Speaking Similarly: Team Personality Composition and Acoustic-Prosodic Entrainment

Abstract

This study introduces the concept of acoustic-prosodic entrainment (ways people speak similarly). We review prior research on entrainment theory and methods from computational linguistics, and then apply this concept to team research by examining the relationship between team personality composition and subsequent entrainment in an exploratory case study. With 62 teams playing a cooperative board game, team average Agreeableness and team Agreeableness diversity positively, and Openness to Experience diversity negatively, preceded different kinds of entrainment. This study suggests entrainment is not a singular construct. Small group researchers could leverage technological, methodological, and conceptual advances in computational linguistics to study emergent team processes.

Keywords

entrainment group dynamic-based weighted convergence personality composition acoustic-prosodic team composition

The analysis of conversations has emerged and persisted as one method for studying team emergent states and processes (e.g., Ervin et al., 2017; Kane & van Swol, 2022; Lehmann-Willenbrock & Chiu, 2018; Paletz et al., 2017). One aspect of the back-and-forth nature of communication between individuals within teams is that individuals may speak similarly to each other over time. This phenomenon has been described variously across disciplines as mimicry (e.g., in psychology, Chartrand & Lakin, 2013), accommodation and adaptation (in communication, Giles et al., 1991), or entrainment (in computational linguistics, Levitan et al., 2012). Entrainment research from computational linguistics, which is our focus, has involved both the automation of measurement—an increasingly important technique to speed up communication research—and the development of conversational robots and tutors with the hope that their entrainment to their human partners will increase human rapport and learning (e.g., Gessinger et al., 2021; Kory-Westlund & Breazeal, 2019; Lubold et al., 2018). Connecting this research to questions and methods within small group research can benefit both the computational linguistics study of communication and team science.

We have two goals for this paper. First, we introduce and explain how acoustic-prosodic entrainment is conceptualized and measured within computational linguistics to the Small Group Research readership. We contribute by conceptualizing acoustic-prosodic entrainment as a team construct, which is rarely done in its current form and could be a benefit to team science. We detail how entrainment as we define it differs from entrainment as described in Ancona and Chong (1992, 1996). In addition, our field of small group research too rarely measures behavioral processes directly (Lehmann-Willenbrock & Allen, 2018) or conducts language analysis (Kane & van Swol, 2022). Measures of dynamic processes can be difficult and/or rare (Luciano et al., 2018). By drawing on new measures of acoustic-prosodic entrainment, we heed calls to bridge computer science, particularly sensor-based research, and social science to examine interactions (Lehmann-Willenbrock & Hung, 2023; Lehmann-Willenbrock et al., 2017; Müller et al., 2019). While developed in a different field for different purposes, acoustic-prosodic entrainment involves concepts and measures of team conversations to the benefit of theorizing and hypothesis-testing research within small group research.

The second goal of this paper is to demonstrate how using the concept and measurement of entrainment can achieve new insights via a case study. This case explores potential relationships between team personality composition and subsequent entrainment. Research on why and under what circumstances team entrainment may occur is still in its infancy. Small group research often presumes an input-process-output-input model (e.g., Ilgen et al., 2005), even though teams are increasingly considered to be multilevel, dynamic, networked, and complex (Mathieu et al., 2019). In team science, there are a variety of relationships between team personality composition (as inputs) and team outcomes such as performance (e.g., Barrick et al., 1998; Bell, 2007; Mohammed & Angell, 2003; Peeters et al., 2006; Prewett et al., 2009)—with emergent team processes presumed to link the two. While much research on entrainment has examined subsequent outcomes (e.g., Beňuš, 2014; Kory-Westlund & Breazeal, 2019), only some studies have examined potential antecedents to entrainment (e.g., Danescu-Niculescu-Mizil et al., 2012; Kämpf et al., 2018; see discussion later). These studies also generally examine entrainment as the behavior of one individual toward another (focusing on the individual level, e.g., a participant entraining to someone speaking another language) or interacting dyads (e.g., dyadic level of two people converging), rather than as a team level emergent process. From the perspective of those who wish to increase entrainment in human teams due to its theorized connection to rapport and other positive social outcomes, the literature on team composition antecedents is limited. Thus, our study explores whether and what types of team personality composition might impact what types of acoustic-prosodic entrainment, as communication emergent processes. In addition, this case study gives us a greater understanding of what types and measures of acoustic-prosodic entrainment are distinct from each other, and their potential association with different team personality precursors.

As a quantitative, multidisciplinary, exploratory study that can help to build new theories about processes (Bamberger & Ang, 2016), this research tackles the relationships between variables from two fields: Our entrainment measures were created, and corpus collected, using one approach—computational linguistics/intelligent systems—but we ask questions based in the social/personality psychology of teams. For this case study, we leverage the Teams Corpus (Litman et al., 2016), lab data collected for multidisciplinary purposes. Within this paper, we also reflect on the challenges and opportunities of conducting cross-disciplinary team science.

Acoustic-Prosodic Entrainment Theory and Research

Sociolinguistics researchers have long suggested that individuals in dyads and teams automatically start to speak more similarly to each other (Giles et al., 1973). Communication Accommodation Theory suggests that this similarity occurs, such that “individuals adapt to each other’s communicative behaviors in terms of a wide range of linguistic-prosodic-nonverbal features” (Giles et al., 1991, p. 7). This theory proposes that (even) unconscious entrainment¹ occurs due to social identity and social integration motives, based on research suggesting that similarity leads to greater liking and social attractiveness (Giles et al., 1991) and that similarity indicates social closeness versus distance (Beňuš, 2014). Not all people speak similarly all the time on all dimensions, however, as entrainment may not be present due to social norms, status, identity maintenance through language/accent use, communication goals (e.g., a parent trying to calm a child), or complementarity (e.g., women on dates in past research both entraining and diverging by enacting “feminine” communication styles; Giles et al., 1991).

Accommodation or entrainment used here is not the same as Ancona and Chong’s (1992, 1996) definition in organizational theory, which was borrowed from biology: “the adjustment or moderation of one behavior to synchronize or to be in rhythm with another” (Ancona & Chong, 1992, p. 166). Specifically, Ancona and Chong examined tempo entrainment, or when processes operated at the same speed or pace; synchronic entrainment, which involved similar rhythms/patterns and pace; and harmonic entrainment, or when observers consider two processes to be in harmony even if their pace or rhythm were different (Ancona & Chong, 1992). Their conceptualization of entrainment is broader than the one we describe from computational linguistics; the Ancona and Chong (1992, 1996) conception refers to an array of synchronized activities, not just communication, over longer and larger time frames (e.g., academic semesters, product development cycles, Zellmer-Bruhn et al., 2003). As such, entrainment in their theory is a sociological concept of choreography aligned with multiple intergroup rhythms at once (e.g., with top management, organizational division, and customers; Ancona & Waller, 2007). For example, “teams may become entrained to the pace of activity occurring on other parts of the organization or other levels of the team context” (Zellmer-Bruhn et al., 2003, p. 137). In other words, the concept championed by Ancona and colleagues is related to, but not the same as, the concept of entrainment within computational linguistics, where entrainment is conceptualized as how individuals communicate, including speaking similarly over time—and is usually examined in shorter time frames, such as a meeting.

While studies drawing on communication accommodation or entrainment have used a variety of research paradigms, disciplines, and communication media (e.g., face to face, phone, online), most research in sociolinguistic settings involves dyadic interactions or what we would consider individual-level entrainment—an individual’s entrainment toward others (e.g., immigrants and their speech patterns; Giles et al., 1991). Entrainment has been examined on a variety of dimensions including nonverbal (e.g., smiles, gaze), linguistic and lexical (e.g., specific words), and acoustic-prosodic signals such as loudness (e.g., Anzalone et al., 2015; Chartrand & Lakin, 2013; Giles et al., 1991; Lubold et al., 2018; Tomprou et al., 2021; Van Swol & Kane, 2019). Moment-by-moment entrainment between infants and mothers on a variety of nonverbal aspects is considered important for secure attachment, and entrainment between adults can promote perceptions of being a social unit and build rapport (Delaherche et al., 2012). Despite the variety in measures and potential explanations for the phenomenon, the overarching theory remains consistent: people often unconsciously match how others speak and behave over time, and this matching generally has positive implications for liking and social outcomes (Chartrand & Lakin, 2013; Delaherche et al., 2012; Doré & Morris, 2018; Sinha & Cassell, 2015; Van Swol & Kane, 2019). Regardless of how it is measured, entrainment is a dynamic behavioral construct that involves changes over time, rather than being a static or episodic phenomenon (Luciano et al., 2018). Even when it is measured as a single number, that number is meant to capture the degree of change that has occurred.

Recent advances in computational linguistics have enabled the creation of new acoustic-prosodic measures of entrainment over the past decade (e.g., Gálvez, Gauder et al., 2020; Levitan & Hirschberg, 2011). Acoustic-prosodic dimensions include the rhythm, frequency, loudness (intensity), and intonation of speech, such as pitch. Pitch is the perceived height of tone (high vs. low notes), and intensity is the loudness of the sound of speech (Lubold & Pon-Barry, 2014). This literature also defines different types of entrainment such as proximity, convergence, and synchrony. Proximity is similarity between the speakers’ dimension (e.g., pitch) compared to a baseline of non-conversational partners, whereas convergence is similarity change over time, which is a more truly dynamic measure as it captures whether speakers become closer to each other in specific dimensions toward the end of a conversation versus the beginning (Levitan & Hirschberg, 2011). Synchrony is moment-by-moment coordination between speakers (Levitan & Hirschberg, 2011). In other words, while convergence studies yield a single measure of the extent of convergence occurred over a conversation or meeting, synchrony involves the moment-by-moment changes in behavior matching in a dynamic and reciprocal manner (Delaherche et al., 2012).

Another aspect of entrainment is what features of a given acoustic-prosodic dimension are being examined, such as the standard deviation of pitch (i.e., the variety of high/low tones as sampled within a time period) and mean intensity (average loudness across samples within a time period). Acoustic-prosodic features are typically extracted automatically and separately for each speaker from high-quality audio recordings of conversations using software tools such as OpenSMILE (Eyben et al., 2010) or Praat (Boersma & van Heuven, 2001). Features can be extracted from any segment of speech as input. Speech segments can be associated with linguistic constructs such as speaker turns, or with temporal windows such as each minute of a recording. While research in dyadic entrainment has often examined short-term entrainment across speaker turns, in multi-party (group or team) conversations, a response to a speaker is not guaranteed to be in the next turn of the conversation. Thus, longer-term measurements based on temporal windows may be more appropriate units of analysis for team research, and this is the approach we adopt, as discussed below.

