Exploring Text Recombination for Automatic Narrative Level Detection

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 3346–3353 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 3346 Exploring Text Recombination for Automatic Narrative Level Detection Nils Reiter♠, Judith Sieker♦, Svenja Guhr♣, Evelyn Gius♣, Sina ZarrieSS♦ ♣Technical University of Darmstadt, Residenzschloss, Marktplatz 15, 64283 Darmstadt ♠University of Cologne, Albertus-Magnus-Platz, 50931 Cologne, nils.reiter@uni-koeln.de ♦University of Bielefeld, Universitätstr. 25, 33615 Bielefeld, sina.zarriess@uni-bielefeld.de Abstract Automatizing the process of understanding the global narrative structure of long texts and stories is still a major challenge for state-of-the-art natural language understanding systems, particularly because annotated data is scarce and existing annotation workflows do not scale well to the annotation of complex narrative phenomena. In this work, we focus on the identification of narrative levels in texts corresponding to stories that can be embedded (stories within stories) or otherwise coordinated within narratives. Lacking sufficient pre-annotated training data, we explore a solution to deal with data scarcity that is common in machine learning: the automatic augmentation of an existing small data set of annotated samples with the help of data synthesis. We present a workflow for narrative level detection, that includes the operationalization of the task, a model and a data augmentation protocol for automatically generating narrative texts annotated with breaks between narrative levels. Our experiments suggest that narrative levels in long text constitute a challenging phenomenon for state-of-the-art NLP models, but generating training data synthetically does improve the prediction results considerably. Keywords: computational literary studies, narrative levels, training data induction 1. Introduction The availability of annotated resources is still a major bottleneck for natural language processing research that is concerned with the analysis of long texts, such as book-length stories or narratives. To comprehend sto- ries, one must not only be able to recognize elements at the phrase or sentence level, e.g., entities, characters, direct speech, but also global structural elements, such as plot arcs or narrative levels. The latter (to be intro- duced more thoroughly below) commonly represents stories within a story, e.g., told by a character. The analysis of stories in terms of these structures is essen- tial in research areas ranging from journalistic writing to storytelling in social media to interviews in the so- cial sciences and literary fiction. While corpora of raw texts covering these domains and types of narratives are available, annotated corpora of narratives are small and cover only some of the relevant phenomena: While the ProppLearner corpus by Finlayson (2015) contains annotations of plot elements in folktales, Bamman et al. (2020) have published a data set with annotated corefer- ence chains on 2000-word samples of stories. Both cor- pora are potentially helpful for processing long narrative texts, but there is no way to combine them in a fruitful way. Hence, automatizing the process of understanding the global narrative structure of long texts/stories is still a major challenge for state-of-the-art natural language understanding systems, particularly because annotated data is scarce. This scarcity of annotated data, however, is not easy to fix, …

Content truncated to 500 words