Using a computational film studies framework, this article examines a major Swedish newsreel archive—the Journal Digital collection—deploying among others: signal archaeology, named entity recognition, and geocoding. We also apply a broad algorithmic toolbox to convert both intertitles and speech into a transcribed and timestamped corpus of the entire collection. Our basic idea is to construct a number of mid-sized datasets from the Journal Digital collection in different modalities, and proceed with an examination using various approaches. Consequently, our intention is to increase the scholarly capacity of media historical sources, while at the same time critically scrutinising AI and algorithmic toolboxes for the multimodal study of the past.

Media History, Newsreels, Multimodal Analysis, Digital humanities, Film studies