Abstract
The study examined the relationship between language use and perception of group processes. In an experiment, participants discussed their views about climate change in a group chat. Afterward, participants (n = 239) filled out their perception of themselves and group processes. Participants who perceived more similarity among group members used less complex language (cognitive processes language) and more assenting language. As participants felt more knowledgeable and credible about the topic, their use of “we” pronouns and word count increased and use of “I” pronouns decreased. Replicating past research, participants with more extreme opinions used more “you” pronouns, and participants who reported engaging in more perspective-taking used more complex language and “we” pronouns. Results are integrated within an input–process–output model of group processes and suggest that language is reflective of individual inputs and perception of group processes.
Holtgraves and Kashima (2008) state that “language use frequently involves the recoding of implicit, nonlinguistic representations into explicit, linguistic ones” (p. 74). Group communication researchers are increasingly examining how language that individuals use in group discussions may reflect implicit individual states and perception of group processes. In a recent review of language in group discourse (Van Swol & Kane, 2019), granular level use of language, such as use of pronouns, related to group processes and individual differences such as status and conflict. In their proposed model of language use and groups, Van Swol and Kane (2019) identify how inputs to the group, such as member opinion or status differences, may be reflected in language use. Group inputs and interactions among members may then affect group processes and perception of the group, such as cohesion and conflict, which can also be reflected in language use. Language may reify initial individual inputs to the group or can shape perception within the context of the group. In this article, we focus on how language variables such as use of pronouns, language complexity, or assenting language may reflect initial inputs, like individual member opinion, and also reflect perception of the interaction and group members, like perception of similarity of members or taking the perspective of others.
Van Swol and Kane (2019) note that research on language and group processes is in the beginning stages, and there are many gaps in the literature. Research has often focused on a specific group processes such as cohesion and conflict, but not others, such as perspective-taking, or research has focused on certain inputs such as status or opinion, but less on perceived expertise (Van Swol & Kane, 2019). Research on language and group processes has not examined how language used in a group can also affect individual outputs, such as opinion change. In addition, given the emerging research in the field, many effects have yet to be replicated. This article aims to fill in some of these gaps by introducing new group processes and outputs, including similarity and perspective-taking into the study of language and group, and by trying to replicate previous research on extremity of opinion and language use. Below we detail our conceptual framework.
Conceptual Framework
Group Similarity
If a group member perceives other group members as similar to oneself, this can be reflected in the language used during interaction. Similarity can be based on having similar opinions, discussing similar information, or other factors such as interaction style (Good & Nelson, 1973); for example, groups that perceive they have more homogeneous personality traits have higher perceptions of similarity among members (Civettini, 2007). Groups with similar members tend to have a higher level of cohesiveness and attraction to the group and stronger group identity (Good & Nelson, 1973; Hogg et al., 1995; Hogg & Williams, 2000). For example, research has found that a perception of ingroup homogeneity is linked to stronger group identity (Brewer & Gardner, 1996; Simon & Pettigrew, 1990), and group members from groups with homogeneous members tend to retain a stronger identity to their group than more heterogeneous groups (Deffa, 2016).
