Post

An interesting report of a project at the Hertziana includes this note. I would argue that access to historical archives is actually hindered when you throw chat at something to transcribe handwritten metadata, which produces errors. But let's say you're willing to make this compromise. +

In that case, every published report of a project like this has the responsibility to publish sequel reports with updates on the progress of human review & correction of errors. Otherwise, we've just produced a digital layer that actively impedes historical research. Source: zfdg.de/2026_013#fna12

I'm skeptical because I've seen this happen too many times now. Computer vision is used in historical archives & image collections, we're told it's imperfect and creates errors; but don't worry, people will go back and correct them.

OK, but it's been six years since CV was used to add metadata to these photos of LA, and 1100+of them still have "snow" in them. 12sunsets.getty.edu/contact-shee...

A story: I once was responsible for cataloguing items that were deliberately misrepresented during WWII as being from Spain when they were actually from Germany so that those items would be preserved. I did this work 70 years after the miscataloguing. So 7 decades of inaccessibility in a sense.