Capturing turn-by-turn lexical similarity in text-based communication

Abstract

Speakers often come to use similar words during conversation; that is, they come to exhibit lexical similarity. The extent to which this occurs is associated with many positive social outcomes. However, existing measures of lexical similarity are either highly labor intensive or too coarse in their temporal resolution. This limits the ability of researchers to study lexical similarity as it unfolds over the course of a conversation. We present a fully automated metric for tracking lexical similarity over time, and demonstrate it on individual conversations, explore general trends in aggregate conversational dynamics, and examine differences in how similarity tracks over time in groups with differing social outcomes.