A Straightforward Approach to Narratologically Grounded Character Identification

Proceedings of the 28th International Conference on Computational Linguistics, pages 6089–6100 Barcelona, Spain (Online), December 8-13, 2020 6089 A Straightforward Approach to Narratologically Grounded Character Identification Labiba Jahan, Rahul Mittal, W. Victor H. Yarlott, Mark A. Finlayson School of Computing and Information Sciences Florida International University, Miami, FL 33199 {ljaha002, rmitt008, wyarl001, markaf}@fiu.edu Abstract One of the most fundamental elements of narrative is character: if we are to understand a nar- rative, we must be able to identify the characters of that narrative. Therefore, character iden- tification is a critical task in narrative natural language understanding. Most prior work has lacked a narratologically grounded definition of character, instead relying on simplified or im- plicit definitions that do not capture essential distinctions between characters and other referents in narratives. In prior work we proposed a preliminary definition of character that was based in clear narratological principles: a character is an animate entity that is important to the plot. Here we flesh out this concept, demonstrate that it can be reliably annotated (0.78 Cohen’s κ), and provide annotations of 170 narrative texts, drawn from 3 different corpora, containing 1,347 character co-reference chains and 21,999 non-character chains that include 3,937 animate chains. Furthermore, we have shown that a supervised classifier using a simple set of easily computable features can effectively identify these characters (overall F1 of 0.90). A detailed error analysis shows that character identification is first and foremost affected by co-reference quality, and fur- ther, that the shorter a chain is the harder it is to effectively identify as a character. We release our code and data for the benefit of other researchers1. 1 Introduction Characters are some of the most central elements of narratives, and the concept of character plays an important role in most definitions of narrative. As an example, Monika Fludernik defines a narrative as “a representation of a possible world ...at whose centre there are one or several protagonists of an anthropomorphic nature . . . who (mostly) perform goal-directed actions ...” (Fludernik, 2009, p.6; emphasis ours). This definition clearly states that characters are central to stories per se. Therefore, it is natural to assume that character identification is an important step in automatic approaches to story understanding. A number of approaches have been proposed for automatically identifying characters. Some ap- proaches, for example, have sought to solve the character identification task using domain-specific on- tologies (Declerck et al., 2012) or reasoning by reference to an existing case base (Valls-Vargas et al., 2014). Others have taken supervised machine learning approaches (Calix et al., 2013; Barros et al., 2019), where a classifier is trained over data annotated by people. Some approaches, e.g., examin- ing characters’ social networks (Sack, 2013), take character identification for granted, implementing heuristic-driven approaches over named entities or coreference chains that are not examined for their efficacy. Regardless of approach, all prior work of which we are aware has, unfortunately, had a rela- tively impoverished concept of character, at least from a narratological point of view. …

Content truncated to 500 words