Please Clap: Modeling Applause in Campaign Speeches

Proceedings of NAACL-HLT 2018, pages 92–102 New Orleans, Louisiana, June 1 - 6, 2018. c⃝2018 Association for Computational Linguistics Please Clap: Modeling Applause in Campaign Speeches Jon Gillick School of Information University of California, Berkeley jongillick@berkeley.edu David Bamman School of Information University of California, Berkeley dbamman@berkeley.edu Abstract This work examines the rhetorical techniques that speakers employ during political cam- paigns. We introduce a new corpus of speeches from campaign events in the months leading up to the 2016 U.S. presidential elec- tion and develop new models for predicting moments of audience applause. In contrast to existing datasets, we tackle the challenge of working with transcripts that derive from un- corrected closed captioning, using associated audio recordings to automatically extract and align labels for instances of audience applause. In prediction experiments, we find that lexical features carry the most information, but that a variety of features are predictive, including prosody, long-term contextual dependencies, and theoretically motivated features designed to capture rhetorical techniques. 1 Introduction Every public speech involving a large audience can be seen as a game of coordination (Asch, 1951): at each moment, each individual mem- ber of the audience must decide in a split second whether to applaud at what has just been said. Ap- plause is a potentially risky action: if an individual spontaneously claps but no one joins in, they suf- fer some negative social cost; the game is to judge from their own private information and content of the speech whether the rest of the audience will applaud at the same time they do. Because of this cost, audiences respond to sev- eral interacting factors in a speaker’s behavior: a.) the content of the message; b.) their deliv- ery (so that changes in pitch, duration and gaze signal salient moments for which applause may be licensed); and c.) the verbal design of the message—those rhetorical strategies that speakers use to signal that applause is welcome (Atkinson, 1984; Heritage and Greatbatch, 1986). In this work, we attempt to model all three of these dimensions in developing a computational model for applause. While past work has focused on these elements in isolation (Guerini et al., 2015; Liu et al., 2017) or for related problems such as laughter detection (Purandare and Litman, 2006; Chen and Lee, 2017; Bertero and Fung, 2016), we find that developing a holistic model encompass- ing all three aspects yields the most robust predic- tor of applause. We focus on political speeches, and in particular those at campaign rallies, which lend themselves well to analysis of rhetorical strategies for several reasons. First, the speakers at these events prior- itize maintaining the crowd’s attention (Strangert, 2005). Motivated to drum up excitement and fer- vor among their supporters that they hope will carry beyond the event and into the voting booth, speakers pull out their strongest rhetorical tactics. Second, campaign speeches usually consist of a series of self-contained messages that can be fully expressed within a few utterances (Heritage and Greatbatch, 1986), yielding a well-defined obser- vation …

Content truncated to 500 words