The American Revolution is getting a new kind of index

A name is a small piece of evidence. In a museum catalogue it may identify a maker; in a newspaper it may belong to a shopkeeper; in a pension file it may appear in someone else’s account of a war. Discovering that those fragments might belong in the same investigation is a substantial part of the historian’s work.

Revolution Crossroads, a collaboration between the Smithsonian and the Library of Congress, is trying to make those connections easier to find. The project focuses on people and communities whose lives intersected with the American Revolution, broadly across 1770–1810. It brings the Smithsonian’s object-centred collections into conversation with documentary collections held elsewhere.[1]

One example reported by the Associated Press’s Steven Sloan on 3 October involves a newspaper advertisement for a silver creamer from around 1774 and a similar creamer in the Smithsonian’s collection. The connection offers a way to investigate an object’s commercial setting. Similarity between an advertisement and an object is a lead, rather than proof that the advertised item is the surviving one.[4]

From images of pages to searchable evidence

The source pool described by the Library of Congress includes historic newspapers, records from four Smithsonian museums and military and pension material from the National Archives. That combination matters. An object catalogue, a newspaper and a later claim for a pension were created for different purposes. Their accounts may overlap without using the same descriptions or organising information in the same way.[2]

The project’s released processing documentation shows how AI starts to bridge those differences. Optical character recognition turns page images into machine-readable text. The system retains the positions of text blocks, page numbers and file references, so a passage can remain attached to the place from which it was extracted. The published data offer block, page and file views of the same material; adding those row counts would count it repeatedly.[5]

A second stage extracts names, places, organisations, dates and other entities. Its published table contains 24,035,075 occurrences. These are mentions in records, including repeated mentions, rather than a count of distinct people. Most Smithsonian material at this stage comes from catalogue text; the newspaper and pension material comes from page images.[6]

The extractor reads short target passages with neighbouring text for context. Before keeping a returned name, the processing code looks for that string in the target text and records its location. A result that cannot be found is discarded. This creates a route back from a proposed connection to the passage that prompted it.[6]

That route is valuable because the released entity data remain experimental. The team has not established extraction accuracy against a manually reviewed reference set. Text-recognition errors can carry forward, and a correctly copied name can still be misclassified. Shared spelling alone also does not establish that two records refer to the same person.[6]

A better starting point for a different question

Imagine following a craftsperson through a collection. A name attached to an object could lead to an advertisement, which could supply an address, which could place a workshop near another business. Those connections would still need checking. But they could turn a search for one object into an investigation of a neighbourhood, a trade or a network of customers. This is an illustrative research path, not a discovery claimed by the project.

The distinction is useful: digitising a collection makes its contents easier to retrieve; connecting collections can change what a researcher thinks to ask. Someone looking only for a known name begins with a candidate. Someone exploring recurring places, dates and associations may find a candidate they would never have typed into a search box.

The Smithsonian’s discussion of the project also recognises that surviving records carry the biases of the societies and institutions that produced them. A larger index cannot recover every voice that was never recorded. It can, however, help researchers investigate ordinary lives that are scattered through material catalogued under more prominent people, institutions or objects.[1]

The Library of Congress describes the collaboration as an effort to make shared collections usable across institutional boundaries. Releasing the underlying datasets gives other researchers a chance to follow different questions, rather than accepting a single finished account of the Revolution.[3]

A museum object can be catalogued once and still acquire another life when someone finally finds its neighbour in an archive.

Report an error or suggest a correction