In sum, taking these three aspects together—dimensions, types, and features—researchers can examine any combination of the three to capture acoustic-prosodic entrainment, such as convergence on standard deviation of pitch, or synchrony of mean intensity. Drawing on an existing corpus (Litman et al., 2016) and promising directions from prior acoustic-prosodic research on dyad entrainment (e.g., Levitan & Hirschberg, 2011; Lubold & Pon-Barry, 2014), this study focuses on pitch and intensity (dimensions), convergence (type), and mean, max and standard deviation (features).

Entrainment as a Measurable Team Process

Research and measurement on acoustic-prosodic entrainment is still under development, with mixed findings. To measure acoustic-prosodic team (larger number of discussants than dyadic) convergence with respect to each potential combination of feature and dimension, our research builds upon a dyadic measure that calculates the difference between the dissimilarity of speakers in two nonoverlapping time intervals (Levitan & Hirschberg, 2011). The size of the time interval is a parameter of the convergence measure. If the dissimilarity in the second interval is less than in the first, the pair is considered to be converging, with the difference value indicating the strength of the convergence.

In the beginning of our research effort using our corpus, we initially proposed a team-level version of this dyadic measure via a non-weighted average across each dyad’s convergence (Litman et al., 2016). The partner difference for a speaker in a dyad within a given time interval was the absolute difference between the feature values for the speaker and their partner (Levitan & Hirschberg, 2011). For each team, we averaged these absolute values for all team members (see Equation 1, Litman et al., 2016). In addition, since differing time intervals had been examined in the dyadic entrainment literature, we explored the impact of comparing features extracted from the first versus last 3, 5, and 7 min of each conversation, as well as from the two conversational halves as in Levitan and Hirschberg (2011). Our results (Litman et al., 2016) suggested that when teams in our corpus converged on a feature, they did so earlier in the conversation; evidence of convergence was strongest when comparing the more distant temporal intervals (i.e., first vs. last 3 min).

However, as we continued our research, we realized we needed to consider aspects of the communication dynamics in our measurements of team entrainment. We thus modified the weighting scheme for the entrainment variables to better reflect group dynamics as compared to simply averaging dyad convergence. This modification was done by decreasing the contribution of speakers whose convergence behaviors differed from the rest of the group (e.g., Speaker 3 in Figure 1 for an illustration). In other words, the weight for a speaker should be the percentage of individuals who have the same convergence behavior as the speaker (see Methods for details). This decision improved performance on predicting social outcomes and differentiating between real and permuted conversations, two types of benchmark tasks previously used by the computational linguistics community to evaluate convergence measures (Rahimi & Litman, 2018).

Figure 1.

Illustration of a four-person team where all speakers except one are converging; lines connect the feature value for each individual.

Antecedents and Outcomes of Entrainment

Entrainment in general has been examined in human interaction processes such as social integration, perceived likability, dominance, and friendship formation (e.g., Beňuš, 2014; Kovacs & Kleinbaum, 2020). Measured in various ways, entrainment has generally shown positive effects on social (and occasionally task) outcomes, in line with Communication Accommodation Theory. For example, social robotics researchers have examined the potential effects of acoustic-prosodic entrainment by a robot on human outcomes (e.g., Lubold et al., 2021). Social robots that used entrainment (pitch and rate) led children to express more positive and fewer negative emotions in a storytelling task (Kory-Westlund & Breazeal, 2019), and robot pitch entrainment combined with social language resulted in greater mathematics learning in the human students than with non-social robots (Lubold et al., 2018). Gálvez, Gravano, Beňuš, et al. (2020) had mixed findings when a conversational avatar used entrainment: Intensity entrainment had a positive effect on trust, but pitch entrainment had a negative effect on trust, suggesting that entrainment on pitch may not be beneficial in the context of social robots. However, designing robot-to-human social interactions is different from naturally occurring human teamwork (e.g., Van Swol & Kane, 2019).

In human dyads, acoustic-prosodic entrainment has been linked to positive social outcomes, but the results depend on the type and dimensions of entrainment (e.g., Gálvez, Gauder et al., 2020). Acoustic-prosodic entrainment in dyads is positively related to conversational quality (Michalsky et al., 2018) and rapport (Lubold & Pon-Barry, 2014). However, there are mixed findings, as the relationship between entrainment and these outcomes depends on the acoustic dimensions (e.g., pitch but not intensity) and, as suggested by Communication Accommodation Theory, their relevance to social identity (e.g., accents) and perceptual salience to listeners (Levitan, 2020).

In addition, most of this research has examined dyads rather than teams, and further studies are needed to explore antecedents, such as personality, in addition to the presence and consequences of entrainment (e.g., Levitan & Hirschberg, 2011). The Teams Corpus (Litman et al., 2016) provided an opportunity to do both. Past research with this corpus found that both intensity maximum and standard deviation features of convergence entrainment were positively related to social outcomes, as measured by a composite of self-reported team emergent state variables (e.g., cohesion, group efficacy; Rahimi & Litman, 2018). A lexical measure of entrainment based on the Linguistic Inquiry and Word Count (LIWC, Pennebaker et al., 2015) had negative associations with both task and process conflict; there were also relationships between lexical entrainment, team demographic diversity, team size, and social outcomes (M. Yu et al., 2019). For instance, teams with higher gender diversity had higher levels of unweighted minimum lexical convergence (M. Yu et al., 2019). Different versions of the team acoustic-prosodic metrics of entrainment (e.g., proximity measures of intensity standard deviation vs. proximity pitch mean vs. different convergence measures) were positively and negatively related to self-reported social outcomes; were positively and negatively correlated with lexical (non-LIWC) entrainment metrics; and when used in a multi-modal (combined) manner, were better able to predict social outcomes (Rahimi et al., 2017). In a separate set of analyses, a more local lexical (using LIWC) measure of entrainment also predicted social outcomes (Rahimi & Litman, 2020).

One challenge in most entrainment research is that, while entrainment is examined using many types of dimensions, types, and features, the predictions tend to be the same despite empirical evidence suggesting differences between them (Levitan, 2020). In other words, many studies take a high-level view of Communication Accommodation Theory and suggest that entrainment on any dimensions, features, and types will predict of positive social outcomes, despite results supporting this hypothesis only for certain measures (e.g., Beňuš, 2014; Levitan, 2020; Rahimi et al., 2017).

Case Study: Linking Team Personality Composition to Team Entrainment

While research on Communication Accommodation Theory examines when and why individuals entrain (Giles et al., 1991), research on under what circumstances team entrainment may occur is still in its infancy. This study explores the potential associations of team personality composition on a set of new measures of team (multi-party, rather than dyadic) acoustic-prosodic entrainment.

Personality is a set of relatively stable individual differences that impacts behavior across situations (John & Srivastava, 1999). Most studies of team personality composition draw on the Big Five Model, a rigorously-developed model of personality that distinguishes five main dimensions of personality (John & Srivastava, 1999; John et al., 2008). Extraversion is the degree to which the person is gregarious, assertive, talkative, enthusiastic, and seeks excitement and stimulation. Agreeableness involves being trusting, compliant, kind, and altruistic. Conscientiousness involves how reliable, thorough, and organized a person is. Openness to Experience includes being curious, artistic, imaginative, and having wide interests. Neuroticism (also termed Negative Emotionality) is the degree to which a person experiences high levels of anxiety, depression, and general emotionality/emotional volatility. The Neuroticism dimension is sometimes called Emotional Stability, which is the opposite (or flipped) version of this dimension and indicates high level of calmness and even-temperedness (John et al., 2008). Team personality composition can be examined in different ways (e.g., Barrick et al., 1998), but is commonly examined as an amount, level, or elevation (e.g., average, maximum, minimum) and as its diversity or variance (e.g., measured via standard deviation, Neuman et al., 1999).

Relationships between team personality and outcomes (e.g., team performance) tend to be small, whereas it may be “easier to find empirical support for relations between team personality and team processes” (Prewett et al., 2009, p. 289)—and entrainment is a type of team emergent process. Although multiple meta-analyses examined the effects of team composition on outcomes (e.g., Bell, 2007; Peeters et al., 2006), the effects of team personality composition on communication dynamics are understudied in terms of both empirical knowledge and theory. If particular personality composition variables are related to certain entrainment variables but not others, it helps us unpack differences between various acoustic convergence phenomena and suggests which dimensions, types, and features could be pursued for further investigation regarding entrainment as a potential team process mediator between team personality and outcomes. In addition, while this study does not make strong causal claims, there is no possibility for reverse-causal effects.

A challenge in our current theorizing is that the extant theory does not describe why or how teams entrain. Prior research describes why one individual might entrain to others (Lewandowski & Jilka, 2019), why dyads might entrain with each other (e.g., mother/child pairs, Delaherche et al., 2012), or how dyadic pairs entrain (e.g., Levitan & Hirschberg, 2011), but not how team entrainment comes about. There is, however, research on the associations between individual personality and tendencies to entrain (see Table 1). There are individual differences in the tendency to mimic, and initial liking may also lead to mimicry in interactions (Kämpf et al., 2018). Individual Agreeableness is likely related to individual entrainment behaviors because affiliative motivation is related to mimicry (Chartrand & Lakin, 2013). According to Communication Accommodation Theory (Giles et al., 1991), people who want to be liked are more likely to entrain to others, and those who resist conformity would have a divergent communication style to highlight their individual identities. In one study of humans entraining to robots, Agreeableness was not found to be associated with entrainment (Brandstetter et al., 2017), although this finding may be because the targets in this study were robots rather than humans. Conscientiousness may or may not be related to entrainment. While entrainment has been occasionally associated with task performance (e.g., Lubold et al., 2018), and Conscientiousness (both individual and team-level) are task-oriented and related to team performance (e.g., Prewett et al., 2009), these findings do not imply much in terms of predictions about entrainment and Conscientiousness.

Table 1.

Traits and Entrainment Literature Summary.

