A Multidimensional Lexicon for Interpersonal Stancetaking

Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 884–895 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1082 Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 884–895 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1082 A Multidimensional Lexicon for Interpersonal Stancetaking Umashanthi Pavalanathan Georgia Institute of Technology Atlanta, GA umashanthi@gatech.edu Jim Fitzpatrick University of Pittsburgh Pittsburgh, PA jim.fitzpatrick@gmail.com Scott F. Kiesling University of Pittsburgh Pittsburgh, PA kiesling@pitt.edu Jacob Eisenstein Georgia Institute of Technology Atlanta, GA jacobe@gatech.edu Abstract The sociolinguistic construct of stancetak- ing describes the activities through which discourse participants create and signal re- lationships to their interlocutors, to the topic of discussion, and to the talk it- self. Stancetaking underlies a wide range of interactional phenomena, relating to formality, politeness, affect, and subjec- tivity. We present a computational ap- proach to stancetaking, in which we build a theoretically-motivated lexicon of stance markers, and then use multidimensional analysis to identify a set of underlying stance dimensions. We validate these dimensions intrinsically and extrinsically, showing that they are internally coherent, match pre-registered hypotheses, and cor- relate with social phenomena. 1 Introduction What does it mean to be welcoming or standoffish, light-hearted or cynical? Such interactional styles are performed primarily with language, yet little is known about how linguistic resources are arrayed to create these social impressions. The sociolin- guistic concept of interpersonal stancetaking at- tempts to answer this question, by providing a con- ceptual framework that accounts for a range of in- terpersonal phenomena, subsuming formality, po- liteness, and subjectivity (Du Bois, 2007).1 This 1Stancetaking is distinct from the notion of stance which corresponds to a position in a debate (Walker et al., 2012). Similarly, Freeman et al. (2014) correlate phonetic features with the strength of such argumentative stances. framework has been applied almost exclusively through qualitative methods, using close readings of individual texts or dialogs to uncover how lan- guage is used to position individuals with respect to their interlocutors and readers. We attempt the first large-scale operationaliza- tion of stancetaking through computational meth- ods. Du Bois (2007) formalizes stancetaking as a multi-dimensional construct, reflecting the re- lationship of discourse participants to (a) the au- dience or interlocutor; (b) the topic of discourse; (c) the talk or text itself. However, the multi- dimensional nature of stancetaking poses prob- lems for traditional computational approaches, in which labeled data is obtained by relying on anno- tator intuitions about scalar concepts such polite- ness (Danescu-Niculescu-Mizil et al., 2013) and formality (Pavlick and Tetreault, 2016). Instead, our approach is based on theoretically-guided application of unsuperviseda learning, in the form of factor analysis, applied to lexical features. Stancetaking is characterized in large part by an array of linguistic features ranging from discourse markers such as actually to backchannels such as yep (Kiesling, 2009). We therefore first compile a lexicon of stance markers, combining prior lexicons from Biber and Finegan (1989) and the Switchboard Dialogue …

Content truncated to 500 words