From Easter 1565, customs officials in England’s ports wrote down every ship that came and went: its name, home port, master, cargo and the merchants who paid duty. About 20,000 port books survive from the period, each in the hand of whichever clerk kept it, and the team scanned the leather-bound originals at the National Archives in Kew. On an earlier project, the Southampton team typed in about 53,000 voyages by hand.

The clever bit

The team photographed some 100,000 pages and taught AI to read them. In its 2023 project blog, the team described how it works: historians transcribe sample pages by hand as “ground truth,” and neural networks learn from them to turn pixels into text. The AI had to learn each official’s handwriting and spellings that drifted over the decades, then pull out the ship, port, master and cargo from entries that clerks didn’t all write in the same order.

The finished database and interactive map cover hundreds of thousands of voyages and more than 3,000 ports at home and abroad, over 1,000 of them in Britain. Coal appears in more than 101,000 cargo entries, the most of any commodity. The cargo lists also turn up children’s toys from China, sheep guts, bowstrings and Venice treacle, a remedy people believed could cure almost anything. In 1696–97, European markets made up about 80 percent of the recorded value of England’s overseas trade.

Without AI extracting and organising the data, this would have taken years.

— Professor Craig Lambert, historian and lead researcher, University of Southampton, via southampton.ac.uk

Hooray

Two centuries of handwritten customs paperwork is now a free map anyone can explore, right down to lost places like the town of Dunwich and Ravenspurn, a once-thriving Yorkshire port that sank into the North Sea.

What comes next

Professor Jason Sadler of Southampton’s GeoData Institute says the map lets the public, scholars and citizen historians dig in by period, trading hub, ship, goods and route. Lambert says the AI’s output still needed manual checking, and the team hasn’t published an error rate, so treat any single record as a lead back to the original port book.