„Nuit, correspondance, sentiment”

Topic Modeling auf einem Korpus von französischen Romanen 1750-1800

Abstract

How can Romance corpora be digitally researched exploratively with regard to their literary topics? In the context of the project “Mining and Modeling Text”, topic modeling with MALLET (McCallum 2002) was applied to a corpus of 80 French novels 1750-1800 (Röttgermann et al. 2020). The aim of the topic modeling approach is to generate statements about the topics of the novels, which then are imported into a knowledge graph based on Wikibase. The overriding, interdisciplinary and novel idea is to practice data-based literary historiography. In addition to information extraction on primary texts, the knowledge network is also fed from digitized bibliographic data (Martin, Mylne & Frautschi 1977; Lüschow 2019). In the interplay of these two data types, a comparison can be carried out: Which “topics” of the novels were identified by the bibliographers and which topics are revealed by the algorithm? Two case studies on Choderlos de Laclos and Xavier de Maistre exemplify this.