by Benjamin Krautter, Janis Pagel, Nils Reiter und Marcus Willand.
November 2018
Within literary studies, there is a coexistence of different perspectives on protagonists, heroes or main characters in dramatic texts, which provide different definitions and strategies for the identification of those characters. Essentially, most of these definitions can be translated into a set of machine-readable character traits. Characters that correspond to these traits may then be classified as protagonists of the drama in question, and be distinguished from other characters (e.g. minor, secondary, supporting characters). Designing an applicable classification is the central objective of this article. Part of the problem lies in identifying eponymous characters, which is related to classifying protagonists, but involves its own presuppositions. We start by approaching both tasks from a theoretical perspective and suggest our own definition of a protagonist, which can be operationalized for the purpose of machinable classification but still draws on existing research in literary studies and follows its definitions. An attempt at manual annotation shows however that this type of definition possesses only a limited potential for intersubjectivity. Using a variety of features such as token count of characters, topic modeling and network sizes, we then train a random forest classifier that separates characters into protagonists and non-protagonists or eponymous heroes and non-eponymous heroes, respectively. The results show that protagonists and eponymous heroes are in fact reliably identifiable using simple features because of their usually prominent position within the play. Following the literary studies perspective, a conclusive analysis of the classification of specific characters using the examples of Die Verschwörung des Fiesko zu Genua, Maria Stuart, and Emilia Galotti makes clear that machine learning models offer interesting starting points for more in-depth reflections on protagonists and eponymous heroes.
[ Download (DE) ]
[ Download (ENG) ]
In addition to the pamphlet, Janis Pagel has released a detailed description of the applications in a blog post.
Cite:
