Christian Wolff & Katrin Dennerlein
Although the depiction and evocation of emotions are at the heart of dramatic production and poetological reflection since Aristotle quantitative drama analysis has neglected them so far. This is why the project aims at extracting emotions of characters and analyzing their appraisal with sentiment analysis. As emotions are highly depended on cultural and historical circumstances we chose to look at the historical phase starting from the end of Thirty Years’ War up to restoration period in the 19th century during which German drama was established. This is a story of developing various subgenres which vary with respect to their thematization and presentation of emotions, mostly termed ‘Affekte’ in that period. The project relies on a corpus of German drama which is significantly enlarged compared to the available one for that period, the Textgrid corpus. It now comprises another 25 libretti of the Hamburger Gänsemarktoper from the period between 1678 and 1730, another 20 libretti from around 1800 as well as 20 texts from the strolling players. This enables us to research the most important dramatic (sub-)genre for the development of affect expression, the libretti of musical theatre, in its influences on spoken theatre. In a first step models for the analysis of emotion in dramatic texts shall be evaluated with regard to their use for digital analysis. Retrievable linguistic as well as rhetorical means are specified and extracted for whole texts as well as for elements of dramatic texts specifically dedicated to the depiction of affects: stage directions, soliloquies, arias and endings of drama. Human annotators will specify for genre-specific subcorpora which emotions are mentioned (for example happiness, love, sorrow, envy). They will also annotate if these emotions are evaluated positively or negatively. In a next step algorithms of sentiment analysis shall be trained on this material. It is an important goal of the project to approve the method of sentiment analysis for its use on historical, especially fictional, corpora. Having done all this work we should be able to identify pathos-strategies, id est passages of texts with a high arousal of emotions and to give first hints as to which emotions are thematized with which appraisal. Another objective is the revision of the history of the comic character and its relation to the emotion of happiness.
- Christian Wolff:
- Katrin Dennerlein:
- Thomas Schmidt:
Tasks/Project Plan - Overview
- Development of emotion annotation schemes
- Annotation of plays of different epochs with emotion and source/target information
- Analysis of annotations and creation of emotion annotated corpora
- Adjusting of word embeddings to the historical and poetic language of different epochs
- Development of machine learning algorithms to predict emotions, source and target
- Exploration and evaluation of prediction results
Organizational tasks in the priority programme Computational Literary Studies
- Head of organization: Working Group Sentiment Analysis
- Head of organization: Working Group Drama
- Schmidt, T., Dennerlein, K. & Wolff, C. (2021, accepted). Towards a Corpus of Historical German Plays with Emotion Annotations. In LDK 2021 – 3rd Conference on Language, Data and Knowledge (LDK 2021). Zaragoza, Spain.
- Dennerlein, K., Schmidt, T. & Wolff, C. (2021, submitted). Using Deep Neural Networks for Emotion Analysis of 18th/19th century German Plays (Working Title). In vDHD-Sonderband "Fabrikation von Erkenntnis: Experimente in den Digital Humanities". Melusina Press.
- Keynote at the workshop "Sentiment Analysis in Literary Studies" at the University of Graz in February 2021.
- More Information about the Workshop
- Video to the Keynote: Annotating and quantifying sentiment and emotions in German plays from around 1800
- Official Press Release (in German)
- Participation at the Digital Humanities Day in Leipzig