Study	Trait (Related Big Five Dimension in Parentheses)	Entrainment-Related Finding
Chartrand and Lakin (2013)	Affiliative motivation (high Agreeableness; possibly high Extraversion on enthusiasm)	More likely to entrain
Giles et al. (1991)	People who want to be liked (high Agreeableness) People who resist conformity (low Agreeableness)	More likely to entrain Less likely to entrain
Brandstetter et al. (2017)	Higher Agreeableness Higher Openness to Experience	Less likely to entrain to robots on pitch or intensity More likely to adopt words introduced by peers
Keltner et al. (2003)	Greater power (low Agreeableness, high Extraversion on dominance)	More disinhibition (less entrainment?)
Berdahl and Martorana (2006)	Greater power (low Agreeableness, high Extraversion on dominance)	More likely to feel can openly express opinions (less entrainment?)
Danescu-Niculescu-Mizil et al. (2012) (see Van Swol & Kane, 2019 for review)	Individuals in a team with others in higher positions of power (Extraversion diversity)	People entrain toward those higher in power on specific function words
Gessinger et al. (2021)	Higher Neuroticism (lower Emotional Stability)	More likely to entrain with a programed tutor with rising intonation
Lewandowski and Jilka (2019)	Higher Neuroticism (lower Emotional Stability) Higher Openness to Experience	More likely to entrain on English pronunciation of interaction partner
A. C. L. Yu et al. (2013)	Higher Openness to Experience	More likely to entrain on voice onset time

Individual Neuroticism, however, has been found to be positively related to individual entrainment: Individuals higher on Neuroticism were more likely to converge with a programed tutor by using rising intonation (Gessinger et al., 2021), and nonnative/German speakers higher on Neuroticism were more likely to entrain on the English pronunciation of their interacting partner (Lewandowski & Jilka, 2019). Individuals high on Neuroticism are more affected by distress and so may seek social approval more, and thus entrain because they are unconsciously trying to be positively evaluated (Lewandowski & Jilka, 2019).

On the individual level, Openness to Experience has been found to be positively related to entrainment via adopting words introduced by peers (Brandstetter et al., 2017) and positively related to voice onset time imitation, or entrainment on the length of time between the end of a word’s consonant and the beginning of the next word (A. C. L. Yu et al., 2013). Lewandowski and Jilka (2019) also found that Openness was positively associated with pronunciation entrainment. Those with high Openness to Experience seek out novelty and are also more open to different cultures and settings, and they thus may also be more perceptive and competent with communication in general (Lewandowski & Jilka, 2019).

Extraversion may include dominance/assertiveness, enthusiasm, and talkativeness (John et al., 2008). Power may be related to low Agreeableness and high Extraversion. In particular, power may be related to the Extraversion subdimension of dominance. Based on research that examined entrainment using linguistic style matching (LSM) or the matching of specific lists of function words using LIWC over conversations (Pennebaker et al., 2015; e.g., Gonzales et al., 2010; Kovacs & Kleinbaum, 2020), people may entrain toward someone who is in a higher position of power (Danescu-Niculescu-Mizil et al., 2012; see Van Swol & Kane, 2019 for a review). High-power individuals are more likely to feel they can openly express their opinions compared to low-power individuals (Berdahl & Martorana, 2006), and greater individual power leads to disinhibition in general (Keltner et al., 2003). Individuals with lower assertiveness may entrain to their more assertive teammates (Van Swol & Kane, 2019). However, dominance is only one dimension of Extraversion, which also includes talkativeness and enthusiasm (John et al., 2008).

Our study is exploratory, as this reviewed research does not offer clear predictions for team personality composition and entrainment. Of note, this study is about convergence entrainment, not vocal behaviors in general. For instance, while work stressors are associated with increased rate of speech and vocal intensity (Langer et al., 2022), our study examines convergence, regardless of the overall rates or levels of the acoustic-prosodic features. A team might or might not converge on low or high intensity. We examine potential relationships between team personality composition involving elevation (levels or mean) and diversity (standard deviation) of the Big Five personality dimensions as precursors, and specific aspects of acoustic-prosodic entrainment as dependent variables. Our overarching research question is thus as follows:

Are Big Five personality dimensions (Agreeableness, Extraversion, Conscientiousness, Neuroticism, and Openness to Experience), when measured as elevation/level (mean) and diversity (standard deviation), related to subsequent team convergence entrainment, as measured using pitch and intensity?

The Teams Corpus

To explore the effects of team personality composition on team entrainment, we drew on the Teams Corpus.² The goals of the Teams Corpus were to create a dataset that would enable the development of team, rather than dyad, acoustic-prosodic entrainment measures and to examine the relationships between entrainment and variables typically studied in small group research (Litman et al., 2016; Rahimi & Litman, 2018).³ The Teams Corpus entailed face-to-face teams of three or four participants playing a cooperative board game, Forbidden Island™, twice, under experimental conditions, after being trained on how to play the game (Litman et al., 2016). Each individual was audio-recorded through headset microphones and videorecorded as well. The project was a collaboration between a computational linguist/computer scientist and a social/personality psychologist. The computer scientist was interested in developing measures that could help detect entrainment in real groups, but also could be deployed by intelligent agent tutors in team situations. The psychologist was interested in whether entrainment could be a behavioral mediator for existing findings between team composition and outcomes, and whether entrainment, as a behavioral measure, would correlate with common self-report measures of team processes and emergent states (e.g., cohesion). The original corpus therefore includes a variety of before- (e.g., personality, demographics, experience playing games) and after-game self-report measures (e.g., cohesion, shared mental models, perceived team efficacy, team potency). Task success in terms of winning the game ended up being of limited variability (everyone won the second game) and so has not been used.

The Teams Corpus thus serves as a useful case study for exploring the relationships between these variables, and it has been used in several studies. Indeed, one creative use of the corpus involved examining speaking overlap dynamics as a way of identifying individual personality (M. Yu et al., 2019). Timing of analyses was such that the computational questions and development proceeded first, and the final versions of the acoustic-prosodic entrainment variables are thus included in this study (i.e., the convergence measures that performed best on benchmark evaluation tasks in Rahimi & Litman, 2018).

At this time, the corpus has been used for a variety of research questions (see prior findings), but not whether team-aggregated personality is an antecedent of entrainment, which was one of the original questions to explore and why personality variables were included in data collection. Research using this corpus has mainly focused on developing a series of lexical and acoustic-prosodic measures of team entrainment (proximity and convergence, but not synchrony), a previous rarity in the literature (Litman et al., 2016; Rahimi, 2019; Rahimi et al., 2017; Rahimi & Litman, 2018, 2020; M. Yu et al., 2019). Over the course of the project, novel versions of entrainment measures were developed that weighted for small group dynamics for acoustic-prosodic (Rahimi, 2019; Rahimi & Litman, 2018; see Measures, below) and lexical (Rahimi & Litman, 2020) measures. The current study tackles the research question, using the novel, more sophisticated versions of the acoustic-prosodic entrainment measures, and specifically those for convergence, which are measures of dynamic phenomena rather than slices of time similarity (i.e., proximity).

Methods

Participants

The Teams Corpus consisted of 216 participants, comprising 63 teams of three or four players (62.5% female; 37.5% male). Because the recruitment sought out non-students via flyers both around a college campus and commercial districts in a city on the U.S. East Coast, the average age was 25 years old (SD = 11.3; Minimum = 18, Maximum = 67). In terms of ethnicity, 166 participants reported being Caucasian, 31 Asian/Pacific Islander, 24 Black, 10 Latino/Hispanic, and one Native American (multiple ethnicities could be reported). Participants read and signed consent forms before the study and received compensation after. One audio file was not saved properly, resulting in 62 teams of 213 participants, with 35 (56.5%) 3-person teams and 27 (43.5%) 4-person teams. One individual did not answer the question about cooperative game knowledge, so the maximum team knowledge of cooperative games variable was calculated from the other team members.

Procedure

Teams played the game Forbidden Island™ twice. Forbidden Island™ was chosen because it requires cooperation, communication, and strategy in real time; each player takes on different roles and there is game-specific terminology, making the game analogous to real-world multifunctional and multidisciplinary teams. Each player goes in turn, but they can freely discuss each other’s actions. Half the teams had a minimal team training intervention before the first game, which involved a review of teamwork concepts based on a needs analysis (Gregory et al., 2013) and a reflection exercise after the first game (Schippers et al., 2014). Neither this experimental intervention nor a balanced order variable for the two games had significant effects on any of the entrainment variables, so they were not included in these analyses. In this study, we examined entrainment during the first game only.

Measures

Acoustic-Prosodic Entrainment

Each participant was video- and audio-recorded using separate headsets and cameras. Focusing on features and dimensions found to be promising in prior computational linguistics research, we used the Praat software program (Boersma & van Heuven, 2001) to extract different types of convergence measures (e.g., Levitan, 2020), specifically the mean, maximum (max), and standard deviation (SD) of both pitch and intensity (e.g., pitch-SD, intensity-mean) for two time slices from the conversation from each individual (not across the teams). First, we calculated the feature values for individuals and for each time slice—in this case the first and last (fourth) quadrants of each of the first games (Rahimi & Litman, 2018). As noted, this decision to compare first and last quadrants was was based on our prior research demonstrating that convergence was strongest when feature value differences were compared between more distant temporal intervals (Litman et al., 2016). Then, we computed the difference in feature values between two individuals, and then the difference of those feature value differences for the two time slices (see Equation 1, Litman et al., 2016). Then, we applied a weighting scheme (detailed below) to calculate a weighted average of such pairwise/dyadic convergence, creating the final team convergence metric used in our prior research (Rahimi, 2019; Rahimi & Litman, 2018) as well as in this study.

For example, convergence on intensity maximum was calculated by (1) determining the maximum loudness of each speaker for the first and fourth time quadrants (see Figure 2 for an illustration of sampling periods)—the stars and dots show each individual’s acoustic-prosodic feature (intensity) maximum value in the first and fourth time quadrant, respectively; (2) creating a measure of the difference of maximum loudness between the two individuals in each dyad during the first and last quadrants separately; (3) subtracting the difference score between the dyads’ differences in the last period from the difference score between the speakers in the first period; and (4) taking a weighted average of those differences to more accurately measure trends in entrainment across the team. For intensity standard deviation (or SD), we examined each individual’s variation in loudness. For interpreting these measures, higher difference numbers indicate more convergence over the game session, and thus higher entrainment.

Figure 2.

Illustration of first and last quadrant sampling for convergence measures.

Previous work with this corpus suggests that there was significant entrainment within the first game on several features (Litman et al., 2016), but also that simple averaging across dyadic similarity measures would be insufficient to capture small team entrainment dynamics (Rahimi, 2019; Rahimi et al., 2019; Rahimi & Litman, 2018). Because different patterns were observed in our data when creating plots such as Figure 2 (which is illustrative), we developed the novel weighted averaging scheme used here, which was introduced and fully detailed in Rahimi and Litman (2018). Instead of simply averaging across dyadic similarities (i.e., average of person 1 to person 2, and person 2 to person 3, etc.) as in Litman et al. (2016), this weighting scheme takes into account small team dynamics by considering relative (converging or diverging) behavior of each speaker with respect to the common behavior in the team (but did not involve any adjustments for gender; see Rahimi, 2019). The weight for a speaker was the percentage of individuals who had the same team convergence behavior (either converging, diverging, or mixed) as the speaker, normalized by the group size. More specifically, if all conversational dyads that a speaker was involved in had positive or negative convergence values, the speaker would be assigned converging or diverging behavior, respectively. Otherwise, the speaker’s behavior was mixed. Thus, the current study uses the novel, more sophisticated versions of the acoustic-prosodic convergence entrainment measures.

