Hard to believe this is my final update of January already, but here we are!

Gender

Monday meant a slight pause for the gender investigation, as our partners at Stepstone (Total Jobs) revealed they may not currently have the expertise to share the API with us. This may mean going back to the drawing board a little, although as the original dictionary Stepstone used to create the API is public (https://ideas.wharton.upenn.edu/wp-content/uploads/2018/07/Gaucher-Friesen-Kay-2011.pdf), we can at very least search for the terms and start to get an idea of how relevant these are for the archive.

We are also still waiting on a reply from Lucy about her work, so perhaps the next stage of this investigation will be more focused on creating our own tools than using what is already out there (which is not much, and apparently quite hard to access).

Personal researcher notes

I have tried four different methods of transforming nine sets of notes this week, three of which were successful. I will be asking for opinions on these at Mondays investigation meeting, so I will not go into too much detail here to avoid repetition.

I took the opportunity to read through a file myself and decide if the description generated fits with the file contents. The file in question was one covering Wireless Messages for Mars, sent from 1928 by Dr Robinson who firmly believed he was receiving replies from a lady Martian. The generated descriptions seem to cover the content well.

While it is a totally crackers file, and I have a newfound respect for the Post Office secretaries that manged to write such polite and serious letters about the whole endeavour, it is also a nice case of linkage. The equipment used to send the message is now in the centre of Information Age at SCM, and detailed drawings of Rugby Radio Station are held by BT (and were even proposed by Jon as a case for digital reconstruction). It’s a good example of a crazy human story at the centre of a tangled web.

Computer vision

It was great to speak to Kaspar on Tuesday about where we see the work between Heritage Weaver and comms going. Ideas that were floating around, once the database is up and running, were the potential to link our object images to photographs, as well as to textual sources such as the GPO Circulars. There’s also the potential to trial Meta’s segment-anything to pull out details and multiple technologies in larger more complex images (for example, screens, aerials, and vehicles in television detector van images):

BT Archives visit

I visited James in BT on Wednesday to see the work they have been doing with moving image material, which was very interesting and hopefully will lead to a mutual learning session on the topic soon.

In short, James has been using ‘Subtitle Editor’ to automatically transcribe around 1000 moving image files, and has then fed these transcriptions to LLAMA2 to generate descriptions. The output is a large series of subtitled films for the archive to host, huge amounts of (time-stamped) transcripts, and around 1000 archival descriptions.

The time-stamped transcripts seem ripe with potential for linkage, and include a wide range of material from 30 second adverts to 3 hour long recordings of management meetings. It would be great to do something with this data in the future, any ideas from readers are most welcome.

Exhibit

Tim, Alex and myself are working up an exhibit brief for the upcoming installation at NSMM in 2025. I will start sharing more regular updates on the exhibits notion page as soon as there’s more to say, but for now the main idea is to celebrate the potential of linkage, get visitors interacting with AI, and to make all of this highly technical work make sense to my mum (okay, maybe that last part is just me).

It was good to hear at the fellows meeting that there is great interest in the work of the exhibit, and I look forward to many more discussions to come!