New Narratives: Stories and Storytelling in the Digital Age

6 Digitized Corpora as Theory-Building Resource

6 Digitized Corpora as Theory-Building Resource New Methods for Narrative Inquiry andrew salway and david herman Introduction: Key Questions and Core Concepts Whereas previous work at the intersection of media studies and narrative theory has explored the impact of the digital media on processes of storytelling and on the interpretation (or experience) of storyworlds,1 our chapter outlines a different approach, examining how computer-assisted analysis of digitized texts can provide new resources for developing a theory of narrative itself. The chapter derives from our collaborative work, which we have christened the Corpus Narratology Initiative (http:// people.cohums.ohio-state.edu/herman145/CNI.html) and which centers on the following key question: will coming to terms with large narrative corpora—not single narratives or even groups of stories but rather multimillion -word collections of narratively organized texts—alter the foundational concepts of narrative theory? Or, to put the question in somewhat more specific terms, what methods for studying large amounts of textual data have been developed in other fields, such as corpus linguistics, and how might incorporation of those methods into narrative inquiry afford new foundations for the study of stories, and perhaps also new applications for narratological research? Thus, rather than trying to develop narratological tools to analyze the narratives now being conveyed or coenacted in digital environments, we seek to show how understandings of what stories are and how they work may need to be rethought in light of concepts and methods used to study digitized corpora.2 To assess how corpus-analytic methods bear on the core concepts and explanatory aims of narrative inquiry, we examine two broad approaches : top-down or hypothesis-driven approaches, and bottom-up or datadriven approaches. Top-down methods have been used in stylistics-based research that begins with categories of structure proposed in advance 120 by analysts and then seeks to (dis)confirm their existence—and study their patterns of distribution—in textual corpora. In the area of narrative research, the top-down approach has been exploited, as Wynne (2006) notes, by Semino and Short (2004); this work builds on earlier studies by Semino, Short, and Culpepper (1997) and Semino, Short, and Wynne (1999). Collectively, these researchers take the modes of speech and thought representation identified by Leech and Short 2007) as their starting point and then test the degree to which the initial hypothesis about the range of available modes (including direct, indirect, and free indirect modes) is borne out by distributional patterns found in an actual corpus of narrative texts—a corpus that in fact suggests the need to make scalar distinctions between modes that were cast as discretely different in the original model. More generally, corpus-enabled research methods of this kind fall into a first family of approaches that Adolphs (2006, 38–39) characterizes as testing hypotheses; Tognini-Bonelli (2001) terms this same family of approaches corpus-based. In our next section, to clarify the possibilities and limitations of such methods, we revisit as a case study an earlier research project by one of the authors (Herman 2005) that likewise draws on top-down or hypothesis-driven strategies for analysis. Focusing on the level of the story or fabula (= the “what” of the narrative) rather than the level of discourse or sjuzhet (= how that “what” is conveyed), this pilot study uses quantitative evidence to test hypotheses about genrebased preferences for representing actions and events. Our discussion below indicates how our collaborative work has yielded new perspectives on the problems as well as the potentials of this earlier research. What is more, revisiting the earlier project allows us to throw into relief the scope and aims of the second broad family of methods for analyzing digitized corpora. This second category of studies, which Adolphs (2006) associates with generating hypotheses and which Tognini-Bonelli (2001) characterizes as corpus-driven methods, can be described as bottom-up in its general orientation. Corpus-enabled research of this second kind seeks to remain as much as possible at the surface level of the texts included in corpora, rather than assuming beforehand that some features will be more relevant than others for the analysis of those texts. Translated into the domain of research on narratively organized discourse, the bottomup approach begins with textual features that are computationally tractable , aiming to work up from there to an account of the structures and Andrew Salway and David Herman 121 functions of narrative...