Beyond Canonical Texts: A Computational Analysis of Fanfiction

Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 2048–2053, Austin, Texas, November 1-5, 2016. c⃝2016 Association for Computational Linguistics Beyond Canonical Texts: A Computational Analysis of Fanfiction Smitha Milli Computer Science Division University of California, Berkeley smilli@berkeley.edu David Bamman School of Information University of California, Berkeley dbamman@berkeley.edu Abstract While much computational work on fiction has focused on works in the literary canon, user-created fanfiction presents a unique op- portunity to study an ecosystem of liter- ary production and consumption, embodying qualities both of large-scale literary data (55 billion tokens) and also a social network (with over 2 million users). We present several em- pirical analyses of this data in order to illus- trate the range of affordances it presents to re- search in NLP, computational social science and the digital humanities. We find that fan- fiction deprioritizes main protagonists in com- parison to canonical texts, has a statistically significant difference in attention allocated to female characters, and offers a framework for developing models of reader reactions to sto- ries. 1 Introduction The development of large-scale book collections— such as Project Gutenberg, Google Books, and the HathiTrust—has given rise to serious effort in the analysis and computational modeling of fiction (Mo- hammad, 2011; Elsner, 2012; Bamman et al., 2014; Jockers, 2015; Chaturvedi et al., 2015; Vala et al., 2015; Iyyer et al., 2016). Of necessity, this work often reasons over historical texts that have been in print for decades, and where the only relationship between the author and the readers is mediated by the text itself. In this work, we present a computa- tional analysis of a genre that defines an alternative relationship, blending aspects of literary production, consumption, and communication in a single, vi- brant ecosystem: fanfiction. Fanfiction is fan-created fiction based on a previ- ously existing, original work of literature. For clarity we will use the term CANON to refer to the original work on which a fanfiction story is based (e.g. Harry Potter) and the term STORY to refer to a single fan- authored story for some canon. Although stories are based on an original canoni- cal work and feature characters from the canon, fans frequently alter and reinterpret the canon—changing its setting, playing out an alternative ending, adding an original character, exploring a minor character more deeply, or modifying the relationships between characters (Barnes, 2015; Van Steenhuyse, 2011; Thomas, 2011). In this work, we present an empirical analysis of this genre, and highlight several unique affordances this data presents for contemporary research in NLP, computational social science, and the digital human- ities. Our work is the first to apply computational methods to fanfiction; in presenting this analysis, we hope to excite other work in this area. 2 Fanfiction data Our data, collected between March–April 2016, originates from fanfiction.net.1 In this data, AUTHORS publish stories serially (one chapter at a time); REVIEWERS comment on those chapters. A summary of data is presented in table 1. The scale of this data is large for …

Content truncated to 500 words