The GIVE-project is a digitisation effort of impressive proportions: over 800.000 newspaper pages, masterpieces and glass plate photographs were digitised and sustainably preserved. With the help of artificial intelligence, over 130 collections with audio and video received more accurate descriptions, making them more searchable. Impressive results, but the project also initiated intensive cooperation between heritage organisations, experts, technical professionals and more.
In this article, we put the spotlight on the people behind the project. This time, we asked our 10 questions to Matthias Priem, Archiving Manager at meemoo.
Can you tell us about your involvement in the GIVE project and your specific role?
I lead the metadata track in the GIVE project, focusing on how we can create metadata for audiovisual content. I have a sort of umbrella role, overseeing the big picture, but I also helped to develop project requests for specific components, like facial recognition. We’ve gained experience with this technology in previous projects, so the technology gap isn’t as wide as it was before, and we’re confident we can do more with it now.
I’ve been involved in the working groups throughout the entire process. Artificial intelligence is still new for many when it comes to large-scale application, especially in the cultural heritage sector, and that’s why we bring lots of organisations together in these working groups. They include not just our partners, such as the heritage institutions, but also technicians, legal experts and ethical committees. Everyone comes together to share insights and find solutions.
Why do you think digitising heritage is so important and valuable?
There are many aspects that make it valuable. The direct link to the internet, for example, immediately makes the processed archives very accessible to the general public. Once digitised, you can make it available and instantly reach a huge audience.
What impact do you hope the GIVE project will have?
I hope the impact will be massive! The metadata track is vast, and we’ve really taken a significant step in using cutting-edge technology to create metadata. In the past, an archivist had to list and describe everything manually, but now we have bots and systems to help us. And the error margins are low because the technology is mature enough. Thanks to ethical and legal frameworks, we can then deploy this mature technology on a larger scale, and start to gradually remove some of the question marks.
The error margins are low because the technology is mature enough.
What challenges did you encounter in your work on the project, and how did you deal with them?
There were significant legal and ethical challenges, for example with facial recognition. You’re dealing with photos of people, after all. It involves individuals or even children, so everything must fit within an ethically and legally acceptable framework. You’re never going to let the machine simply identify politicians’ children, for example, so you have to configure the reference sets correctly. And then there’s still the question of where to draw the line: do you want to recognise certain individuals if they’re at a party or gathering outside their official capacity? The many heritage institutions involved consider all this extremely thoroughly.
The scale of the GIVE project is also unprecedented: we’re talking about 120,000 hours of video that we have to process through facial recognition. An algorithm looked at all these hours of video and extracted some 3.5 million faces – before another algorithm compares these faces with a reference set. It’s a huge technical challenge. Believe me: it’s not easy creating metadata for all that heritage content and processing it correctly. But thanks to the partners, we made it!
What have you learned or discovered during this project?
We learned as a sector or group of people that there’s really a lot to gain from collaboration, from the collective.
We saw this very concretely in the facial recognition aspect. Each archivist for each institution used to work with their own reference set, so each had their own, organisation-specific list of names and photos that were only relevant to them.
But in the GIVE project, by putting our heads together, we came up with a collective reference set from all 120 partners. From politics to performing arts: it’s all in there. One colleague added politicians, while another archivist included dancers or authors. The gain here is enormous because when someone with their expertise recognises a person and adds them to the list, this entity can then also be recognised across all archives – leading to new insights that you wouldn’t have discovered so quickly before.
Which collaborations within the project have made the biggest impression on you, and why?
The collaboration that resulted in a collective reference set for facial recognition was particularly impressive. Thanks to this effort, everyone can now contribute individuals from their niche to our reference set. Cycling experts know all about riders, archives have insights into political figures, and performing arts organisations are familiar with actors. With a shared set, a politician can now be spotted in a performing arts archive, or a cyclist in the Flemish Parliament. Before, these individuals might not have been annotated, which means end-users lose out.
We added Josse De Pauw to the reference sets, for example, so now end-users can find all relevant information about this actor, regardless of the archive it’s stored in. I think that’s where the real value lies.
What is your favourite masterpiece, newspaper article, glass plate or piece of metadata, and why?
My favourite piece of metadata is everything that hasn’t been named or processed yet. Beyond well-known figures like Jan Jambon and Josse De Pauw, there are many culturally or socially significant individuals not yet in the reference sets, which is why we created an interface for archivists that displays the top 100 unidentified faces from the archives. If someone appears in hundreds of hours of video, they must be significant, so we then show these ‘unknowns’ to archivists, hoping they can identify them and then name them across the entire archive. It’s uncharted territory but incredibly efficient!
