In his pioneer study on drama, Manfred Pfister strongly demands a way to identify protagonists: “At the current level of research it is not possible to grade the figures of the dramatis personae precisely according to their importance for...
moreIn his pioneer study on drama, Manfred Pfister strongly demands a way to identify protagonists: “At the
current level of research it is not possible to grade the figures of the dramatis personae precisely according to
their importance for the development of the plot [. . . M]ore subtle distinctions such as those between ‘major
figures’ [and] ‘minor figures’ [. . . ] can only be guessed intuitively and not defined operationally” (Pfister 1988,
166). In this paper, we explore such a way to identify protagonists in German plays. Casting the problem
as a classification task allows to inspect the influential factors for the decision, and at the same time offers
a clear evaluation method. Diverging definitions of “protagonist” and “hero” circulate in literary studies,
whereby different aspects of characters are mentioned as defining criteria. In certain aspects, the definitions
of “protagonist” and “hero” clearly overlap, e.g. their plot relevance. Other aspects are distinctive: According
to ancient understanding, the only hero of a play must be of good nature, but not flawless (Aristotle 1982,
7), whereas later interpretations of Aristotle allow heroes which can have mixed (Lessing, Hamburgische
Dramaturgie, 86. Stück and Martus (2011), 15) or even overly negative characteristics (Asmuth 1997, 94).
Since we want to identify those characters automatically that are the most relevant for the plot in a given
drama (Pfister 1988, 234), including negative heroes like Woyzeck and Macbeth, a value-based evaluation of
a single hero is not feasible. Instead, we opted for value-neutral protagonist-detection and define protagonist
as characters that have a central scope of action either by acting themselves or by triggering the action.
Consequently, more than one character can be a protagonist, and we make no assumptions on polarity.
There are already some publications that focus on sub-classifying literary characters on a formal basis or
automatically. Several research projects follow the character classification proposed by Propp (1958) (e.g.,
Declerck, Koleva, and Krieger (2012), Finlayson (2015)). It distinguishes seven character types in folk tales
by their plot function. Moretti (2011, 2013) describes experiments to formalize the notion of protagonists,
but makes use of network-based features only. Algee-Hewitt (2017) and Fischer et al. (2018) concentrate
on the development of protagonists throughout literary history using different network metrics for corpus
analyses of English and German plays. In contrast, Jannidis et al. (2016) try to automatically detect main
characters in German novels. For the classification, they use summaries of the novels as a gold standard.