We note that entrainment measure names such as entrainment on standard deviation (SD) of pitch, or entrainment pitch (SD), do not indicate the standard deviation across the team (unlike the personality measures, e.g., standard deviation of Extraversion). These entrainment measures are created using the standard deviation of an acoustic-prosodic feature within a time slice per individual, which is then combined to create an over-time convergence measure.

Team Personality Composition

All team members filled out individual difference questionnaires before training and before playing the game. This study took the two common ways of measuring team personality composition: means for levels or elevation and standard deviation for diversity (Halfhill et al., 2005). Measures of z-scored mean and standard deviation team composition were created for each of the Big Five dimensions for ease of comparison. We measured the Big Five using the 44-item Big Five Inventory measure (BFI, John et al., 1991).⁴ Participants rated the degree to which they agreed or disagreed with each item with the stem of “I am someone who. . .” on a 5-point Likert scale: 1 (disagree strongly), 2 (disagree a little), 3 (neither agree nor disagree), 4 (agree a little), or 5 (agree strongly). Each Big Five dimension has eight or nine items. For instance, for Extraversion (Cronbach’s α = .89), items include “Is full of energy,” “Is talkative,” and “Is reserved” (reverse scored). For Agreeableness (Cronbach’s α = .79), items include “Is helpful and unselfish with others,” “Is considerate and kind to almost anyone,” “Is sometimes rude to others” (reverse scored), and “Is cold and aloof” (reverse scored). Examples of Conscientiousness (Cronbach’s α = .79) include “Does a thorough job,” “Makes plans and follows through with them,” and “Tends to be lazy” (reverse scored). Items for Neuroticism (Cronbach’s α = .83) include “Worries a lot,” “Can be moody,” and “Is relaxed, handles stress well” (reverse scored). Examples of Openness to Experience (Cronbach’s α = .80) include “Is curious about many different things,” “Likes to reflect, play with ideas,” and “Has few artistic interests” (reverse scored).

Team Demographic Composition and Other Covariates

Prior studies using this corpus or other data suggested that demographic composition was significantly related to some entrainment variables (e.g., Levitan, 2020; M. Yu et al., 2019), so we examined the following team-level measures as potential covariates: percent female, standard deviation of age, ethnic diversity using Blau’s index, team size (three or four-person team), and the mean and standard deviation of each participant’s highest level of education (e.g., high school, some college, college, etc. on an ordered scale; Ouyang et al., 2017). Blau’s index is a commonly used metric for racial/ethnic diversity, where higher numbers indicate a greater variety of ethnic/racial categories represented more equally (Solanas et al., 2012). Average age and standard deviation of age within teams were highly correlated (r_s = .93), so only standard deviation of age was used. The standard deviation of education, or educational diversity, was included as a covariate because prior analyses with this corpus found that it was positively related to a task-specific measure of team performance (Ouyang et al., 2017). We also asked the participants to describe their cooperative board game experience, which could plausibly be a relevant covariate, on a 5-point frequency of play scale: 1 (never), 2 (play rarely, once or twice), 3 (play sometimes), 4 (play often), and 5 (play very often). These individual responses were combined to be mean and maximum team level of cooperative board game experience, and then those two variables were z-scored.

Analyses

All analyses were conducted using SPSS 28.0.0.0 at the team-level, because the covariates, entrainment, and personality composition variables were all at the team level. Given the novelty of the entrainment variables, we used correlations to determine if data reduction indices could be created (Table A1, Appendix A). Because many variables were not normally distributed, Spearman correlations were used. The set of entrainment variables were not highly correlated (all |r_s| <.30, see Table A1).⁵ While some of these correlations were statistically significant, they were too modest to suggest that the entrainment variables represented the same construct, thus not justifying combining entrainment variables or choosing just some entrainment variables.

First, we examined which covariates were significantly related to each entrainment variable using linear regression. All covariates were entered simultaneously, and non-significant covariates for each entrainment variable were eliminated using backward stepwise elimination, an exploratory, data-driven analytic approach. Second, we then conducted regression analyses for each of the six entrainment variables that included only the significant covariates (if any) and all 10 of the team personality variables, and we eliminated non-significant variables also using backward stepwise elimination. We present the final regression models with the retained variables. We could thus transparently explore which team personality and entrainment variables were significantly related while controlling for the other team personality variables and any significant covariates in the model.

Results

We first present the descriptive statistics for continuous variables (Table 2), and then present the results of our regression analyses.

Table 2.

Descriptive Statistics (N = 62 teams).

Variable	M	Median	SD	Min	Max
Percent Female (0–1.0)	0.63	0.66	0.25	0.00	1.00
Age Diversity (SD)	8.08	4.12	8.33	0.50	26.63
Ethnic/Racial Diversity	0.35	0.38	0.26	0.00	0.75
Education Average	5.29	5.00	1.37	3.25	9.00
Education Diversity (SD)	1.60	1.73	1.03	0.00	3.51
Play Cooperative Board Games (Team Average, z-scored)	0.00	−0.14	1.01	−2.17	3.12
Play Cooperative Board Game (Team Maximum, z-scored)	0.00	−0.14	1.01	−1.65	2.86
Agreeableness Avg.	3.92	3.95	0.36	3.15	4.52
Extraversion Avg.	3.30	3.31	0.40	2.25	4.21
Conscientiousness Avg.	3.84	3.87	0.32	3.03	4.44
Neuroticism Avg.	2.70	2.66	0.40	1.75	3.54
Openness Avg.	3.81	3.80	0.32	2.95	4.43
Agreeableness Diversity (SD)	0.53	0.48	0.26	0.11	1.34
Extraversion Diversity (SD)	0.81	0.84	0.36	0.07	1.70
Conscientiousness Diversity (SD)	0.54	0.49	0.25	0.06	0.97
Neuroticism Diversity (SD)	0.70	0.69	0.36	0.07	1.44
Openness Diversity (SD)	0.54	0.50	0.25	0.10	0.99
Agreeableness Avg., z-scored	0.01	0.11	1.01	−2.16	1.70
Extraversion Avg., z-scored	−0.02	0.00	0.99	−2.66	2.23
Conscientiousness Avg., z-scored	−0.02	0.08	0.99	−2.53	1.87
Neuroticism Avg., z-scored	0.03	−0.05	0.98	−2.28	2.08
Openness Avg. z-scored	−0.01	−0.04	1.01	−2.74	1.94
Agreeableness Diversity (SD), z-scored	0.02	−0.17	1.00	−1.57	3.14
Extraversion Diversity (SD), z-scored	0.00	0.09	1.01	2.08	2.50
Conscientiousness Diversity (SD), z-scored	0.00	−0.20	1.01	−1.93	1.71
Neuroticism Diversity (SD), z-scored	0.02	0.00	1.00	−1.73	2.10
Openness Diversity (SD), z-scored	0.03	−0.13	0.98	−1.67	1.76
Entrainment Pitch Mean	−1.93	−0.67	6.53	−30.29	22.70
Entrainment Intensity Mean	−0.05	−0.02	1.60	−5.24	4.46
Entrainment Pitch Maximum	7.61	1.25	30.39	−22.93	165.70
Entrainment Intensity Maximum	0.00	0.16	2.61	−8.81	7.83
Entrainment Pitch SD	0.86	0.19	5.17	−10.71	25.38
Entrainment Intensity SD	−0.02	−0.01	0.47	−1.40	1.21

Note. N = 62; The Big Five Inventory raw data were measured on a 1 to 5 scale.

No team covariate or personality composition measures were found to be significantly associated with entrainment on mean or maximum pitch. Team education diversity had a significant negative relationship with entrainment in terms of intensity standard deviation (unstandardized B = −0.13, SE = 0.06, lower bound = −0.24, upper bound = −0.01], p = .029). However, no personality composition variable was retained in the backward stepwise elimination for intensity standard deviation.

For convergence entrainment for standard deviation of pitch, team education diversity had a significant positive relationship as a covariate, so it was retained for the analyses with the personality composition variables. The final model after backward stepwise elimination, R²_adj = .23, F(3, 58) = 7.04, p < .001, had a positive association with the team education diversity (standard deviation of education level) and a negative one with team Openness to Experience diversity. The model also retained team Average Conscientiousness, although this variable was not statistically significant (Table 3).

Table 3.

Final Model for Pitch Standard Deviation (Entrainment) After Backward Stepwise Elimination.

Effect	Estimate^a	SE ^a	95% CI^a	p
Effect	Estimate^a	SE ^a	[LL, UL]	p
Intercept	−2.41	1.08	[−4.59, −0.24]	.03
Education Diversity (SD)	2.09	0.58	[0.94, 3.24]	<.001
Conscientiousness Avg.^b	1.07	0.60	[−0.13, 2.27]	.08
Openness Diversity (SD)^b	−1.56	0.59	[−2.75, −0.37]	.01

Unstandardized beta and standard error for unstandardized error.

z-scored.

The final covariate models for entrainment on intensity mean and intensity maximum both had a significant covariate, the age diversity of the team. The 10 personality composition variables, along with age diversity, were then analyzed for both dependent variables. Of these, for intensity mean, age diversity, team Average Agreeableness, and team Agreeableness diversity (SD) were all retained by the model, R²_adj = .23, F(3, 58) = 7.04, p < .001, with Agreeableness diversity and age diversity having significant (positive and negative, respectively) associations with entrainment on intensity mean (Table 4).

Table 4.

Final Model for Intensity Mean (Entrainment) after Backward Stepwise Elimination.

Effect	Estimate^a	SE ^a	95% CI^a	p
Effect	Estimate^a	SE ^a	[LL, UL]	p
Intercept	0.63	0.25	[0.13, 1.13]	.02
Age diversity (SD)	−0.09	0.02	[−0.13, −0.04]	<.001
Agreeableness Avg.^b	0.36	0.20	[−0.03, 0.75]	.07
Agreeableness Diversity (SD)^b	0.44	0.20	[0.05, 0.83]	.03

Unstandardized beta and standard error for unstandardized error.

z-scored.

Just as with the analyses of intensity mean entrainment (Table 4), we conducted a backward stepwise elimination regression with team age diversity as the initially retained covariate and the 10 personality composition variables as predictors for entrainment intensity maximum. In this case, both age diversity and team average Agreeableness were retained, with the latter having a positive association, final model R²_adj = .11, F(2, 59) = 4.84, p = .011 (Table 5).

