TL;DR

Historians at the University of Southampton have used AI to pull data from around 100,000 pages of handwritten “port books” kept by the National Archives at Kew, the university announced on 19 October. The records, which start in 1565 and run unbroken for 200 years, now feed a public interactive map of hundreds of thousands of voyages. Among the early findings: Europe, not the colonies, carried most of Britain’s overseas trade.

Teaching a machine old handwriting

The port books log individual voyages, ships, shipmasters, ports and cargoes. The difficulty was reading them. Each port official wrote in a different hand, and spellings drifted over the decades, so the team trained its software on both. Lead researcher Professor Craig Lambert said that without AI pulling out and organising the data, the job “would have taken years”. Manual checking was still needed, he added, but the process was much faster.

The university’s GeoData Institute combined the extracted records with other historical sources to build the database behind the map, which lets users explore trade by period, port, ship, route or commodity.

What the records show

The team identified more than 3,000 ports in Britain and abroad, over 1,000 of them British. In 1696-97, about 80% of England’s recorded overseas trade by value was with European markets, and Europe supplied 10 of the 15 countries with the most recorded trading locations. France led with 157 sites, the Netherlands followed with 68, and India came third with 59.

Coal was the most frequently recorded cargo, in more than 101,000 entries. Malt came second and wheat third, followed by iron, wool and wine. The 1696-97 records alone list upwards of 500,000 animal skins brought in, from wolf and otter to bear, while stranger entries include sheep guts, bowstrings, children’s toys from China and Venice treacle, an early modern cure-all.

Looking forward

This is a light story, but a useful one for anyone sitting on paper archives. In our view, it is a good example of handwriting recognition applied at scale to a large, messy collection, with people still checking the output. Anyone holding handwritten ledgers, from councils to family firms, faces a similar problem: the information exists but cannot be searched. The project also shows where AI earns its keep, by turning a years-long transcription job into one that historians can review and correct.