In terms of language use, use of collective (e.g., we) versus singular pronouns (e.g., I) indicates differences in where a speaker is focusing attention and in a speaker’s psychological state (Brewer & Gardner, 1996; Chung & Pennebaker, 2007). Group members with a stronger sense of group identity and cohesion may have a more group focus rather than inward and self-focus, and therefore use more collective pronouns (Pennebaker, 2011). For example, research has found a positive relationship between use of “we” pronouns and perception of cohesiveness in a group (Lieberman et al., 2005; Matthews et al., 2015), although Gonzales et al. (2010) found that more cohesive groups used fewer “we” pronouns. Furthermore, newcomers can encounter more social acceptance in a group using collective pronouns because these pronouns serve a social integration function (Kane & Rink, 2015). Brewer and Gardner (1996) found that exposure to “we” pronouns increased perceptions of collective identification and facilitated judgments of similarity to ambiguous statements more than exposure to “they” or “it” pronouns. This research suggests there could be a relationship between use of “we” pronouns and perceptions of greater similarity. Participants who perceive themselves as similar to their group members will likely use more “we” pronouns during discussion, as people use more “we” pronouns with stronger group identity (Pennebaker, 2011; Van Swol & Kane, 2019). For example, research has found that use of “we” pronouns increases as people form a stronger group identity in online groups (Bäck et al., 2018). In essence, use of “we” pronouns signals proximity to interacting partners (Goffman, 1981). It is formally stated as follows:
Use of assenting language (e.g., agree, yes, OK) can indicate agreement and has been linked to less conflict and a positive group environment (Fischer et al., 2007; Huffaker et al., 2011; Sapru & Bourlard, 2013; Yilmaz & Peña, 2015). Huffaker et al. (2011) found that as negotiators communicating online converged on a similar solution, they used more assenting language. Huffaker et al. (2011) note that in online communication assenting language can replace nonverbals used to indicate agreement. Given the relationship between similarity and group attraction and cohesion, use of assenting language should be higher with a higher perception of group similarity (Good & Nelson, 1973; Hogg et al., 1995). In this regard, some researchers suggest that assenting language could indicate more groupthink-like processes and less critical analysis in the group (Yilmaz, 2016). For example, group members who do not delve into critical analysis of an issue may be less likely to understand its nuances, that is, recognizing fewer details that actually contradict their own view, and maintain a superficial consensus, which could increase perception of similarity (Baron, 2005). Thus, we hypothesize the following:
Conversely, critical analysis would lead to more nuanced understanding of an issue and the differences among members’ opinions. This could upset consensus, even superficial consensus, and at the individual level engender feelings of dissimilarity from one another. Thus, use of complex language and language implying cognitive processes could be related to reduced individual perception of group similarity. Within the language software program Linguistic Inquiry Word Count (LIWC; Pennebaker et al., 2015), complex thinking is measured through cognitive processes language, with language categories such as insight (e.g., think, know, consider), causation (e.g., because, effect, hence), discrepancy (e.g., should, would, could), tentative (e.g., maybe, perhaps, guess), certainty (e.g., always, never), inhibition (e.g., block, constrain), inclusive (e.g., with, and, include), and exclusive (e.g., but, except, without). Cognitive process language can indicate reflection, rational argumentation, analysis, and evaluation (Lin et al., 2016; Pennebaker et al., 2003; Tausczik & Pennebaker, 2010) and has been linked to integrative complexity of arguments (Van et al., 2016). For example, Gelfand et al. (2015) found that negotiators who used more facts, logic, and rational persuasion in their tactics used more cognitive processes language than negotiators who used more appeals to emotion. Thus, cognitive process language can indicate more complexity of thought, and participants who are thinking more complexly about an issue are engaging in less superficial consensus and may perceive more differences and nuances in others’ views. It is formally stated as follows:
Perception of Self-Credibility
People’s perception of their own competence and credibility about an issue is an input factor that is likely related to the language they use when discussing the issue in a group. Past research has found that participants who view themselves as more competent and knowledgeable often talk more in group discussion, which we operationalize as word count (Anderson et al., 2012; Bonito, 2006; Sniezek, 1992; Van Swol, 2009; Van Swol et al., 2016). Thus, we try to replicate and hypothesize the following:
Perception of similarity and perception of credibility are related. Often the perception that others agree with us or confirm our initial opinions (Chira et al., 2008; Sniezek & Buckley, 1995) or share similar information to ourselves (Wittenbaum et al., 1999) increases a sense of credibility and confidence. Given our hypothesized relationship between similarity and use of “we” pronouns due to greater group identity, perception of credibility could also be related to use of “we” pronouns.
