PO-EMO: Conceptualization, Annotation, and Modeling of Aesthetic Emotions in German and English Poetry

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 1652–1663 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 1652 PO-EMO: Conceptualization, Annotation, and Modeling of Aesthetic Emotions in German and English Poetry Thomas Haider1,3, Steffen Eger2, Evgeny Kim3, Roman Klinger3, Winfried Menninghaus1 1Department of Language and Literature, Max Planck Institute for Empirical Aesthetics 2NLLG, Department of Computer Science, Technische Universitat Darmstadt 3Institut für Maschinelle Sprachverarbeitung, University of Stuttgart {thomas.haider, w.m}@ae.mpg.de, eger@aiphes.tu-darmstadt.de {roman.klinger, evgeny.kim}@ims.uni-stuttgart.de Abstract Most approaches to emotion analysis of social media, literature, news, and other domains focus exclusively on basic emotion categories as defined by Ekman or Plutchik. However, art (such as literature) enables engagement in a broader range of more complex and subtle emo- tions. These have been shown to also include mixed emotional responses. We consider emotions in poetry as they are elicited in the reader, rather than what is expressed in the text or intended by the author. Thus, we conceptualize a set of aesthetic emotions that are predictive of aesthetic appreciation in the reader, and allow the annotation of multiple labels per line to capture mixed emotions within their context. We evaluate this novel setting in an annotation experiment both with carefully trained experts and via crowdsourcing. Our annotation with experts leads to an acceptable agreement of κ = .70, resulting in a consistent dataset for future large scale analysis. Finally, we conduct first emotion classification experiments based on BERT, showing that identifying aesthetic emotions is challenging in our data, with up to .52 F1-micro on the German subset. Data and resources are available at https://github.com/tnhaider/poetry-emotion. Keywords: Emotion, Aesthetic Emotions, Literature, Poetry, Annotation, Corpora, Emotion Recognition, Multi-Label 1. Introduction Emotions are central to human experience, creativity and behavior. Models of affect and emotion, both in psychology and natural language processing, commonly operate on pre- defined categories, designated either by continuous scales of, e.g., Valence, Arousal and Dominance (Mohammad, 2016) or discrete emotion labels (which can also vary in intensity). Discrete sets of emotions often have been motivated by theories of basic emotions, as proposed by Ekman (1992)— Anger, Fear, Joy, Disgust, Surprise, Sadness—and Plutchik (1991), who added Trust and Anticipation. These categories are likely to have evolved as they motivate behavior that is directly relevant for survival. However, art reception typi- cally presupposes a situation of safety and therefore offers special opportunities to engage in a broader range of more complex and subtle emotions. These differences between real-life and art contexts have not been considered in natural language processing work so far. To emotionally move readers is considered a prime goal of literature since Latin antiquity (Johnson-Laird and Oatley, 2016; Menninghaus et al., 2019; Menninghaus et al., 2015). Deeply moved readers shed tears or get chills and goose- bumps even in lab settings (Wassiliwizky et al., 2017). In cases like these, the emotional response actually implies an aesthetic evaluation: narratives that have the capacity to move readers are evaluated as good and powerful texts for this …

Content truncated to 500 words