Table 5.

Final Model for Intensity Maximum (Entrainment) after Backward Stepwise Elimination.

Effect	Estimate^a	SE^a	95% CI^a	p
Effect	Estimate^a	SE^a	[LL, UL]	p
Intercept	−0.63	0.44	[−1.51, 0.25]	.16
Age Diversity (SD)	0.08	0.04	[0.001, 0.15]	.047
Agreeableness Avg.^b	0.65	0.32	[0.02, 1.28]	.04

Unstandardized beta and standard error for unstandardized error.

z-scored.

Table 6 shows a summary of our findings. We found no personality composition associations with entrainment measured via pitch mean, pitch maximum, or intensity standard deviation, suggesting that different entrainment features may indicate different constructs. For entrainment measured via pitch standard deviation, we found a negative association with diversity in Openness to Experience when controlling for a positive relationship with team education diversity (and team average Conscientiousness). Further, Agreeableness diversity was positively related to entrainment on intensity mean, and team average Agreeableness was positively related entrainment on intensity maximum, both when controlling for team age diversity (but that covariate was significant in opposite directions). These findings warrant further discussion.

Table 6.

Summary of Findings: Relationship between Entrainment and Team Personality.

Team Personality	Findings
Agreeableness Avg.	+ with Entrainment Intensity Mean but n.s. + with Entrainment Intensity Maximum
Agreeableness Diversity (SD)	+ with Entrainment Intensity Mean
Extraversion Avg.
Extraversion Diversity (SD)
Conscientiousness Avg.	+ with Entrainment Pitch SD, but n.s.
Conscientiousness Diversity (SD)
Neuroticism Avg.
Neuroticism Diversity (SD)
Openness Avg.
Openness Diversity (SD)	− with Entrainment Pitch SD

Note. + indicates a positive relationship and – indicates a negative relationship.

Discussion

This paper makes a contribution by describing the concept and measurement of acoustic-prosodic entrainment in computational linguistics and then demonstrating how it can be used in a team research study. By conducting multidisciplinary team science, we identified a new problem space, that being the relationships between team personality composition and acoustic-prosodic speaking convergence. Our key empirical findings suggest that different kinds of team composition may be related to subsequent entrainment, but the relationship varies depending on the exact type of personality composition and entrainment. We found that average Agreeableness was positively related to entrainment on intensity maximum, and diversity in team Agreeableness was positively related to entrainment on intensity mean—both measures of convergence in terms of loudness. Diversity in team Openness was negatively related to convergence on pitch standard deviation. Previous research suggests that team personality compositions have different implications for team processes and outcomes. Our quantitative exploratory study (Bamberger & Ang, 2016) further implies that acoustic-prosodic convergence entrainment is similarly not a single phenomenon. As small group researchers, we can benefit from both the utility of entrainment as a possible variable of interest and by incorporating relevant constructs from a larger variety of disciplines. This study leveraged data and metrics generated within computational linguistics, bringing them to questions, theory, and disciplinary practices in social science (Allen et al., 2017). By mixing disciplines, we identified a space to sketch out knowledge gaps and theory.

Implications

Team researchers have long sought to better measure and understand team processes and emergent states (Kozlowski, 2015; Marks et al., 2001), and many argue that we should measure behavior directly (e.g., Baumeister et al., 2007). It is common in the history of science to develop and then widely use a measurement paradigm, but this tendency often results in the exclusion of other potential metrics, which in turn limits theorizing. One example is survey research on team processes such as conflict, where the conceptualization, study, and theorizing of conflict becomes that of a large or lingering memorable phenomenon rather than moment-by-moment conversational events (Paletz et al., 2011). While some might exclude moment-by-moment events from conflict definitionally, there is no theoretical reason to do so—it is due to limitations imposed by methodological norms. A theory of conflict could be more comprehensive by encompassing and distinguishing between both types of conflicts. Similarly, language use and conversations have long been recognized as key to small group processes, and language analysis is a promising method for understanding these processes (Kane & van Swol, 2022). Such studies can open the “black box” of team processes, so that these processes can be measured and conceptualized beyond what has traditionally been researched (Kane & van Swol, 2022; Lehmann-Willenbrock & Allen, 2018; Lehmann-Willenbrock & Hung, 2023).

Further research is necessary to understand whether entrainment is correlated with and/or reflective of other group processes and emergent states that are typically measured via surveys, such as cohesion. If different kinds of entrainment correlate highly with cohesion or other constructs (e.g., lexical entrainment with shared mental models), it may be possible to use automated measures of entrainment as a proxy for those other variables when questionnaires are impossible or difficult to obtain. Mixing social science methods and computational, sensor-based methods can be challenging given the different levels of granularity in terms of how to align on different and similar constructs (Lehmann-Willenbrock & Hung, 2023; Müller et al., 2019). Nevertheless, as we expand our measurement paradigms, we can also extend our theorizing about team processes.

Even if entrainment is not correlated with other constructs, it could still be studied as an important phenomenon in its own right, standing with participation and information sharing among other communication phenomena as a variable of interest for small group researchers. Both the concept and the measurement of acoustic-prosodic entrainment have been understudied within small group research but can be integrated into our existing models. For example, in a new integrative model of team effectiveness research that takes into account complexity and advocates for a “dynamic, multilevel” perspective, entrainment is not mentioned at all, nor is any other accommodation-like construct like synchrony (Mathieu et al., 2019). Perhaps entrainment serves as a mediator for the relationships between team personality and perceived social outcomes such as cohesion, but we will not know without theorizing, measurement, and empirical studies, all of which should lead to improved models that incorporate entrainment.

In terms of theoretical implications, our study raises the question of why certain measures of entrainment are related to personality composition while others are not. Just as creativity and emotion need to be measured and conceptualized in highly specific ways to truly understand how each construct works and the relationships between them (Ivcevic, 2022), so too does entrainment need to be unpacked and better conceptualized. We found three types of team personality composition (two regarding team Agreeableness) were related to three types of convergence entrainment. Of note, these six constructs were mostly distinct. Agreeableness diversity and team average Agreeableness were negatively related—modestly, but significantly (r_s = −.35, p < .01). Personality dimensions, while correlated modestly, are discrete constructs (John et al., 2008), and prior research suggests that entrainment metrics on different acoustic-prosodic dimensions and features may similarly be distinct, as they are not correlated with each other (Levitan, 2020; Schoenherr et al., 2019). Entrainment findings and correlates may vary depending on tasks, time slices, and types of features measured (Levitan, 2020). For instance, the maximum loudness (intensity) that people might express in different time slices, versus their average loudness, could be indicative of different social communication processes. Mean intensity could reflect general levels of excitement and engagement with the task and team, whereas maximum intensity could indicate intermittent levels of excitement.

To better understand the nature of different types of acoustic-prosodic entrainment, we may need to return to qualitative research. Researchers could watch and compare the interactions in videos of teams with high versus low entrainment (e.g., on pitch standard deviation) to understand what is happening in those conversations over time in terms of both what acoustic features can be perceived by an observer and the content of the communication. In addition, whether entrainment can be perceived by an observer has interesting implications as to whether entrainment must be perceived by the speakers themselves, or if it is an entirely unconscious team phenomenon.

One fruitful way to incorporate entrainment into broader theory is through associating it with other communication constructs such as participation. The relative participation of individuals with different personality patterns could explain a team’s overall entrainment. In other words, entrainment may be constrained or even enacted by participation dynamics within a group over time. Future work to develop a theory of team acoustic-prosodic entrainment should consider team dynamics that might result from not only team personality, but from the interaction between personality, the presumed motives involved in entrainment (e.g., affiliation, Giles et al., 1991), task demands, and settings, and how those factors impact both participation and entrainment. It is possible that the personality composition variables have an impact on different kinds of entrainment dimensions, such as lexical entrainment, or in different settings (Levitan, 2020).

While suggestions for practice are premature from our case study findings, we provide some potential practical insights for small group researchers in their study of entrainment. Most literature on entrainment has been dominated by examinations of lexical measures (see Van Swol & Kane, 2019 for a review), and studies of acoustic-prosodic entrainment have largely focused on dyadic entrainment or on individual tendencies to entrain (e.g., Gessinger et al., 2021; Lubold & Pon-Barry, 2014; Tomprou et al., 2021). Our study suggests the utility of expanding beyond lexical measures and dyads to acoustic-prosodic measures and teams. It is not without potential logistical or professional difficulties, however. Measuring entrainment using sensor and language data requires high-fidelity audio-recordings that can capture the different elements (e.g., specific acoustic-prosodic dimensions and features). Further, researchers must calculate the convergence of team members into one number for each meeting or conversation, or track moment-by-moment matching of changes if measuring synchrony. While our research team worked collaboratively and was able to communicate internally the benefits and limits of these entrainment metrics, situating our research for the psychology/organizational behavior audience has been difficult. In our experience with other multidisciplinary work, importing a research methodology to a new discipline can be more difficult than importing constructs and relevant research questions. Despite calls to be creative by mixing disciplines (e.g., with creativity research, Berg et al., 2023), combining research paradigms comes with the risk of being so counter-normative as to have difficulty finding a disciplinary home or at least difficulty in finding reviewers or audiences who can evaluate all the different aspects of a study.

Limitations and Future Research

Given the novelty of the convergence measures and the sample size of teams for the large number of potential variables, the case study findings are necessarily exploratory. In addition, backward stepwise regression is a data-driven exploratory technique, such that these findings should be replicated before being cited as established science. While considered a large sample for computational linguistics and computer science studies of group interactions (Allen et al., 2017), our sample was not large enough for more complex statistical analyses used in psychology and organizational behavior (e.g., structural equation modeling). These measures should be used in a larger dataset and these findings replicated with different tasks. Because team composition effects may change over time (Mathieu et al., 2014), additional research could examine longer-term teams, such as work teams with a variety of task types. Researchers could test more complex compilation models, which are only possible to examine with a huge variety of team personality combinations in large samples. For instance, the kind of team likely to have the highest entrainment in this task might have a combination of moderately high Agreeableness, high average Openness to Experience, low Extraversion diversity, and moderate-to-high Extraversion. With a larger dataset, we could also explore different combinations with different sub-dimensions of the Big Five (e.g., talkativeness, dominance). In addition, some effects may be curvilinear. Perhaps moderate average Extraversion, rather than high or low average Extraversion, would be predictive of greater team entrainment. This possibility may explain why our linear models did not find a significant association between Extraversion and entrainment. Also, weighting entrainment by team dynamics raises the question as to whether it would be useful to weight team personality composition by team dynamics, such as by participation. Future research could examine the utility and construct validity of weighting personality composition by particular individuals in teams, which to our knowledge has rarely been done in studies of team personality composition.