In addition, researchers have found a strong link between behaviors related to confidence and credibility and status in groups (Anderson et al., 2012). Anderson et al. (2012) found that those who perceived themselves as more knowledgeable on a task managed to achieve more status in a group, often through higher participation rates. Those who perceived themselves as more knowledgeable in Anderson et al.’s studies were not actually more competent than their peers, but their pursuit of status was related to higher self-perceptions of competence and to behaviors in the group like higher participation rates that would increase perception of competence. Relatedly, whereas “we” pronouns are used more by members with higher status because they may attain status by focusing on the group rather than self (Van Swol & Kane, 2019), lower status group members use more “I” pronouns (Pennebaker, 2011; Van Swol & Kane, 2019) for multiple reasons. They may be more focused on themselves or trying to gain the attention of higher status members. Also, given the positive relationship between status and confidence (Anderson et al., 2012; Kennedy et al., 2013), they may refrain from representing the group due to their low confidence. For example, in online discussion boards, Dino et al. (2009) found that lower status members posted less and used more first-person singular pronouns than higher status members (for additional research support, see also Chung & Pennebaker, 2007; Kacewicz et al., 2014; Krifka et al., 2003; Reysen et al., 2010; Sakai & Carpenter, 2011; Scholand et al., 2010; Sexton & Helmreich, 2000). Thus, we hypothesize the following:
Extremity
One input into a group discussion is members’ differing opinions. Group members with extreme opinions on issues often think and behave differently than more moderate members, and this difference can be reflected in language use. People with extreme opinions rate their opinion on the far end points of a scale, rather than near the middle. Compared with people with more moderate views, people with extreme views often view themselves as more knowledgeable about the issue, view the issue as more important, have more commitment, and are less likely to change their opinion (Eagly & Chaiken, 1993; Ewing, 1942; Judd & Brauer, 1995; Krosnick et al., 1993; Osgood & Tannenbaum, 1955; Van Swol, 2009). Given their perception of their own knowledge and their commitment and confidence, it is not surprising that past research has found that extreme members talk more than non-extreme members (Van Swol, 2009; Van Swol et al., 2016, 2018). Thus, we seek to replicate this past research and hypothesize the following:
In addition, Van Swol et al. (2016) found that group members with extreme opinions use more “you” pronouns during discussion. In terms of attentional focus, use of “you” pronouns can reflect a speaker’s focus on others, especially a focus on others in comparison or relative to the speaker. Pennebaker (2011) notes that “you” pronouns occur more in interactions that are toxic and have more anger and conflict; he said “you” pronouns are “the equivalent of pointing your finger” (p. 175). Use of “you” pronouns can indicate distancing oneself from others, blaming, and differentiation (Kane & Rink, 2015; Simmons et al., 2008). Given extreme members’ perception of the superiority of their opinion and issue understanding, extreme members may engage in more accusation or more toxic conversations as they focus on others to explain their perceived superior opinion.
Although extreme members have more confidence entering a group discussion, perceive themselves as more credible and knowledgeable, and talk more during a group discussion (Van Swol, 2009; Van Swol et al., 2016), their behavior does not usually translate into status and influence in a group (Van Swol et al., 2018). For example, Van Swol et al. (2016) found that extreme members entered a group discussion with higher levels of confidence. However, while moderate group members became more confident after group discussion, extreme members actually became less confident. Van Swol et al. (2016) suggested that extreme opinions can be difficult to explain to more moderate members. In addition, use of “you” pronouns often reduces one’s influence in a group (Van Swol & Carlson, 2017; Van Swol et al., 2016). Thus, while extreme members enter the group discussion with higher perceived credibility and tend to talk more, their perception of credibility is based on their perception of the superiority of their views in comparison to less extreme members. Hence, their use of “you” pronouns might highlight the differences between self and others—“you” in comparison with me. This is different than status and its relationship to perceived credibility, and thus, we do not predict that extreme members will use more “we” pronouns. In conclusion, we seek to replicate Van Swol et al.’s (2016) finding that group members with more extreme opinions use more “you” pronouns:
Perspective-Taking
Perspective-taking is the tendency to try to understand others’ point of view and thinking (Davis, 1983). Perspective-taking may affect perception of a group discussion and be reflected in language use. When group members try to understand the opinions of other members and take their perspective, their focus of attention is going to be more outward than inward. Use of “we” pronouns reflects a more outward focus for a speaker (Pennebaker, 2011). For example, one reason that leaders in groups may use more “we” pronouns than lower status members is because a leader’s focus is more distributed across the group than concentrated on oneself (Van Swol & Kane, 2019). Kacewicz et al.’s (2014) research on status states that collective pronoun use is related to other-directedness. Thus, we hypothesize the following:
