See report_gutenberg.pdf for full anaylsis
Reading literature is often equated to travel, allowing the reader to participate in the geographic imagination of a culture. Literature has the potential to create geographies of desire, migration, colonialism, and fantasy. This project explores geographies of imagination through mentions of places in English literature. The Gutenberg Corpus provides documents written in and translated into English. This analysis studies mentions of place-names in the documents over time and investigates the relationships between places mentioned. The specific research points addressed in this report are which places are mentioned in the corpus, with what frequency they are mentioned, and how that frequency changes over time. The Vermont Advanced Computing Core (VACC) was utilized to support large-scale computing of the entire Gutenberg Corpus.
Time series plots were created from the data under different conditions such as truncation to certain time periods and selection of most-mentioned places. Examination of these plots can provide insight as to which countries were post well known at certain times.
- ideas.txt: file provides a logbook-like account of our work. It also contains
files that were generated or used during the intermediate process.
- report_YellowOctopus.pdf: scientific report outlining our findings.
- gutenberg_analysis.py: All data manipulation and plotting code
Data:
- geo_alldata.json: all data collected from the Gutenberg corpus
- timeseries.csv: time series data for country mentions by book (once per book)
- timseriestotal.csv: time series data for country mentions by book
- timeseriescities.csv: time series data for city by book (once perbook)
VACC_code: used to collect original data
- gutenberg_VACC.py: gathers place names from a list of gutenberg books
- geo_script.script: submits one job to the VACC
- submit_jobs.script: submits all jobs to the VACC
- combine_outputs: combines outputs from the VACC into one json file
images: .png files used within final report
The main source of data was the Gutenberg Corpus of online eBooks. This provided a nearly exhaustive collection of every literary work written or translated into English (about 60,000 works in all). The corpus is available online and is free for everyone. It was developed so that literature would be more accessible to people. Most of the books included have expired copyrights, so there is no need to pay to read them. Each book can be read in HTML format, and included with each book is publication information. GeoNames is a database of geographical names with over 25 million entries, andis free under creative commons. (Geonames) We used a subset of the GeoNames database consisting of all countries, and cities with a population of over 15,000. The statistical program R contains a specific library devoted to the Gutenberg Corpus, “gutenbergr”. Contained in this library is data about every book in the corpus, including information about the authors. This was used to create the file “authordata.csv”, which contains information on the name, birth date, death date, and Gutenberg identification number for every author who appears in the corpus.
The “geograpy” module in Python is a machine-learning named entity recognition application that searches through text for place names. This was run on every book from the Gutenberg corpus, generating a json file of data for each book. We merged the datasets for author birth year and place mentions or each book. We then created a table of counts for each place name per year. Counts were recorded for both the number of books mentioning a place name, and number of individual mentions of the place name. We binned the data to show counts for each decade rather than each individual year, and performed time series analysis on the data. A pipeline to the UVM Vermont Advanced Computing Core (VACC) was created, allowing these computations to be completed quickly and in parallel.
As previously mentioned, it is likely that some of the apparent patterns in the analysis are the results of noise. This in part exemplifies the difficulty of computational named entity recognition False positives may also have occurred due to occasional failures of the header
and footer stripping function, allowing copyright and digitization information with place names to be recorded. Another place these errors could have occurred is during parsing. Every book in the Gutenberg corpus includes information about the author and the date and place of copyright. It is likely that this data was not trimmed properly for some of the books, so publication cities (and countries) may have been mixed in with names of places mentioned in the books. Another source of error could have been the way the dates were recorded. As previously mentioned, “date” refers to the author’s birth date. We did not expect this to affect large-scale analysis of trends over time because the data spans such a large time period. In future work it will be important to refine the data collection process to reduce false positives. It will also be important to find a way to be more lenient in filtering country names without significantly increasing false positives. This would allow the inclusion of alternate and historical country names (e.g. Gaul and the Holy Roman Empire, or America and the Indies)
A related area of future research would include network analysis. A network could be created with place names as nodes with edges between places mentioned in the same book. This project provided a novel, though broad, analysis of the gutenberg dataset, and illuminated several interesting areas for future work in the digital humanities.
Cooper, D. and Gregory, I. N. (2011), Mapping the English Lake District: a literary GIS.
Transactions of the Institute of British Geographers, 36: 89-108.
doi:10.1111/j.1475-5661.2010.00405.x
Hughes, J. M., Foti, N. J., Krakauer, D. C., & Rockmore, D. N. (2012). Quantitative patterns
of stylistic influence in the evolution of literature. Proceedings of the National
Academy of Sciences, 109(20), 7682–7686. DOI: 1 0.1073/pnas.1115407109
Murrieta-flores, P., & Howell, N. (2017) Towards the Spatial Analysis of Vague and
Imaginary Place and Space: Evolving the Spatial Humanities through Medieval Romance, Journal of Map & Geography
Libraries, 13:1, 29-57, DOI: 1 0.1080/15420353.2017.1307302
Murrieta-Flores, P., and Martins B. (2019) The geospatial humanities: past, present and future, International Journal of Geographical Information Science, 33:12, 2424-2429, DOI: 10.1080/13658816.2019.1645336
Murrieta-Flores, P., Donaldson, C., & Gregory, I.N. (2017). GIS and Literary History:
Advancing Digital Humanities research through the Spatial Analysis of historical travel
writing and topographical literature. Digital Humanities Quarterly, 11.
Porter, C., Atkinson, P., and Gregory, I.N. (2018) Space and Time in 100 Million Words: Health and Disease in a Nineteenth-century NewspaperInternational Journal of Humanities and Arts Computing 2018 12:2, 196-216
Moretti, F. (1999). Atlas of the European Novel: 1800-1900. New York: Verso.