It cost 2.5 million euros and five years of work by a team of fifty professionals and more than five hundred volunteers, but this Monday the time has come: the database goetgevonden.nl is online. From now on, anyone can search the 680,000 resolutions (decisions) that the States General took between 1576 and 1796 – the period of the revolt against Spain, the subsequent Golden Age and the eventual demise of the Republic of the Seven United Netherlands after the invasion. by the armies of the French Revolution.
“The rapid development of technology has made it possible for us to do things that were unthinkable five years ago, when we started this adventure,” says project leader Joris Oddens of the Huygens Institute for Dutch History and Culture. “By making this database available, the Netherlands is leading the world.”
During the time of the Republic, the States General formed the highest administrative council of what was called the Generality, the central level of government. The individual provinces sent representatives to The Hague to make decisions that affected the entire country. Oddens: “The ambition to make all resolutions of the States General available already existed at the beginning of the last century, but it was only when we received a grant from research funder NWO in 2019 that we could really get started.”
The name of the database is taken from the phrase ‘is wellfound and understood’which flowed daily from the pen of the clerk of the States General when he noted down a decision of the meeting. All these resolutions are bound in more than six hundred thick books that are stored in the warehouse of the National Archives in The Hague. The first step of the project was to scan all the pages from these volumes. “There were about a million,” says Oddens. “After this the difficult work began.”
2 percent reading errors
His team was dealing with two types of resolutions, Oddens says. “The decisions from the eighteenth century were printed, but the clerk wrote down those from the seventeenth century by hand. We initially thought that the computer would be able to read those handwritten resolutions with a margin of error of 20 percent. Then it’s actually of no use to you. But ultimately we are now below 2 percent reading errors for these resolutions, and even less than 1 percent for printed decisions.”
That success is partly due to the work of the volunteers, says Oddens. “They corrected the first transcriptions that the computer had made by hand. The program learned from that, which allowed it to do its work better.”
After reading and transcribing there was a huge knitting of letters available. Oddens: “If you want to be able to use it, you will have to divide the text into the parts you are interested in: the individual resolutions. We did this by working with volunteers to look for sentences that were characteristic of the beginning of such a noted decision. We used this to train algorithms and have succeeded in distinguishing around 680,000 different resolutions.”
After this, the material could be enriched, says Oddens. “We recognized all kinds of text elements in the transcripts of the resolutions. We did this with a technique that already worked with modern texts, but not yet with historical material.”
Developments have also been very rapid in this area, according to Oddens. There is one now large language model – an AI-driven program such as the well-known ChatGPT – that can work with historical Dutch. “We asked the volunteers to include the names of places, people and organizations in a small part of the material tag. The program then went to work finding those same words throughout the corpus. It has never been done this way anywhere in the world.”
/s3/static.nrc.nl/images/gn4/stripped/data125287415-2c135d.jpg|https://images.nrc.nl/rH4_wVR0LQRN3Jyrby0urFE2r5U=/1920x/filters:no_upscale()/s3/static.nrc.nl/images/gn4/stripped/data125287415-2c135d.jpg|https://images.nrc.nl/p-Ai-FkVGtmH4BmNW7rTfOSHCB0=/5760x/filters:no_upscale()/s3/static.nrc.nl/images/gn4/stripped/data125287415-2c135d.jpg)
Generated image, Huygens Institute/Stefan Klut (Laypa/Loghi)
Artificial intelligence
The big problem here was that there was no consistent, general spelling during the more than two centuries in which the decisions were made, says Oddens. “There were sometimes dozens of ways to write a place name. But as a researcher you naturally want to find all references to ‘Amsterdam’ when you type that word on the search screen. Normally, linking all these spellings together is done manually, but we have managed to do that automatically.”
This combination of the manual work of volunteers and the most modern artificial intelligence has now created an enormous database that is accessible to everyone via the internet. Oddens: “For historians who are interested in long-term developments, it is now possible to do research that was previously unthinkable, because it was difficult to read all those volumes page by page.”
But goetgevonden.nl may also be a treasure trove for other interested users, says Oddens. “Think of people who research the history of their family or place of residence. They can now see at the touch of a button whether anything has been said about this in the States General.”