If a group member is engaging in perspective-taking, they may think more complexly about the issue due to more exposure to others’ viewpoints. Researchers have found a positive correlation between perspective-taking ability and cognitive complexity (Clark & Delia, 1977; Hale & Delia, 1976; Ku et al., 2015). Cognitive complexity can increase perspective-taking (Alcorn & Torney, 1982; Hale & Delia, 1976; Lutwak & Hennessy, 1982; Suedfeld et al., 1992), and also perspective-taking can increase cognitive complexity (Epley et al., 2004; Todd et al., 2012; Vescio et al., 2003). Thus, we seek to replicate this research and extend it by being the first to examine this robust finding by operationalizing cognitive complexity with cognitive process language use:
Polarization
An outcome of group discussion is that members may change their opinion. Individuals often polarize and become more extreme in their opinion after group discussion; this has been termed group polarization (Hinsz & Davis, 1984; Myers & Lamm, 1975). Individuals’ attitude polarization after group discussion is often accompanied by increases in confidence in one’s views (Heath & Gonzalez, 1995; Sunstein, 2009). In fact, increases in confidence during discussion created by organizing and elaborating on one’s rationales or by corroboration from fellow group members can drive polarization (Heath & Gonzalez, 1995; Sunstein, 2009). Confidence in discourse can be measured in certainty language which can indicate confidence, competence, and credibility (Corley & Wedeking, 2014). We seek to understand whether an outcome such as attitude polarization is reflected with increased use of certainty language, and our final hypothesis is as follows:
In this study, participants’ contributions to an online chat about climate change were analyzed using LIWC software program. After the chat, participants reported their perception of group processes and self-credibility, so language use in the discussion could be related to self-reported perceptions of self and the group discussion.
LIWC (Pennebaker et al., 2015) counts and then categorizes words into more than 80 categories and provides a measure of percentage of language use in each category for transcripts analyzed. A validated internal dictionary of words comprises each category. LIWC was rigorously developed in a process in which judges read more than 2,000 words or word stems and then judged their placement to specific categories. When a transcript is put through LIWC, every word in the transcript is compared with LIWC’s “dictionaries.” Then a percentage of the total words falling into each category is calculated. The content and construct validity of LIWC have been established (Francis & Pennebaker, 1992; Stirman & Pennebaker, 2001); interrater reliability in the discrimination of categories varies from 86% to 100%, supporting content validity. Although LIWC provides data on more than 80 different language categories, this study focuses only on pronoun use, cognitive processes language, word count, assenting language, and certainty language because there are theoretically derived hypotheses for these categories based on previous research.
Method
Participants
There were 239 undergraduate participants at a large, public Midwestern university in the United States. This exceeded the number of participants (n = 197) suggested to find a significant effect at p < .05 for a moderate effect size (.25) for a two-tailed t test. For participation, they received extra credit in a communication class. The study received Institutional Review Board approval from the first author’s university.
Procedure
Participants signed up for an experimental time to participate at a computer lab. Depending on attendance rates, there were four to nine people participating in the lab. Experimental times were staggered with participants starting the experiment every 20 min. There were dividers between computers to ensure privacy. Upon arrival for the experiment, participants were seated at a desktop computer and given a paper informed consent form to sign if they desired. First, their attitude about whether climate change is really happening was assessed (see measures). Then, participants were given nine pieces of information in randomized order. Information was manipulated in this study but not analyzed here and reported in Van Swol et al. (2019). This previous paper did not analyze content of participants’ chat responses. Six of the nine pieces supported the view that anthropogenic climate change is real, and three supported the view that anthropogenic climate change is not real. Information was obtained from procon.org (n.d.) and was pilot-tested to guarantee that information on either side of the debate was not significantly different in terms of valence and quality. An example of information arguing against the view that anthropogenic climate change is really happening is as follows: In testimony to the US Senate, William Gray, PhD, stated that the rising temperatures are caused primarily by water vapor, the most abundant greenhouse gas in the atmosphere, not by CO2. According to a 2010 study by researchers at NOAA, water vapor in the atmosphere was responsible for increasing the rate of warming during the 1990s by 30%.
