Sioux football team, circa 1910.
1 of 100s of thousands of images in the @librarycongress collection with almost no associated information. Here there's a guess at a year, and the text written on the photo - listed as 'unverified data'.
1 of 100s of thousands of images in the @librarycongress collection with almost no associated information. Here there's a guess at a year, and the text written on the photo - listed as 'unverified data'.
I think a lot of these neglected objects.
There's a kind of magnifying effect at work here - the things with the least data are the least likely to be surfaced in a traditional search. The least likely to be flagged by researchers, to be added to, to have their stories told.
There's a kind of magnifying effect at work here - the things with the least data are the least likely to be surfaced in a traditional search. The least likely to be flagged by researchers, to be added to, to have their stories told.
The more data an object in the collection has, the more empowered it is– to be referenced, to be written about, to become a piece in a history as told by an author in a paper or in a book.
More simply than that, objects with more data are privileged to be found by search.
More simply than that, objects with more data are privileged to be found by search.
The best book I read last year was Lauret Savoy's Trace. In it she speaks about how so many of this country's histories have been erased. "Neglect," Savoy writes, "may reflect many things: commitment but lack of means, amnesia or apathy; perhaps forces more complex and sinister."
The first thing I tell my students when they are examining a new data set is to look at what isn't there. What fields are missing, what pieces of information the schema doesn't allow. As Savoy says, there can be many reasons for these things being absent.
Along with asking what isn't there, it's crucial to ask why. Lack of means? Apathy? Something more sinister? This is not a question limited to archives. As @mer__edith tells us in this talk, it's central to understanding the dangers of AI and ML: vimeo.com/287094149
Back to archives. The ways in which we find things online in collections do a particular disservice to objects who are already 'undata-ed'. Keyword search in particular acts to amplify the narratives that are already written, over those which have been neglected.
The Sioux Football image came to me from Flynn Shannon's excellent Free to Use extension which populates new browser windows with random images from the LOC's Flickr sets: labs.loc.gov/experiments/fr…
The Free to Use extension neatly circumvents data bias, as any image is as likely to appear as any other. It's egalitarian, for sure - but with 14,897,266 in the @librarycongress collection, a lot of images are still going to stay hidden.
This isn't a thread that's going to end tidily (I'm thinking aloud, but my work w/ @LC_Labs has been thinking about 3 things:
1) How can we make better mechanisms for finding in large collections which don't magnify existing data biases
1) How can we make better mechanisms for finding in large collections which don't magnify existing data biases
2) How can the public (ie. the Očhéthi Šakówiŋ people) be empowered to endata objects in ways that aren't obstructed by political and schematic gatekeepers (I've been particularly inspired by the TK Labels project - localcontexts.org/tk-labels/)
My time at the @librarycongress is very nearly up (only two months left!) so I'm surely not going to solve these problems in any real way. But with my upcoming ITP course and work in other archives (hint, hint, other archives) I hope I can at least mark some trails. /thread
