The National Archives, through a partnership with FamilySearch, has used AI to extract text for all 2,322,137 pages of the Case Files of Pension and Bounty-Land Warrant Applications Based on Revolutionary War Service, ca. 1800 to ca. 1912. The records are now available in the National Archives Catalog.
FamilySearch used 30,000 pages of pensions transcribed by National Archives Citizen Archivists to teach their AI language model how to transcribe this 19th Century cursive writing. The transcriptions generated by their AI language model for the entire series are now available in an “Extracted Text” panel on the document record.
The Archives still needs review and clean-up of the AI extracted text by Citizen Archivist volunteers. They purposefully placed the AI extracted text in the Extracted Text panel because it is not considered a transcription until a human volunteer interacts with it.
The AI transcriptions are perfect. Extracted text is a tool to help transcribe a record. A page is not considered transcribed until the text has been copied to the transcription panel, reviewed against the handwritten record, edited, and saved.
You can watch a video tutorial about the new records access.
More information can be found at the National Archives Revolutionary War Pension File Transcription Mission webpage.
Illustration: Revolutionary War Pension and Bounty Land Warrant Application File for Peter Hartless of Virginia, showing AI transcription (National Archives).
Source link