An example of information arguing for anthropogenic climate change is as follows: According to the World Glacier Monitoring Service, since 1980, glaciers worldwide have lost nearly 40 feet (12 meters) in average thickness (measured in average mass balance in water equivalent).
After six pieces of the information, participants were given attention check questions (multiple choice or fill-in) to check that they read the information. If they failed the question, they were asked to reread the information.
After receiving the information, participants were told that they would discuss the issue with other participants in an online chat room. The chat room opened in a different internet browser window. Participants were told they had 12 to 15 min to discuss the issue of anthropogenic climate change with four other peers who were participating in the study. They were instructed that they could use the information they had read or any other information they knew about climate change and that they did not have to come to consensus. After clicking “enter the chat room,” the chat window opened and participants waited 25 s for it to “fill.” The chat continued for approximately 12 min and then closed automatically, and participants were instructed to complete a survey.
In reality, there were no other chat members and only the participant was actively participating in the chat; the chat was scripted and contributions of four bots to the chat were inserted at random intervals of 2 to 11 s. Participants’ typed contributions were included with these scripted chat contributions. Thus, the discussion appeared as a real, functioning chat. The scripted chat contributions were written from a real exchange among undergraduate research assistants about climate change, but then more information about climate change was included at fixed places. By keeping the chat the same for participants, we were better able to control variation in chat content, information, and opinions to which participants were exposed. See the appendix for examples of a chat.
Three of the scripted chat members voiced a view that anthropogenic climate change is real, and one expressed skepticism toward this view. This proportion reflects the sample population; in pilot testing (n = 17) and consultation with undergraduate research assistants, we identified that a strong majority of students had the opinion that anthropogenic climate change is really happening. The minority member in the chat mentioned and repeated more information supporting the view that anthropogenic climate change is not real (mention/repeat 75% contrary information), expressed skepticism toward information about anthropogenic climate change (e.g., “But some people were trying to say it is just [natural] changing ocean currents”), and stated that climate change is in fact not due to anthropogenic causes. Majority members, in contrast, did the same for information supporting the anthropogenic view of climate change and expressed this opinion. After the chat, participants completed a questionnaire online (see below). Then, participants were debriefed and thanked. The experimenters were asked to record participants doubting the legitimacy of the chat, but no participants expressed suspicion during or after the conclusion of the experiment. In addition, participants could write comments about the study in an open-ended portion of the questionnaire, but no participants expressed suspicion in their comments. However, five participants never contributed anything to the online chat, possibly due to suspicion, but their data were by default not analyzed.
Measures
Group similarity
Three items (e.g., My group is similar to me) were used to measure perception of the group’s similarity to the participant on a scale from 1 (strongly agree) to 5 (strongly disagree). The reliability was good (Cronbach’s α = .84), and the mean was used as a measure of group similarity.
Self-perception of credibility
Three items (e.g., I was competent at this task) measured self-perception of credibility on a scale from 1 (strongly agree) to 5 (strongly disagree). Reliability was good (Cronbach’s α = .77), and the mean was used as a measure of self-perception of credibility.
Attitude toward the climate change
Before and after the chat, participants rated whether they thought anthropogenic climate change was real on a scale from 1 (strongly disagree) to 7 (strongly agree) with the midpoint labeled as “neither agree nor disagree.” To measure polarization in their opinion after the chat, we measured how much participant’s opinion polarized in the direction of the scale end points. A positive score showed more opinion change in the direction of the scale end points, whereas a negative score showed more opinion change away from the end points of the scale and toward the midpoint. Seven participants rated themselves as neutral before the discussion, and polarization for these participants was measured in either direction. See Table 1 for distribution of opinions before and after the chat and for polarization frequencies. To measure extremity of opinion, participants who rated their opinion before the chat as either 1 (strongly disagree) or 7 (strongly agree) on the scale were coded as extreme (Van Swol et al., 2016). There were 98 participants who used extreme ends of the scale in the pre-chat opinion, and these participants would not be able to polarize any more in their opinion after the chat, as they had already used end of the scale (see Table 2).