We also argue for additional research to conduct a fuller convergent-divergent correlational study of different entrainment dimensions, types, and features across many team settings and tasks, and with different team processes and emergent states. Those dimensions—be they pitch and intensity, gestural, lexical, phonetic, and so on—that correlate together could help us build theory around which types of communication are likely to be entrained on by which teams and on which tasks. These types of entrainment, measured in different ways, could also be associated with the typical self-report measures of team constructs—shared mental models, team cohesion, and so on—to determine which aspects of entrainment correlate with them, if any.

In addition, while our measures captured convergence, measures that are on a finer-grained time scale and multilevel could be fascinating to deploy and build theory around. The current convergence measures compare the first and last quadrants of conversations, developed by comparing different potential first and last segments. Future research can explore the optimal time slices and durations of conversations for detecting convergence and other types of entrainment. Do conversations need to be 15 min, 30 min, or an hour to detect convergence? Further, while quadrants were the best time slice for 30-minute conversations, would the first and last 5 min be sufficient for longer or shorter conversations? Future work could also examine multiple time slices rather than just two, such as a 5-min time slice compared to the next 5 min compared to the third set of 5 min, and so on, to calculate change over time in a different manner. For example, in a study of lexical entrainment, instead of comparing just two fixed intervals, we iterated over all two intervals of a given time-slice in chronological order and computed convergence based on properties such as maximum difference across all comparisons (M. Yu et al., 2019).

Finally, while convergence measures are not the same as synchrony measures, discovering the relationship between them could also be useful for conceptual clarity of entrainment. Theory on team dynamics suggests that dynamic processes may have feedback loops, and also that some variables may have asymmetric influence: for the latter, higher levels of one variable can increase a dependent variable, but lower levels of that variable do not result in the decrease of the dependent variable (Cronin & Vancouver, 2019). When examined at a finer-grained time scale, entrainment processes may be built via feedback loops with individual and team personality, participation, and silence.

Conclusion

This study used new advances in measures of team convergence entrainment to explore whether personality composition influences entrainment. By introducing these constructs and measurements to this audience, we offer an integrated, creative (Berg et al., 2023) perspective on team research. Before building theory, exploratory studies like this one are necessary to even realize we need to ask new questions. We recommend that the conceptual and measurement clarity of entrainment should be improved, and then small group researchers can associate those entrainment variables with other constructs of interest. Through iterative studies of quantitative and qualitative exploration, theory-building, and empirical testing, we can better understand the role of entrainment as a communication phenomenon within improved models of team processes.

Footnotes

Appendix

Table A1.

Spearman Correlations of All Variables Entered into the Models.

Variable	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21	22	23
1. Team size	—
2. Percent female	−.08	—
3. Age SD	.08	-.30*	—
4. Ethnic/ racial diversity	-.29*	.10	−.16	—
5. Education Avg.	.19	−.18	.65**	-.31*	—
6. Education SD	.10	−.18	.61**	−.10	.64**	—
7. Play Coop Board Games Avg. (Z)	−.07	.00	-.26*	.23	−.19	−.11	—
8. Play Coop Board Games Max (Z)	.20	.01	−.13	.06	−.01	.01	.66**	—
9. Agreeableness Avg. (Z)	.04	−.05	.10	−.01	.18	.04	.07	−.22	—
10. Agreeableness SD (Z)	.13	.03	.07	.10	.06	.05	.17	.39**	-.35**	—
11. Extraversion Avg. (Z)	.11	−.06	.20	−.11	.18	.21	.11	.17	.14	.06	—
12. Extraversion SD (Z)	.20	.02	.13	−.10	.11	.04	.11	.22	−.19	.21	−.24	—
13. Conscientiousness Avg. (Z)	.14	.05	−.21	−.15	.13	−.20	.14	.07	.15	.17	.04	.22	—
14. Conscientiousness SD (Z)	.10	.06	−.13	−.03	−.12	−.17	.04	.16	−.07	.02	−.03	−.07	-.33**	—
15. Neuroticism Avg. (Z)	−.16	.33**	−.09	−.09	−.20	−.13	−.15	−.12	-.36**	.06	-.27*	.05	-.26*	.08	—
16. Neuroticism SD (Z)	.15	.28*	.03	.11	.13	−.04	−.00	.09	−.07	.12	−.20	.09	−.06	.15	.29*	—
17. Openness Avg. (Z)	.02	−.15	.12	.14	−.04	.16	−.08	−.08	−.15	.14	.04	−.14	−.12	−.09	−.05	−.01	—
18. Openness SD (Z)	−.12	.02	.00	−.12	−.08	.01	−.01	−.11	−.03	−.20	−.25	−.02	.02	-.37**	.13	−.11	-.32*	—
19. Pitch Mean	.08	.10	−.19	.04	−.15	−.10	.02	−.01	.07	.13	.06	−.07	.23	−.09	.09	.10	−.00	−.02	—
20. Pitch Max	.11	.01	.05	−.01	.17	.13	.03	.10	.11	.05	.20	.03	.11	−.10	−.09	−.01	.01	−.14	.14	—
21. Pitch SD	.14	−.09	.13	−.07	.12	.30*	.22	.13	.22	.10	.19	−.06	.09	−.00	.06	−.01	.15	−.16	.29*	.27*	—
22. Intensity Mean	−.01	.17	-.44**	.01	-.38**	-.33**	.09	.06	.07	.21	−.15	.02	.24	.03	.01	−.09	.01	−.18	.25	.28*	.22	—
23. Intensity Max	−.06	-.32*	.15	.07	.18	.07	.03	−.14	.11	−.17	−.09	−.05	.06	−.09	0	−.19	.02	−.02	−.10	−.09	−.01	−.19	—
24. Intensity SD	−.08	−.07	−.08	−.05	−.18	−.19	−.07	−.06	−.06	−.16	-.30*	.15	.08	−.02	.07	−.13	.17	.08	−.13	−.01	−.16	.09	.18

Note. N = 62. Uncorrected p-values: *p < .05, ** p < .01. Correlations are rounded to two digits.

Acknowledgements

The authors are grateful to Stefani Allegretti and Caitlin Rice for their assistance in iterating the study materials, data collection, and data processing, as well as Mingzhi Yu for tabulating the ethnic diversity measure. We are also grateful for the reviewers and, especially, the editor for their helpful comments on earlier drafts of this paper. Some of this work was conducted when Susannah Paletz was at the University of Maryland Center for Advanced Study of Language. Zahra Rahimi was at the University of Pittsburgh. Susannah Paletz is currently also affiliated with the Social Data Science Center (University of Maryland) and the Applied Research Laboratory for Intelligence and Security (University of Maryland). Diane Litman is faculty at the Intelligent Systems Program and a Professor in the Department of Computer Science, as well. Some versions of the analyses were presented at the 2018 and 2022 Interdisciplinary Network for Group Research annual convention. Anonymized data and analysis scripts are available upon request.

Correction (June 2023): This article has been updated with the affiliation of Zahar Rahmini to University of Pittsburgh instead of Pandora/Sirius since its original publication.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by National Science Foundation (NSF) grants IIS-1420377 to the first author and 1420784 to the second author. This project was also supported by internal funds through the University of Maryland Applied Research Laboratory for Intelligence and Security and start-up faculty funds to the first author from the University of Maryland College of Information Studies. The opinions expressed are those of the authors and not necessarily those of the University of Maryland or the U. S. government.

ORCID iD

Susannah B. F. Paletz

Notes

Author Biographies

Susannah B. F. Paletz is an Associate Professor at the College of Information Studies, University of Maryland, College Park (PhD., University of California, Berkeley, Social/Personality Psychology, USA). Her research involves teams, culture, and creativity, and also applied psychology. She is the founding director of Organizational Teams and Technology Research Society.

Diane Litman, Professor in the School of Computing and Information at the University of Pittsburgh (USA) and senior scientist with the Learning Research and Development Center, completed her PhD at the University of Rochester (USA). Her research focuses on natural language processing, and applications such as education and law.

Valerie Karuzis is a Principal Faculty Research Specialist at the Applied Research Laboratory for Intelligence and Security, University of Maryland. She received her M.A. in Measurement, Statistics, and Evaluation from the University of Maryland and her B.A. in Neuroscience from Oberlin College.

Kelly M. Jones, PhD, is an Assistant Research Scientist in experimental and social psychology at the Applied Research Laboratory for Intelligence and Security, University of Maryland. Her research focuses on individual and social perceptions and behaviors, including cognitive biases and attitudes and behaviors regarding cultural and social movements.

Zahra Rahimi completed her PhD in Artificial Intelligence in the School of Computing and Information at the University of Pittsburgh (USA), where she focused on Natural Language Processing and Dialogue Systems.

References

Allen

J. A.

Fisher

Chetouani

Chiu

M. M.

Gunes

Mehu

Hung

(2017). Comparing social science and computer science workflow processes for studying group interactions. Small Group Research, 48(5), 568–690. https://doi.org/10.1177/1046496417721747

Ancona

D. G.

Chong

C.-L.

(1992). Timing is everything: Entrainment and performance in organization theory. Academy of Management Proceedings, 1992(1), 166–169. https://doi.org/10.5465/ambpp.1992.17515371

Ancona

D. G.

Chong

C. L.

(1996). Entrainment: Pace, cycle, and rhythm in organizational behavior. Research in Organizational Behavior, 18, 251–284.

Ancona

D. G.

Waller

M. J.

(2007). The dance of entrainment: Temporally navigating across multiple pacers. In Rubin

B. A.

(Ed.), Workplace temporalities: Research in the sociology of work (Vol. 17, pp. 115–146). Elsevier.

Anzalone

S. M.

Boucenna

Ivaldi

Chetouani

(2015). Evaluating the engagement with social robots. International Journal of Social Robotics, 7, 465–478. https://doi.org/10.1007/s12369-015-0298-7

Bamberger

Ang

(2016). The quantitative discovery: What is it and how to get it published. Academy of Management Discoveries, 2(1), 1–6. https://doi.org/10.5465/amd.2015.0060

Barrick

M. R.

Stewart

G. L.

Neubert

M. J.

Mount

M. K.

(1998). Relating member ability and personality to work-team processes and team effectiveness. Journal of Applied Psychology, 83(3), 377–391. https://doi.org/10.1037/0021-9010.83.3.377

Baumeister

R. F.

Vohs

K. D.

Funder

D. C.

(2007). Psychology as the science of self-reports and finger movements: Whatever happened to actual behavior? Perspectives on Psychological Science, 2(4), 396–403. https://doi.org/10.1111/j.1745-6916.2007.00051.x

Bell

S. T.

(2007). Deep-level composition variables as predictors of team performance: A meta-analysis. Journal of Applied Psychology, 92(3), 595–615. https://doi.org/10.1037/0021-9010.92.3.595

10.

