Dialogical Signals of Stance Taking in Spontaneous Conversation

Abstract

This is one of the first computational studies to investigate dialogical aspects of stance taking in spontaneous, spoken dialogue with a focus on lexical similarities. In any dialogic inter- action, each speaker influences the others’ lexical choices and aspects of their grammatical style (Brennan, 1996; Niederhoffer and Pennebaker, 2002). For this study, I leverage two distinct corpora. The ATAROS corpus (Freeman et al., 2014; Freeman, 2015) contains a series of task-oriented, collaborative tasks recorded in a controlled laboratory environment. The other corpus is drawn from a United States Homeland Security Subcommittee hearing regarding the 2007 - 2008 financial crisis. As such, it represents stance taking in an inherently adversarial environment, where high-stakes, real-world issues are being discussed. Both are annotated at the spurt-level with a 3-way stance strength annotation. I will show, through various experimental studies, that speakers show different patterns of dialogical behaviour when expressing stance versus when they are not; they show a higher level of engagement, as demonstrated by the use of similar or related terminology, and the rate of convergence in linguistic style, as measured by function word use, is also higher when expressing stance.