The (In-)Consistency of Literary Concepts

Abstract

This paper explores how both annotation procedures and automatic detection (i.e. classifiers) can be used to assess the consistency of textual literary concepts. We developed an annotation tagset for the "literary comment"—a frequently used but rarely defined concept—and its subtypes (interpretative comment, attitude comment and metanarrative/metafictional comment) and trained a multi-output and a binary classifier. The multi-output classifier shows FScores of 28% for attitude comment, 36% for interpretative comment and 48% for meta comment, whereas the binary classifier achieves FScores up to 59%. Crucially, both our annotation and the automatic classification struggle with the same subtypes of comment, although annotation and classification rely on completely different features. Our findings suggest an inconsistency in the overall literary concept "comment" and most prominently the subtypes "attitude comment" and "interpretative comment". As a best-practice-example, our approach illustrates that the contribution of Digital Humanities to Literary Studies may go beyond the automatic recognition of literary phenomena.