How Google & Capgemini Preserve History at the UK's Museums

With 33 million objects across five locations, including 11 million photographs and the UK’s second-largest film archive, Imperial War Museums (IWM) has been facing a fundamental challenge: making its vast collections discoverable.

According to some figures, the Imperial War Museum's five sites (which are spread across the UK) attract more than 2.6 million visitors each year and the proper curation of exhibits and research is the cornerstone of the IWM's popularity.

Organising such a large collection is a difficult task, however, with all physical artefacts requiring proper treatment. But physical objects aren't the only thing that the IWM deals in.

"Among that collection, we also have a huge amount of what we call oral histories, which are audio interviews with people who experienced war and conflict: not just soldiers on the front line, but people at home as well, civilians," says Nick Hodder, Director of Digital Engagement and Transformation at IWM.

The accessibility challenge was immediate. “We had digitised audio that was available to everyone, so if you go online you could find stories manually by just listening to the audio,” Nick explains.

“There are a number of problems with that – one might be accessibility. Perhaps you have hearing loss – that audio file is not available to you in the same way as it is to everyone else.”

Traditional solutions fell short and historical context made the problem worse.

“We’ve been using digital transcription for a number of years, but the accuracy level of that is a lot lower than we would want,” Nick notes.

“Back in the 1940s and 50s, accents were broader than they are now, and you only really realise that when you listen to some of this audio.”

To combat this, the IWM has struck up a couple of corporate partnerships that have helped to streamline these processes.

Contextual understanding

Working with Google and Capgemini, IWM has deployed AI transcription that is able to grasp the full meaning of sentences beyond individual words.

“We have one interview with a sailor who constantly talks about ‘ooks’ and the transcription said ‘double-oos’,” Nick says.

“That’s what digital transcription normally would provide. But the AI transcription understands: this is a sailor, he’s on a ship, when he’s talking about ‘ooks’, he’s hanging things up, this must be ‘hooks’. So it understands context in a way that previous transcriptions didn’t.”

The improvements in accuracy have been dramatic.

“We found with our spot checking we’re reaching an accuracy of 99%, which is a much lower error rate than not just digital transcription, but humans as well,” Nick reports.

The efficiency gains were equally significant. Nick believes that the project saved “more than 20 years of manual transcription time in a matter of weeks”.

A revolution in research

The most significant change may be in how people investigate history. “When you search something in the past, you might say, ‘Does this person mention this date or this place?’” Nick explains.

“But through the LLM, you could say, ‘How did they feel? Were they scared? Did they have any funny stories?’ So you’re now introducing sentiment into research, which is really important, because when you are researching something where you want to tell someone’s story, it’s really important to understand how they felt.”

The implications extend beyond museums. “As we move towards the present era, away from the Second World War, we’ll find ourselves left with very few people alive with that direct experience of that war,” Nick observes.

“Capturing those oral histories is vitally important and enabling discovery of those stories is, I think, pivotal: not just for us as a museum, but I think for other museums and other archives capturing this history and making it discoverable.”