From Sentiment Mining to Mining Embodied Emotions

In the 18 th century actors and playwrights designed a new form of theatre, the bourgeois tragedy or the “Comédie Larmoyante”, usually a play written in prose. There are indications that because of this change, the emotions were also differently expressed and represented. The traditional passions from the tragedy – such as anger, revenge, remorse, shame and despair – were changed for “smaller” emotions – such as sadness, infatuation, bliss, sensibility. Instead of speaking of the passions, people spoke of the “affections”, “sentiments” and “feelings”. The acting style is assumed to have changed as well: while the classical tragedy was predominantly a language art, in bourgeois tragedy, the emotions were best expressed through the body. During the 17th and 18th century the body parts where emotions were located changed: the production centers of emotions could migrate from the blood, to the heart or the nerves and the brain. With these migrations, the bodily expressions of emotions changed.

Would it be possible to trace the supposed change in emotional styles by means of text mining techniques? This project will explore whether this change is measurable on a text level. Firstly, the words for emotions the characters employ will be turned into an “emotional vocabulary”, giving insight in the way people thought about emotions, and in the terms and classifications they used. Secondly, the stage directions accompanying the text (“she weeps”, “collapses”, “shudders”) will be mapped to trace the bodily expression of emotions. Thirdly, the playing text also contains bodily directions, for instance when the characters describe their actions (for example “my nerves collapse”, “I cannot hold back my tears”, “my heart beats out of love”).

Through typographical tracing, performative expressions should be recognized separately from the playing texts. Subsequently, the plays should be mined on word level for emotion terms and body terms. Cluster analysis can determine what emotive acts and bodily references occur in the proximity of emotion terms. By means of machine learning and manual tagging, the digitized theater texts, broken down into recognizable units, will be researched for “emotional styles”. The aim is to develop novel visualization techniques for these styles or embodied emotions.

In the last few years, many disciplines have made a so-called “affective turn”. Historians, linguists, psychologists, anthropologists, jurists, philosophers and theologians study the production, expression and transmission of emotions and the determining role emotions play in political, social, cultural and individual processes. Notwithstanding the growing attention to the formative power of emotions, up to now few digital tools for emotion research have been developed. Sentiment analysis is a much-used method, but usually, this concerns only binary text analyses or opinion mining.

The present project aims to develop a more complex form of sentiment mining. Through a text mining analysis of 17 th and 18th century Dutch theater texts, a method will be developed to trace “emotional styles”. Emotional styles are determined both by the manner in which emotions are treated in writing (and the vocabularies employed), and the way emotions are expressed physically. What body parts are connected with which emotional expressions? How are emotions “embodied”? How do we classify emotions by connecting them with bodily sensations? The early modern period offers an excellent case to investigate the possibilities of “sentiment mining 2.0”, or “mining emotional styles”.