There’s hard data in history and large language models can mine and refine it. A new paper on censorship in Early Modern England shows how.*
Peter Grajz and Peter Murrell wanted to know the impact government political writing had on “cultural innovation” in England in the 16th-17th centuries. So they fed an online archive of 60,000 texts into multiple large language models with a historical narrative providing a context, to create an annual censorship index.
On one level, their work confirmed the obvious; censorship suppressed debate (burning people at the stake under Mary I will do that); but debate blossomed when the government eased up on Protestants with unorthodox opinions in 1689.
Digging into the data they found that censorship did not work over time that “cultural production” bounced back a decade or so after the men with the red pens cut hardest.
Obscurity alert: now they have the model, perhaps Grajz and Murrel could consider using it on the great question of economic history – how did the Brits and Dutch start market cultures at the end of the 17th century. The great Diedre McCluskey explains what happened but not why then – this LLM could track the thinking that created capitalism.
But the biggest thing about the paper is that they set out how they did it – including all the prompts.
It’s a data-revolution that transforms what the discipline can do and scary stuff for scholars used to reading personal papers for evidence. It might be why the authors are in economics, not history departments.