A key feature of the Enlightenment is the development of a discourse on commerce and economy entangled in larger discussions around politics, morality, and progress. Importantly, this debate was not formalised in the way it may be seen today. Instead, it was an emerging subject incorporating diverse and contested ideas. The objective of this case study, then, is to use various methods to identify the boundaries of these emerging economic, political, and moral disquisitions in a data-driven way, using a unified version of the metadata (e.g. publisher, publication year, format of the print product) of the English Short Title Catalogue (ESTC) and full texts of the Eighteenth Century Collections Online (ECCO). We approach the task iteratively, first making a separation between broadly defined economic documents and other eighteenth century documents by modeling the features which separate samples from two collections of historical economic texts from the wider ECCO data. Then, based on the previous step, we distinguish works similar to Hume’s Political Discourses (a text at the heart of the Scottish Enlightenment) from other branches of commercial and economic discourse, and analyse this set of works in more detail. We also experiment on how a purely unsupervised approach – a contextualized topic model using BERT encodings – groups our set of economic texts. Previous historical scholarship has taken the perspective that we ought to identify language use from large corpora of text. The aim has been to contextually understand language from the perspective of a particular group of historical actors or, as is the case with conceptual historians, detect contested and changing concepts. Our approach is different in that we closely link language use to the material and historical circumstances of the individual texts within which these uses can be found. Essentially, we combine computation and social network analysis with the study of changing concepts and word uses over time, taking individual editions rather than language abstracted from them as the basic object of study. This approach allows us to identify additional contextually relevant works by both their linguistic features, and the material and network history of their production. Jointly, the combination of iterative data-driven discourse detection and the focus on manifested editions allows us not only to extract a significant proportion of the debates forming the Scottish Enlightenment in a data-driven manner, but to link them to the social networks and commercial context in which they were produced and in which they evolved. Thus, this approach allows us to evaluate the existence, scope, and contexts of historical discourses (in this case, economic discourse) in the eighteenth century in a way which is both computationally state-of-the-art and relevant to historical practices and interests. It also demonstrates how data-driven analysis and the traditional hermeneutic approach can be combined to study meanings and their changes over time.
Read full abstract