40 Acres and a Lie is a two year investigation that used historic research and new technologies to uncover the names of 1,250 formerly enslaved Black men and women who received land titles after the Civil War – only to have the land taken back.
This ambitious reporting project spans four stories, and is a partnership between the Center for Public Integrity, Reveal, and Mother Jones. It is a rare example of bringing historic research together with new technologies to uncover something that was previously unknown. Today, 40 Acres and a Mule is largely remembered as a broken promise and an abandoned step toward multiracial democracy. Less known is that the federal government actually issued hundreds, perhaps thousands, of titles to specific plots of land between 4 and 40 acres. Freedmen and women built homes, established local governments, and farmed the land. But their utopia didn’t last long. After President Abraham Lincoln was assassinated, his successor, Andrew Johnson, stripped property from formerly enslaved Black residents across the South and returned it to their past enslavers.
Most of the names and land titles had never been published, and the team conducted genealogical research to locate living descendants of many who had received and then lost the land. For the first time, many of these Black Americans were made aware that land had been given to and then taken away from their ancestors.
This project represents an unprecedented and innovative use of nearly 2 million Freedmen’s Bureau records that were digitized by the National Archives. But only about 500,000 had been transcribed by Smithsonian volunteers. We focused our initial research around keyword searches of transcribed documents, but we needed a way to mine the nearly 1.5 million documents that had not been transcribed. We used several recent developments in machine-learning technology to allow us to “search” both the transcribed and non-transcribed documents.
The cursive handwriting and the vocabulary used in the historical documents made it challenging to use traditional tools to employ traditional data extraction tools. We built a workflow to use the transcribed documents to “teach” a model to search the untranscribed documents, allowing us to search the entire collection in several different ways.
