Coming to Terms: A Discourse Epistemetrics Study of Article Abstracts from the Web of Science

This study investigates the relative power and characteristics of a set of social and epistemic terms to distinguish among disciplines of research article abstracts, using a corpus of 928,572 abstracts from 13 disciplines indexed by Web of Science in 2011. Applying the machine-learning approach to discourse epistemetrics using a sequential minimal optimization (SMO) algorithm, and a feature set of terms derived from Hyland’s (2005) metadiscourse studies per Demarest and Sugimoto (2014), the current paper reports subsets of terms that best (and least) distinguish among disciplines, finding that the terms least able to distinguish among disciplines are rarely used and overwhelmingly adjectival or adverbial markers of authorial attitude, reflecting personal positioning, while terms best able to distinguish disciplines are mostly verbs frequently used as engagement markers, framing the generation of knowledge for the readership in ways that are standardized within disciplines (while varying among them). We plan to analyze the findings of the current research-in-progress from discipline-based as well as term-based perspectives, incorporating both into a two-mode network, as well as incorporating finer grained data for specific specializations to compare with the current higher-level disciplinary findings.