Literary Event Detection

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3623–3634 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 3623 Literary Event Detection Matthew Sims School of Information UC Berkeley msims@berkeley.edu Jong Ho Park Computer Science Division UC Berkeley jjhpark@berkeley.edu David Bamman School of Information UC Berkeley dbamman@berkeley.edu Abstract In this work we present a new dataset of lit- erary events—events that are depicted as tak- ing place within the imagined space of a novel. While previous work has focused on event de- tection in the domain of contemporary news, literature poses a number of complications for existing systems, including complex nar- ration, the depiction of a broad array of men- tal states, and a strong emphasis on figurative language. We outline the annotation decisions of this new dataset and compare several mod- els for predicting events; the best performing model, a bidirectional LSTM with BERT to- ken representations, achieves an F1 score of 73.9. We then apply this model to a cor- pus of novels split across two dimensions— prestige and popularity—and demonstrate that there are statistically significant differences in the distribution of events for prestige. 1 Introduction Do events determine the shape of literary nar- ratives? This question reaches back at least as far as the 1920s, when literary theorists from the Russian Formalist school began making distinc- tions between syuzhet (the way in which events are presented in a narrative) and fabula (the chronological sequence of events, distinct from the way they’re represented) (Shklovsky, 1990; Propp, 2010). Even on a far more localized scale, events are often considered to play a fundamental role in how literary narratives progress. Moretti (2013), for instance, describes the inherent pro- ductivity of events in Daniel Defoe’s novel Robin- son Crusoe, where one event invokes another in a chain of occurrences that seem to flow in “micro- narrative sequences.” Such localized sequences in turn relate to the larger architecture of plot, which has its own distinct modes of organization and generation (Forster, 1927; Genette, 1983; Brooks, 1992). The status of events in literature thus in- evitably engages larger questions about scale and narrative technique. At the same time, the representation and iden- tification of events and their participants in NLP have historically focused on the domain of news, including early evaluation campaigns like MUC (Sundheim, 1991), seminal datasets like ACE 2005 (Walker et al., 2006) and the DEFT ERE framework (Aguilar et al., 2014; Bies et al., 2016), as well as other resources that require the identifi- cation of events as a precondition for other activi- ties, such as temporal ordering (Pustejovsky et al., 2003b) or factuality judgments (Saur´ı and Puste- jovsky, 2009; de Marneffe et al., 2012; Werner et al., 2015; Lee et al., 2015; Rudinger et al., 2018). The role of events in literary fiction, however, is very different from their role in fact-based re- porting of events in the real world, including his- torical texts (Sprugnoli and Tonelli, 2017). Nov- els …

Content truncated to 500 words