Beňuš

. (2014). Social aspects of entrainment in spoken interaction. Cognitive Computation, 6, 802–813. https://doi.org/10.1007/s12559-014-9261-4

11.

Berdahl

J. L.

Martorana

(2006). Effects of power on emotion and expression during a controversial group discussion. European Journal of Social Psychology, 36(4), 497–509. https://doi.org/10.1002/ejsp.354

12.

Berg

J. M.

Duguid

M. M.

Goncalo

J. A.

Harrison

S. H.

Miron-Spektor

(2023). Escaping irony: Making research on creativity in organizations more creative. Organizational Behavior and Human Decision Processes, 175, 104235. https://doi.org/10.1016/j.obhdp.2023.104235

13.

Boersma

van Heuven

(2001). Speak and unSpeak with PRAAT. Glot International, 5(9/10), 341–347.

14.

Brandstetter

Beckner

Benitez

Bertneck

(2017). Persistent lexical entrainment in HRI. In Mutlu

Tscheligi

(Eds.) HRI ’17: Proceedings of the 2017 ACM/IEEE international conference on human-robot interaction (pp. 63–72). Association for Computing Machinery. http://doi.org/10.1145/2909824.3020257

15.

Chartrand

T. L.

Lakin

J. L.

(2013). The antecedents and consequences of human behavioral mimicry. Annual Review of Psychology, 64, 285–308. https://doi.org/10.1146/annurev-psych-113011-143754

16.

Cronin

Vancouver

J. B.

(2019). The only constant is change: Expanding theory by incorporating dynamic properties into one’s models. In Humphrey

S. E.

LeBreton

J. M.

(Eds.), The handbook of multilevel theory, measurement, and analysis (pp. 89–114). American Psychological Association.

17.

Danescu-Niculescu-Mizil

Lee

Pang

Kleinberg

(2012). Echoes of power: Language effects and power differences in social interaction [Conference session]. In Mille

Gandon

Misselis

(Eds.) WWW2012: Proceedings of the 21st international conference on the World Wide Web (pp. 699–708). Association for Computing Machinery. https://doi.org/10.1145/2187836.2187931

18.

Delaherche

Chetouani

Mahdhaoui

Saint-Georges

Viaux

Cohen

(2012). Interpersonal synchrony: A survey of evaluation methods across disciplines. IEEE Transactions on Affective Computing, 3(3), 349–365. https://doi.org/10.1109/t-affc.2012.12

19.

Doré

B. P.

Morris

R. R.

(2018). Linguistic synchrony predicts the immediate and lasting impact of text-based emotional support. Psychological Science, 29(10), 1716–1723. https://doi.org/10.1177/0956797618779971

20.

Ervin

Bonito

J. A.

Keyton

(2017). Convergence of intrapersonal and interpersonal processes across group meetings. Communication Monographs, 84(2), 200–2020. https://doi.org/10.1080/03637751.2016.1185136

21.

Eyben

Wöllmer

Schuller

(2010). Opensmile: the munich versatile and fast open-source audio feature extractor [Conference session]. In del Bimbo

Chang

S.-F.

(Eds.) Proceedings of the 18th ACM international conference on multimedia (pp. 1459–1462). Association for Computing Machinery. https://doi.org/10.1145/1873951.1874246

22.

Farrús

Hernando

Ejarque

(2007). Jitter and shimmer measurements for speaker recognition [Conference session]. In Van Compernolle

Boves

(Eds.) 8th Annual conference of the International Speech Communication Association (pp. 778–781). International Speech Communication Association. https://doi.org/10.21437/Interspeech.2007-147

23.

Gálvez

R. H.

Gauder

Luque

Gravano

(2020). A unifying framework for modeling acoustic/prosodic entrainment: definition and evaluation on two large corpora [Conference session]. In Pietquin

Muresan

Chen

Kennington

Vandyke

Dethlefs

Inoue

Ekstedt

Ultes

(Eds.) Proceedings of the SIGdial 2020 conference (pp. 215–224). Association for Computational Linguistics.

24.

Gálvez

R. H.

Gravano

Beňuš

Š.

Levitan

Trnka

Hirschberg

(2020). An empirical study of the effect of acoustic-prosodic entrainment on the perceived trustworthiness of conversational avatars. Speech Communication, 124, 46–67. https://doi.org/10.1016/j.specom.2020.07.007

25.

Gessinger

Möbius

Le Maguer

Raveh

Steiner

(2021). Phonetic accommodation in interaction with a virtual language learning tutor: A Wizard-of-Oz study. Journal of Phonetics, 86, 101029. https://doi.org/10.1016/j.wocn.2021.101029

26.

Giles

Coupland

(1991). Accommodation theory: Communication, context, and consequence. In Giles

Coupland

(Eds.), Contexts of accommodation: Developments in applied sociolinguistics (pp. 1–68). Cambridge University Press.

27.

Giles

Taylor

D. M.

Bourhis

(1973). Towards a theory of interpersonal accommodation through language: Some Canadian data. Language in Society, 2(2), 177–192. https://doi.org/10.1017/s0047404500000701

28.

Gonzales

A. L.

Hancock

J. T.

Pennebaker

J. W.

(2010). Language style matching as a predictor of social dynamics in small groups. Communication Research, 37(1), 3–19. https://doi.org/10.1177/0093650209351468

29.

Gregory

M. E.

Feitosa

Driskell

Salas

Vessey

W. B.

(2013). Designing, delivering, and evaluating team training in organizations: Principles that work. In Salas

Tannenbaum

Cohen

Latham

(Eds.), Developing and enhancing teamwork in organizations: Evidence-based best practices and guidelines (pp. 441–487). Jossey-Bass.

30.

Halfhill

Sundstrom

Lahner

Calderone

Nielsen

T. M.

(2005). Group personality composition and group effectiveness: An integrative review of empirical research. Small Group Research, 36(1), 83–105. https://doi.org/10.1177/1046496404268538

31.

Ilgen

D. R.

Hollenbeck

J. R.

Johnson

Jundt

(2005). Teams in organizations: From input-process-output models to IMOI models. Annual Review of Psychology, 56, 517–543. https://doi.org/10.1146/annurev.psych.56.091103.070250

32.

Ivcevic

(2022). Conceptual and measurement specificity are key: The case of creativity and emotions. Creativity Research Journal, 34(4), 391–400. https://doi.org/10.1080/10400419.2022.2122373

33.

John

O. P.

Donahue

E. M.

Kentle

R. L.

(1991). Big Five Inventory (BFI) [Database record]. APA PsycTests. https://doi.org/10.1037/t07550-000

34.

John

O. P.

Naumann

L. P.

Soto

C. J.

(2008). Paradigm shift to the integrative Big Five trait taxonomy: History, measurement, and conceptual issues. In John

O. P.

Robins

R. W.

Pervin

L. A.

(Eds.), Handbook of personality: Theory and research (pp. 114–158). Guilford Press.

35.

John

O. P.

Srivastava

(1999). The Big Five trait taxonomy: History, measurement, and theoretical perspectives. In Pervin

L. A.

John

O. P.

(Eds.), Handbook of personality: Theory and research (2nd ed., pp. 102–138). Guilford Press.

36.

Kämpf

M. S.

Liebermann

Kerschreiter

Krause

Nestler

Schmukle

S. C.

(2018). Disentangling the sources of mimicry: Social relations analyses of the link between mimicry and liking. Psychological Science, 29(1), 131–138. https://doi.org/10.1177/0956797617727121

37.

Kane

A. A.

van Swol

L. M.

(2022). Harnessing a language analysis perspective to uncover emergent group processes. In Dehghani

Boyd

R. L.

(Eds.), Handbook of language analysis in psychology (pp. 125–140). Guilford Press.

38.

Keltner

Gruenfeld

D. H.

Anderson

(2003). Power, approach, and inhibition. Psychological Review, 110(2), 265–284. https://doi.org/10.1037/0033-295x.110.2.265

39.

Kory-Westlund

J. M.

Breazeal

(2019). Exploring the effects of a social robot’s speech entrainment and backstory on young children’s emotion, rapport, relationship, and learning. Frontiers in Robotics and AI, 6, 54. https://doi.org/10.3389/frobt.2019.00054

40.

Kovacs

Kleinbaum

A. M.

(2020). Language-style similarity and social networks. Psychological Science, 31(2), 202–213. https://doi.org/10.1177/0956797619894557

41.

Kozlowski

S. W. J.

(2015). Advancing research on team process dynamics: Theoretical, methodological, and measurement considerations. Organizational Psychology Review, 5(4), 270–299. https://doi.org/10.1177/2041386614533586

42.

Langer

König

C. J.

Siegel

Fredenhagen

Schunck

A. G.

Hähne

Baur

(2022). Vocal-stress diary: A longitudinal investigation of the association of everyday work stressors and human voice features. Psychological Science, 33(7), 1027–1039. https://doi.org/10.1177/09567976211068110

43.

Lehmann-Willenbrock

Allen

J. A.

(2018). Modeling temporal interaction dynamics in organizational settings. Journal of Business and Psychology, 33, 325–344. https://doi.org/10.1007/s10869-017-9506-9

44.

Lehmann-Willenbrock

Chiu

M. M.

(2018). Igniting and resolving content disagreements during team interactions: A statistical discourse analysis of team dynamics at work. Journal of Organizational Behavior, 39(9), 1142–1162. https://doi.org/10.1002/job.2256

45.

Lehmann-Willenbrock

Hung

(2023, April 19–22). A multimodal social signal processing approach to team interactions [Paper presentation]. 2023 Society for Industrial-Organizational Psychology (SIOP) Annual Conference, Boston, MA.

46.

Lehmann-Willenbrock

Hung

Keyton

(2017). New frontiers in analyzing dynamic group interactions: Bridging social and computer science. Small Group Research, 48(5), 519–531. https://doi.org/10.1177/1046496417718941

47.

Levitan

(2020). Developing an integrated model of speech entrainment [Conference session]. In Bessiere

(Ed.) Proceedings of the twenty-ninth international joint conference on artificial intelligence (IJCAI-20) (pp. 5159–5163). International Joint Conferences on Artificial Intelligence.

48.

Levitan

Gravano

Willson

Beňuš

Hirschberg

Nenkova

(2012). Acoustic-prosodic entrainment and social behavior [Conference session]. In Fosler-Lussier

Riloff

Bangalore

(Eds.), Conference of the North American chapter of the Association for Computational Linguistics (ACL): Human language technologies (pp. 11–19). Association for Computational Linguistics.

