Automatically Analyzing Large Texts in aGISEnvironment: The Registrar General's Reports and Cholera in the 19th Century

Abstract

The aim of this article is to present new research showcasing how Geographic Information Systems in combination with Natural Language Processing and Corpus Linguistics methods can offer innovative venues of research to analyze large textual collections in the Humanities, particularly in historical research. Using as examples parts of the collection of the Registrar General's Reports that contain more than 200,000 pages of descriptions, census data and vital statistics for the UK, we introduce newly developed automated textual tools and well known spatial analyses used in combination to investigate a case study of the references made to cholera and other diseases in these historical sources, and their relationship to place-names during Victorian times. The integration of such techniques has allowed us to explore, in an automatic way, this historical source containing millions of words, to examine the geographies depicted in it, and to identify textual and geographic patterns in the corpus.

Geoparsing history: Locating commodities in ten million pages of nineteenth-century sources

“…A cognate project, Spatial Humanities: Texts, GIS & Places is a collaborative and interdisciplinary endeavor that also uses the Edinburgh Geoparser for historical and literary history research. Their approach, which has produced impressive results and methodological advances, has selected smaller amounts of material than Trading Consequences, largely limited to a subset of the Lancaster Newsbooks Corpus with 870,000 words and the Registrar General's reports for England and Wales with two million words (Gregory and Hardie 2011;Murrieta-Flores et al 2015). The benefit of this approach is that the researchers know a great deal about the contents of the corpus and can focus on the relatively limited geography of England and Wales or Europe.…”

“…Geoparsing makes it possible to automatically recognize the places discussed in text and explore their relationship with other significant terms. This may include identifying the places mentioned in the context of a single word, such as "cholera," to explore the spatial history of disease in a large corpus of texts, or it may involve extracting a wide range of relationships between places and a list of other words pertinent to a subfield of history (Gregory and Geddes 2014;Murrieta-Flores et al 2015). In Trading Consequences, we made use of the Edinburgh Geoparser, which we adapted to historical text Grover et al 2010).…”

A deeply annotated testbed for geographical text analysis

“…Performing this analysis enabled us to test hypotheses about the historical meaning and application of these terms, and, in the process, to create a broader basis for the study of historical landscape aesthetics. We have also used a similar approach in field of historical demography in order to evaluate the places associated with particular diseases in public health reports [19,21].…”

Dealing with heterogeneous big data when geoparsing historical corpora

“…Figure 3 exemplifies an application of this approach within our analysis of the British Library newspaper corpus. In earlier work [3], we scrutinised a corpus drawn from the Registrar Generals reports for terms and concepts associated with the water-borne diseases that were major causes of death in the nineteenth century, such as cholera and dysentery. While we were able to geoparse that (substantially smaller) corpus in full, for the British Library material, the concordance-oriented technique described here is essential due to its massive scope.…”

“…Doing this allows the researcher to extract place-name instances from a text so that it can subsequently be mapped and analysed in geographical information systems (GIS). This, in turn, opens up new potential for analysing texts in ways that stress their geographies, and new sources for use in GIS which has traditionally been restricted to the quantitative and the cartographic [2], [3]. Geoparsing is, however, an error-prone process.…”