Search

We present MapReader, a free, open-source software library written in Python for analyzing large map collections. MapReader allows users with little computer vision expertise to i) retrieve maps via web-servers; ii) preprocess and divide them into patches; iii) annotate patches; iv) train, fine-tune, and evaluate deep neural network models; and v) create structured data about map content. We demonstrate how MapReader enables historians to interpret a collection of ≈16K nineteenth-century maps of Britain (≈30.5M patches), foregrounding the challenge of translating visual markers into machine-readable data. We present a case study focusing on rail and buildings. We also show how the outputs from the MapReader pipeline can be linked to other, external datasets. We release ≈62K manually annotated patches used here for training and evaluating the models.

Although the Ordnance Survey has itself been the subject of historical research, scholars have not systematically used its maps as primary sources of information. This is partly for disciplinary reasons and partly for the technical reason that high-quality maps have not until recently been available digitally, geo-referenced, and in color. A final, and crucial, addition has been the creation of item-level metadata which allows map collections to become corpora which can for the first time be interrogated en masse as source material. By applying new Computer Vision methods leveraging machine learning, we outline a research pipeline for working with thousands (rather than a handful) of maps at once, which enables new forms of historical inquiry based on spatial analysis. Our ‘patchwork method’ draws on the longstanding desire to adopt an overall or ‘complete’ view of a territory, and in so doing highlights certain parallels between the situation faced by today’s users of digitized maps, and a similar inflexion point faced by their predecessors in the nineteenth century, as the project to map the nation approached a form of completion.

This paper proposes a new approach to animacy detection, the task of determining whether an entity is represented as animate in a text. In particular, this work is focused on atypical animacy and examines the scenario in which typically inanimate objects, specifically machines, are given animate attributes. To address it, we have created the first dataset for atypical animacy detection, based on nineteenth-century sentences in English, with machines represented as either animate or inanimate. Our method builds upon recent innovations in language modeling, specifically BERT contextualized word embeddings, to better capture fine-grained contextual properties of words. We present a fully unsupervised pipeline, which can be easily adapted to different contexts, and report its performance on an established animacy dataset and our newly introduced resource. We show that our method provides a substantially more accurate characterization of atypical animacy, especially when applied to highly complex forms of language use.