49.

Levitan

Hirschberg

(2011). Measuring acoustic-prosodic entrainment with respect to multiple levels and dimensions [Conference session]. In Cosi

De Mori

(Eds.) Proceedings of Interspeech 2011 (pp. 3081–3084). International Speech Communication Association. https://doi.org/10.21437/Interspeech.2011-771

50.

Lewandowski

Jilka

(2019). Phonetic convergence, language talent, personality and attention. Frontiers in Communication, 4, 18. https://doi.org/10.3389/fcomm.2019.00018

51.

Litman

Paletz

S. B. F.

Rahimi

Allegretti

Rice

(2016). The Teams Corpus and entrainment in multi-party spoken dialogues [Conference session]. In Su

Duh

Carreras

(Eds.) Proceedings of the 2016 conference on empirical methods in natural language processing (pp. 1421–1431), Association for Computational Linguistics.

52.

Lubold

Pon-Barry

(2014). Acoustic-prosodic entrainment and rapport in collaborative learning dialogues [Conference session]. In Ochoa

Worsley

Chiluiza

Luz

(Eds.), MLA '14: Proceedings of the 2014 ACM workshop on multimodal learning analytics workshop and grand challenge (pp. 5–12). Association for Computing Machinery. https://doi.org/10.1145/2666633.2666635

53.

Lubold

Walker

Pon-Barry

(2021). Effects of adapting to user pitch on rapport perception, behavior, and state with a social robotic learning companion. User Modeling and User-Adapted Interaction, 31, 35–73. https://doi.org/10.1007/s11257-020-09267-3

54.

Lubold

Walker

Pon-Barry

Ogan

(2018). Automated pitch convergence improves learning in a social, teachable robot for middle school mathematics [Conference session]. In Rosé

C. P.

Martínez-Maldonado

Hoppe

H. U.

Luckin

Mavrikis

Porayska-Pomsta

McLaren

du Boulay

(Eds.) Lecture notes in computer science: Vol. 10947. Lecture notes in artificial intelligence: AIED: International conference on artificial intelligence in education (pp. 282–296). Springer. https://doi.org/10.1007/978-3-319-93843-1_21

55.

Luciano

M. M.

Mathieu

J. E.

Park

Tannenbaum

S. I.

(2018). A fitting approach to construct and measurement alignment: The role of big data in advancing dynamic theories. Organizational Research Methods, 21(3), 592–632. https://doi.org/10.1177/1094428117728372

56.

Marks

M. A.

Mathieu

J. E.

Zaccaro

S. J.

(2001). A temporally based framework and taxonomy of team processes. Academy of Management Review, 26(3), 356–376. https://doi.org/10.5465/amr.2001.4845785

57.

Mathieu

J. E.

Gallagher

P. T.

Domingo

M. A.

Klock

E. A.

(2019). Embracing complexity: Reviewing the past decade of team effectiveness research. Annual Review of Organizational Psychology and Organizational Behavior, 6, 17–46. https://doi.org/10.1146/annurev-orgpsych-012218-015106

58.

Mathieu

J. E.

Tannenbaum

S. I.

Donsbach

J. S.

Alliger

G. M.

(2014). A review and integration of team composition models: Moving toward a dynamic and temporal framework. Journal of Management, 40(1), 130–160. https://doi.org/10.1177/0149206313503014

59.

Michalsky

Schoormann

Niebuhr

(2018). Conversational quality is affected by and reflected in prosodic entrainment. In Klessa

Bachan

Wagner

Karpiński

Śledziński

(Eds.) Proceedings of speech prosody 2018 (pp. 389–392). International Speech Communication Association. https://doi.org/10.21437/SpeechProsody.2018-79

60.

Mohammed

Angell

L. C.

(2003). Personality heterogeneity in teams: Which differences make a difference for team performance? Small Group Research, 34(6), 651–677. https://doi.org/10.1177/1046496403257228

61.

Müller

Fàbregues

Guenther

E. A.

Romano

M. J.

(2019). Using sensors in organizational research–clarifying rationales and validation challenges for mixed methods. Frontiers in Psychology, 10, 1188. https://doi.org/10.3389/fpsyg.2019.01188

62.

Neuman

G. A.

Wagner

S. H.

Christiansen

N. D.

(1999). The relationship between work-team personality composition and the job performance of teams. Group & Organization Management, 24(1), 28–45. https://doi.org/10.1177/1059601199241003

63.

Ouyang

Woodworth

Paletz

S. B. F.

Litman

(2017, August 3–6). Will different degrees help you succeed? The role of educational diversity on team performance [Poster presentation]. American Psychological Association 2017 Convention, Washington, DC.

64.

Paletz

S. B. F.

Chan

Schunn

C. D.

(2017). The dynamics of micro-conflicts and uncertainty in successful and unsuccessful design teams. Design Studies, 50, 39–69. https://doi.org/10.1016/j.destud.2017.02.002

65.

Paletz

S. B. F.

Schunn

C. D.

Kim

K. H.

(2011). Intragroup conflict under the microscope: Micro-conflicts in naturalistic team discussions. Negotiation and Conflict Management Research, 4(4), 314–351. https://doi.org/10.1111/j.1750-4716.2011.00085.x

66.

Peeters

M. A. G.

van Tuijl

H. F. J. M.

Rutte

C. G.

Reymen

I. M. M. J

. (2006). Personality and team performance: A meta-analysis. European Journal of Personality, 20(5), 377–396. https://doi.org/10.1002/per.588

67.

Pennebaker

J. W.

Boyd

R. L.

Jordan

Blackburn

(2015). The development and psychometric properties of LIWC2015. University of Texas at Austin. http://hdl.handle.net/2152/31333

68.

Prewett

M. S.

Walvoord

A. A. G.

Stilson

F. R. B.

Rossi

M. E.

Brannick

M. T.

(2009). The team personality–Team Performance Relationship Revisited: The impact of criterion choice, pattern of workflow, and method of aggregation. Human Performance, 22(4), 273–296. https://doi.org/10.1080/08959280903120253

69.

Rahimi

Kumar

Litman

Paletz

(2017, August). Entrainment in multi-party spoken dialogues at multiple linguistic levels. In Lacerda

(Ed.), Proceedings of Interspeech 2017 (pp. 1696–1700). International Speech Communication Association. https://doi.org/10.21437/Interspeech.2017-1568

70.

Rahimi

(2019). Entrainment measures for multi-party spoken dialogues [Unpublished doctoral dissertation]. University of Pittsburgh.

71.

Rahimi

Litman

(2018). Weighting model based on group dynamics to measure convergence in multi-party dialogue. In Komatani

Litman

Papangelis

Cavedon

Nakano

(Eds.) Proceedings of the 19th annual SIGdial meeting on discourse and dialogue (pp. 385–390). Association for Computational Linguistics. https://doi.org/10.18653/v1/W18-5046

72.

Rahimi

Litman

(2020). Entrainment2vec: Embedding entrainment for multi-party dialogues. Proceedings of the AAAI Conference on Artificial Intelligence, 34, 8681–8688. https://doi.org/10.1609/aaai.v34i05.63935.

73.

Rahimi

Litman

Paletz

S. B. F.

(2019). Acoustic-prosodic entrainment in multi-party spoken dialogues: Does simple averaging extend existing pair measures properly? In Eskenazi

Devillers

Mariani

(Eds.), Lecture notes in electrical engineering: Vol. 510. Advanced social interaction with agents (pp. 169–177). Springer.

74.

Schippers

M. C.

Edmondson

A. C.

West

M. A.

(2014). Team reflexivity as an antidote to team information-processing failures. Small Group Research, 45(6), 731–769. https://doi.org/10.1177/1046496414553473

75.

Schoenherr

Paulick

Worrack

Strauss

B. M.

Rubel

J. A.

Schwartz

Deisenhofer

A.-K.

Lutz

Stangier

Altmann

(2019). Quantification of nonverbal synchrony using linear time series analysis methods: Lack of convergent validity and evidence for facets of synchrony. Behavior Research Methods, 51, 361–383. https://doi.org/10.3758/s13428-018-1139-z

76.

Sinha

Cassell

(2015). Fine-grained analyses of interpersonal processes and their effect on learning. In Conati

Heffernan

Mitrovic

Verdejo

M. F.

(Eds.), Lecture notes in computer science: Vol. 9112. Lecture notes in artificial intelligence: AIED: International conference on artificial intelligence in education (pp. 781–785). Springer. https://doi.org/10.1007/978-3-319-19773-9_115

77.

Solanas

Selvam

R. M.

Navarro

Leiva

(2012). Some common indices of group diversity: Upper boundaries. Psychological Reports: Human Resources & Marketing, 111(3), 777–796. https://doi.org/10.2466/01.09.21.pr0.111.6.777-796

78.

Tomprou

Kim

Y. J.

Chikersal

Woolley

A. W.

Dabbish

L. A.

(2021). Speaking out of turn: How video conferencing reduces vocal synchrony and collective intelligence. PLoS One, 16(3), e0247655. https://doi.org/10.1371/journal.pone.0247655

79.

Van Swol

L. M.

Kane

A. A

. (2019). Language and group processes: An integrative, interdisciplinary review. Small Group Research, 50(1), 3–38. https://doi.org/10.1177/1046496418785019

80.

A. C. L.

Abrego-Collier

Sonderegger

(2013). Phonetic imitation from an individual-difference perspective: Subjective attitude, personality and “Autistic” traits. PLoS One, 8(9), e74746. https://doi.org/10.1371/journal.pone.0074746

81.

Gilmartin

Litman

(2019). Identifying personality traits using overlap dynamics in multiparty dialogue. In Kubin

Kačič

(Eds.), Proceedings of Interspeech 2019 (pp. 1921–1925). International Speech Communication Association. https://doi.org/10.21437/Interspeech.2019-1886

82.

Litman

Paletz

(2019). Investigating the relationship between multi-party linguistic entrainment, team characteristics and the perception of team social outcomes [Conference session]. In Barták

Brawner

K. W.

(Eds.) Proceedings of the 32nd International Florida Artificial Intelligence Research Society (FLAIRS) Conference (pp. 227–232). Association for the Advancement of Artificial Intelligence. https://doi.org/10.48550/arXiv.1909.00867

83.

Zellmer-Bruhn

Waller

M. J.

Ancona

(2003). The effects of temporal entrainment on the ability of teams to change their routines. In Blount

(Ed.), Time in groups: Research on managing groups and teams (Vol. 6, pp. 135–158). Emerald Group Publishing.