The Project Dialogism Novel Corpus: A Dataset for Quotation Attribution in Literary Texts

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 5838–5848 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 5838 The Project Dialogism Novel Corpus: A Dataset for Quotation Attribution in Literary Texts Krishnapriya Vishnubhotla∗†, Adam Hammond‡, Graeme Hirst∗ ∗Department of Computer Science, University of Toronto ‡Department of English, University of Toronto †Vector Institute for Artificial Intelligence vkpriya@cs.toronto.edu, adam.hammond@utoronto.ca, gh@cs.toronto.edu Abstract We present the Project Dialogism Novel Corpus, or PDNC, an annotated dataset of quotations for English literary texts. PDNC contains annotations for 35,978 quotations across 22 full-length novels, and is by an order of magnitude the largest corpus of its kind. Each quotation is annotated for the speaker, addressees, type of quotation, referring expression, and character mentions within the quotation text. The annotated attributes allow for a comprehensive evaluation of models of quotation attribution and coreference for literary texts. Keywords: quotation attribution, literature, coreference 1. Introduction Computational analysis of literary texts looks into Nat- ural Language Processing (NLP) techniques to model aspects of narrative, events, and characters (Elsner, 2012; Bamman et al., 2014; Vishnubhotla et al., 2019). Past work in this area has focused mainly on analysing works of fiction, drawn from open-source platforms like Project Gutenberg1 (Brooke et al., 2015; Bamman et al., 2020). The idiosyncrasies of literary text present several challenges to NLP models for named entity recognition, coreference resolution, character cluster- ing, event detection, and speaker identification. The typical length of a text is several thousands of tokens, and the format and structure of the content vary widely depending on the genre, topic, time-period, and author of the text. Characters are referred to by various aliases, often incorporating notions of familial relations (her fa- ther, Mr., Mrs., and Miss Bennet) or social titles (the baron); mentions such as the former also can refer to different entities if used by different speakers (my fa- ther). Consider, for example, the very first quotation in Jane Austen’s Pride and Prejudice: “My dear Mr. Bennet,” said his lady to him one day, “have you heard that Netherfield Park is let at last?” Identifying the speaker of this quotation involves mak- ing several inferences: that the person being spoken to is Mr. Bennet, that the mention “his lady” refers to “Mr. Bennet’s” lady, and that this is a proxy for Mr. Bennet’s wife, who must be Mrs. Bennet (the first explicit men- tion of Mrs. Bennet is only in Chapter 2; several other characters are introduced to us in the meantime). Previous work attempting to solve this task of identi- fying the speaker of a quotation in the text, or quota- tion attribution, has explored both rule-based systems 1https://www.gutenberg.org/ (Glass and Bangay, 2007; Muzny et al., 2017) and ma- chine learning models that are trained on an annotated dataset of quotations and speakers (Elson and McK- eown, 2010; He et al., 2013; O’Keefe et al., 2012). Some of these approaches treat the task as a two-step problem, where quotations are first attached to …

Content truncated to 500 words