This project seeks to re-create the advertised repertoire of Bordeaux’s performance venues. In this reconstructed repertoire, we find works staged at Bordeaux’s premiere venue, the Grand Théâtre de Bordeaux, alongside performances of travelling performers. From acrobatic troupes to equestrian shows, animal menageries to pyrotechnic displays, this database offers a cross section of Bordeaux’s performance culture.

Motivation

This digital work adds context to Scott Sanders’ manuscript project, entitled In Nature’s Darkness: The Black Atlantic on French Stages. Sanders’ work in progress examines how eighteenth-century French theater responded to the increasing presence of the Black Atlantic in French Atlantic spaces. In this study, Sanders traces three stages of representational strategies: from colonial attempts to control the Black and French Atlantic, through an implicit recognition of the Black Atlantic’s presence in domestic spaces, to an overt celebration of the Black Atlantic’s role in Republican life. The Bordeaux Performance Repertoire database will help identify theatrical networks that spanned the French and possibly even Black Atlantic worlds such as the actor, director and playwright, François-Marie Mayeur de Saint-Paul, who performed in Bordeaux as well as Saint-Domingue.

Dataset

The performances in this reconstructed repertoire appear in three newspapers: Journal de Guienne (1784-1790), Journal Patriotique et de Commerce (1790-1792), and Journal de Commerce, de Politique et de Littérature (1792-1793). Combined, these periodicals contain information about theatrical events that took place over 3,000 days, from the end of Louis XVI’s reign until the early phases of the Revolution. Given the nature of the data, advertisements posted before a performance, this reconstructed repertoire does not account for cancelled events; works replaced for another; or last minute substitutions. Instead, this reconstructed repertoire gives a sense of the types of works that venues throughout Bordeaux planned to perform.

Workflow

This project follows the ethos of the minimal computing movement in so far as three individuals, with limited outside resources, produced the dataset, database and resulting website. Here we provide a narrative of the project’s genesis in the hope that it will serve as a roadmap for scholars to create their own projects.

To generate the raw data, Sanders extracted from a collection of over 3,000 newspapers theatrical information that appeared at the end of each issue, under the rubric Spectacles. He first scraped the OCR (optical character recognition) layer of the pdfs and collated them into a dataframe (i.e. spreadsheet). For each date, he removed any text that was unrelated to performance. After the raw data was narrowed down to theatrical information, the next step proved the most difficult. He tried to find an approach that would organize unstructured data. Given that OCR packages such as tesseract unevenly render pre-19th century print into legible texts, he first experimented with processes that would improve the OCR; namely, the software packages Pytesseract, Calamari, OCRopus, and Google Cloud Vision. These attempts, however, led to middling results. His next experiment involved formatting changes. Since these advertisements italicize theatrical works, he turned to Google Doc’s API to identify italicized words. While this approach worked, it led to an unwieldy amount of data manipulation. He finally decided on a Natural Language Processing approach that did not require any improvements to OCR.

Using spaCy, an advanced natural language processing library, he modified code from its api to create a custom entity detector. He randomly generated a training set that included less than 5% of the dates from the dataset. For each date in the training set, he tagged words associated with theatrical categories such as venue, work, genre as well as artists and creators. In the figure below, we see the result of spaCy’s predictive algorithm.

Sanders finally ran spaCy’s entity detector on the entire dataset and transferred the tagged entities into a dataframe.

Given a number of inconsistencies, Sanders along with an undergraduate assistant, Madeleine Wallace, combed through the dataset. They implemented both manual and computational methods to standardize spellings, correct misidentified data, and fill in any missing information. The methods deployed included a feature extraction algorithm from scikit-learn called TfidfVectorizer as well as the text correction site called OpenRefine. Finally, Sanders associated the works, venues and artists with authoritative records.

James Adams first ingested the NER-tagged data output. Using the order of entity tags, Adams set up a series of rules to parse each of the following (hierarchically highest to lowest):
Document
Venue
Performance
Work (including order within performance)
Genre
Person (actors, composers, etc.)

Adams then designed a relational database that captured the shape of the dataset and its objects. Generating unique IDs for the above objects, Adams stored the data in tables as CSV files, with the future option of creating a remote database. Finally, to allow users to interact with the dataset, Adams created RMarkdown templates for each object (Venue, Performance, Work, Genre, Person), which then generated static HTML pages (e.g. each person gets a page from the “person” template, each work gets a page from the “work” template). To see our workflow, please consult our GitHub repository.

Future Plans

Once Sanders and Adams have completed the first phase of this project, Sanders plans to include plays from earlier newspapers; namely, Affiches, Annonces, et Avis Divers (1758-1784) as well as Iris de Guienne (1763 and 1773). In addition to these periodical ressources, Sanders will incorporate performances that the actor, Jean-Baptiste Lecouvreur, recorded from 1773 to 1793 into his manuscript.

After Sanders has completed work on Bordeaux’s repertoire, he hopes to find traces of the theatrical repertoire in Nantes and possibly La Rochelle. With Bordeaux, these cities represent the three most prominent slave-trading cities in eighteenth-century France. The site’s title French Atlantic Theater is partly a nod to this future aspiration. Its name also reflects the reality of Bordeaux’s theatrical milieu, where performers and spectators migrated between Bordeaux and Saint-Domingue.

Acknowledgments

This site would not have been possible without assistance from the Bibliothèque municipale de Bordeaux, which graciously provided access to their digitized periodicals. Thanks are also due to the American Society of Eighteenth-Century Studies, which supported this site through the Theodore E. D. Braun Research Travel Fellowship.