Catching up with maps
Living with Machines uses scanned maps of Great Britain from the long nineteenth century for a suite of experiments in computer vision. These enable us to evaluate how industrialisation was documented cartographically, and what the spatial distribution of certain features of industrialised life was during this period. Maps’ continuous representation of the environment gives us clues about:
- why activities happened in specific places,
- what the experience of those activities may have been,
- and what the consequences of activities could be on the land itself in the future.
Way back in June 2019, I reported on the early work Living with Machines was beginning with historical maps. At that stage, we first asked:
What information about the lived experience of industrialisation is embedded in OS map sheets?
This question led us to set up a series of conversations, readings, and experiments to get to know just where machines and other signs of industry were to be found on nineteenth-century Ordnance Survey (and eventually other) map sheets. You can read about parts of this learning process in the posts on hack days and the Hypothesis Generation group.
One basic insight is that existing methods for generating data from maps perpetuate the ‘extractive’ rhetoric so common to industrial activity itself. This leads back to the second question we asked early on:
Can (and should) we develop new tools to make it easier to automatically extract information from historical maps for DH projects?
Here, the real test has been to develop a method for automatically collecting data from map images that:
a) has the potential to be widely reproducible for humanities researchers and GLAM institutions without data science expertise or funding for traditional GIS project data entry;
b) is flexible enough to adapt to different kinds of maps and research questions;
c) is reliable enough to be useful to historians who would otherwise simply not work at this scale;
and d) creates digital data from maps that supports a humanistic approach to distantly viewing collections, in addition to using the data forms common in geospatial sciences.
We began testing with a data set of 8,765 map sheets provided by the National Library of Scotland, which largely focused on Lancashire, Dorset, and (all of) Scotland. Since then, we have expanded our corpus by accessing images via the tile server where the NLS exposes their map sheets. This has opened the door to many thousand more sheets and allowed us to begin experiments that infer map content at a national scale.
Triangulating the maps work
With this expanded corpus, it was important to strategize how we worked with the maps and what their relationship is to the rest of LwM. We describe the resulting focus areas as:
- Work on map collections and their histories
- Work on how to use maps at scale for digital humanities research
- Work that depends on maps in conjunction with other sources
These three pieces of work create a space to combine the multidisciplinary expertise on the LwM team. We can evaluate how well our computer vision experiments meet the needs outlined above. With such an approach, our work is connected both to the institutions housing map collections and to the rest of LwM, where linking historical experience by place is a priority.
Recently Olivia Vane published an update on techniques for visualizing thousands of polygons (e.g. map sheet bounding boxes!). This is a key stepping stone in developing a method for viewing, selecting subsets of, and analyzing massive map collections.
The maps endeavor is an experiment in inventing a new method for historians to work with large map collections. It will sync up with the other LwM interventions, in particular with assessing different forms of bias in structured and unstructured text data (e.g. the census and newspapers).
Stay tuned for more posts on working with maps at scale.