The Teams Corpus and Entrainment in Multi-Party Spoken Dialogues

Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 1421–1431, Austin, Texas, November 1-5, 2016. c⃝2016 Association for Computational Linguistics The Teams Corpus and Entrainment in Multi-Party Spoken Dialogues Diane Litman University of Pittsburgh Pittsburgh, PA Susannah Paletz University of Maryland College Park, MD Zahra Rahimi and Stefani Allegretti and Caitlin Rice University of Pittsburgh Pittsburgh, PA Abstract When interacting individuals entrain, they be- gin to speak more like each other. To sup- port research on entrainment in cooperative multi-party dialogues, we have created a cor- pus where teams of three or four speakers play two rounds of a cooperative board game. We describe the experimental design and techni- cal infrastructure used to collect our corpus, which consists of audio, video, transcriptions, and questionnaire data for 63 teams (47 hours of audio). We illustrate the use of our corpus as a novel resource for studying team entrain- ment by 1) developing and evaluating team- level acoustic-prosodic entrainment measures that extend existing dyad measures, and 2) investigating relationships between team en- trainment and participation dominance. 1 Introduction Linguistic entrainment1 refers to the convergence of (para)linguistic features across speakers during con- versation (Brennan and Clark, 1996; Porzel et al., 2006). Research has found that speakers entrain to both human and computer conversational partners, with the amount of entrainment often positively re- lated to conversational and task success. However, most prior work has focused on the study of entrain- ment during two-party dialogues, rather than during the multi-party conversations typical of teams. To support the study of entrainment during multi- party cooperative dialogue, we have created a large- scale corpus (over 47 hours of recordings) of teams 1Other terms in the literature include accommodation, adap- tation, alignment, convergence, coordination and priming. of three or four speakers playing a cooperative board game requiring conversation. The corpus consists of audio, video, transcriptions, and questionnaire data for 63 teams. The goal of the corpus is to provide a freely-available data resource for the development and evaluation of multi-party entrainment measures that can be 1) computed using language technolo- gies, 2) motivated and validated by the literature on teams, and 3) associated with measures of task and dialogue success. In this paper, we first describe the experimental design and technical infrastructure used to create our corpus. We then present two case studies illustrat- ing the use of our corpus as a novel resource for studying team entrainment: quantifying acoustic- prosodic entrainment at the team-level rather than the dyad-level, and incorporating a construct from the teamwork literature into the study of entrain- ment. 2 Background and Related Work The development of methods for automatically quantifying entrainment in text and speech data is an active research area, as entrainment has been shown to correlate with success measures or with social variables for a variety of phenomena, e.g., acoustic-prosodic, lexical, and syntactic (Nenkova et al., 2008; Reitter and Moore, 2007; Mitchell et al., 2012; Levitan et al., 2012; Lee et al., 2011; Stoyanchev and Stent, 2009; …

Content truncated to 500 words