about 1.7 million electronic copies of documents from regional archives.