Computational emotion classification for genre corpora of German tragedies and comedies from 17th to early 19th century.

This article focuses on a computational method for analyzing character emotions in German drama from the 17th to the early 19th century, using a fine-tuned German BERT language model trained on 11,939 manually annotated emotion instances across seventeen dramatic texts. The model classifies fourteen emotion categories and was applied to a corpus of 233 tragedies and comedies, revealing that tragedies feature higher proportions of 'suffering' and 'abhorrence,' while comedies show more 'anger' and 'joy,' with a notably high presence of anger in comedies. Further analysis of comedy subgenres—including satirical comedies, Enlightenment comedies (1740–1770), and popular Kasperl plays—demonstrates distinct emotional profiles, such as elevated 'friendship' and 'love' in Enlightenment comedies, increased 'anger' and 'suffering' in satirical comedies, and higher 'schadenfreude' and 'joy' in Kasperl plays. The study highlights the potential of combining manual annotation with transformer-based models to advance large-scale literary emotion analysis and suggests avenues for improving classification accuracy and exploring genre distinctions through emotion patterns.