Aron van de Pol
Latest
I read colonial Korean print with computer vision and language models, across collections too large to page through, and I ask what these methods reveal and what they obscure.
I am a PhD candidate at the Leiden Centre for Digital Humanities and the Leiden Institute for Area Studies. Much of the material I work with survives in large digitized collections that resist conventional close viewing, and my work develops computational methods to make them tractable as historical sources.
My dissertation, Typography and advertising in colonial Korea, takes two cases from the same decades. In the first, an interpretable vision model tells four Seoul printshops apart across some 57,000 pages, not by separate typefaces but by how often each leans on the letterforms they all bought from the same Japanese foundries.
In the second, a detector finds hundreds of thousands of newspaper advertisements and a vision-language model reads them at scale, so the print trade can be followed through the ads it placed.
With Steven Denney, I read every South Korean national-history textbook since 1895 with language models, following how the language of the nation shifts each time the books are rewritten.
