Annotating Characters in Literary Corpora: A Scheme, the CHARLES Tool, and an Annotated Novel

Annotating Characters in Literary Corpora: A Scheme, the CHARLES Tool, and an Annotated Novel Hardik Vala1, Stefan Dimitrov1, David Jurgens2, Andrew Piper1, Derek Ruths1 1McGill University, 2Stanford University 1Montreal, Canada, 2Stanford, USA {hardik.vala, stefan.dimitrov}@mail.mcgill.ca, jurgens@stanford.edu, {andrew.piper, derek.ruths}@mcgill.ca Abstract Characters form the focus of various studies of literary works, including social network analysis, archetype induction, and plot comparison. The recent rise in the computational modelling of literary works has produced a proportional rise in the demand for character-annotated literary corpora. However, automatically identifying characters is an open problem and there is low availability of literary texts with manually labelled characters. To address the latter problem, this work presents three contributions: (1) a comprehensive scheme for manually resolving mentions to characters in texts. (2) A novel collaborative annotation tool, CHARLES (CHAracter Resolution Label-Entry System) for character annotation and similiar cross-document tagging tasks. (3) The character annotations resulting from a pilot study on the novel Pride and Prejudice, demonstrating the scheme and tool facilitate the efficient production of high-quality annotations. We expect this work to motivate the further production of annotated literary corpora to help meet the demand of the community. Keywords: Annotation tool, Collaborative annotation, Annotation scheme, Literary corpora 1. Introduction ”It begins with a character, usually, and once he stands up on his feet and begins to move, all I can do is trot along behind him with a paper and pencil trying to keep up long enough to put down what he says and does.” — William Faulkner Character is fundamental to literary analysis, forming the basis of much computational research in literary domains. They are crucial to investigating various literary aspects: Social networks embedded in narratives, where characters form the nodes and their interactions and relationships form the edges, have been useful in determining character impor- tance and role (Agarwal et al., 2012), understanding genre (Ardanuy and Sporleder, 2014), investigating the impact of setting on the social world described in novels (Jayannavar et al., 2015; Elson et al., 2010; Moretti, 2005), and even generating story text (Sack, 2012). Previous work has also attempted to induce character archetypes using the text as- sociated to character mentions in large corpora (Bamman et al., 2014b; Bamman et al., 2014a). Elsner (2012) outlines representations of narrative centered around characters and their relationships, and then explores differences in various works through these abstractions. Such works rely on the accurate identification of characters and their mentions, but state-of-the-art methods for auto- matically detecting and resolving character mentions per- form poorly (Vala et al., 2015) and there is sparse availabil- ity of manually annotated datasets. Moreover, the growing application of computational methods in researching narra- tive works has been met with a growing demand for anno- tated literary corpora, especially for character-driven work (Agarwal et al., 2010; He et al., 2013). Previous works have used gold mentions (Agarwal et al., 2013; Lee and Ye- ung, 2012), or restrict to using only highly prevalent char- acters (Ardanuy and Sporleder, 2014), for which current au- tomatic systems are …

Content truncated to 500 words