Opinions on Climate Change Before and After Chat.
1 = not at all to 7 = extremely on “I believe that climate change is really happening.”
Amount of After-Task Opinion Polarization on Climate Change.
Note. The shift values from −2 to +2 indicate the degree of opinion change by subtracting the absolute value of the before-chat opinion minus 4 (midpoint) from the absolute value of the after-chat opinion minus 4 (midpoint).
Focus on others’ contributions
Two items (e.g., I tried to take into consideration all possible perspectives) on a 1 (strongly disagree) to 5 (strongly agree) scale were used to measure how much the participant focused on others’ information, that is, perspective-taking. With only two items, some researchers suggest that Cronbach’s alpha is inappropriate and meaningless (Eisinga et al., 2013; Sainfort & Booske, 2000; Verhoef, 2003) and suggest using Pearson’s correlation. Both items were significantly correlated (r = .40, p < .001) at an acceptable level.
Language measures
Transcripts of the participant’s contributions to the chat were analyzed with LIWC software (Pennebaker et al., 2015) to get measures of word count and percentage of language used in the LIWC categories analyzed. Only the participant’s contributions were analyzed. Percentage of words in each LIWC category is determined from total word count from that transcript. Therefore, if a participant uses five “you” pronouns and has a total word count of 200, LIWC will give a perception of 2.5% of use of “you” pronouns.
Results
When testing relationships among variables, correlations are used. Because some language variables are correlated (see Table 3), their relationship to group processes is tested separately. When testing differences between groups, t tests are used. Significance is tested at p < .05.
Means, Standard Deviations, and Correlations.
p < .05 level (two-tailed). **p < .01 level (two-tailed). ***p = .055.
Group Similarity
We hypothesized that use of “we” pronouns (H1: r = .015, p = .818) and assenting language (H2: r = .139, p = .036) would be positively related to perception of group similarity. Only assenting language was positively related to the perception of group similarity, in support of H2. There was no relation between use of “we” pronouns and perception of similarity, failing to support H1. H3 predicted that use of more complex language, operationalized as cognitive processes words (e.g., “cause,” “know,” “ought”), would be negatively related to perception of the group as similar to self (r = −.166, p = .012); this supported H3. See Table 3 for means, standard deviations, and correlations of all variables. See Table 5 for a summary of all hypotheses. Perception of similarity was not related to whether participants’ opinion was in the direction of the majority (r = .001, p = .989). Thus, perception of similarity does not seem based on sharing similar opinions with chat members. In addition, although not hypothesized, we found that use of negations (no, not, never) was negatively related to group similarity (r = −.142, p = .031).
Self-Perception of Credibility
H4 predicted that higher word count and H5 predicted that use of “we” pronouns would be positively related to self-perception of credibility, whereas H6 predicted that use of “I” pronouns would be negatively related to perception of being credible. Perception of self-credibility was positively related to word count, although the effect only approached significance (H4: r = .127, p = .055), and use of “we” pronouns (H5: r = .208, p = .002). Perception of credibility was negatively related to use of “I” pronouns (H6: r = −.149, p = .024). Participants’ perception of credibility and perception of similarity in the group were highly correlated (r = .27, p < .001), but while perception of similarity was not related to whether the participant’s opinion leaned toward the majority, perception of credibility was related to being in the majority opinion (r = .24, p < .001).
Extremity
We hypothesized that members with more extreme opinions would talk more (H7) and use more “you” pronouns (H8). We operationalized extremity by participants who rated their opinion about climate change using the end points (1 or 7) of the scale. Participants with extreme opinions (M = 97.16 words, SD = 53.93) did not have a significantly higher word count than participants who did not use extreme ends of the scale (M = 86.91 words, SD = 62.66), t(234) = 1.34, Cohen’s d = .158, p = .095 (one-tailed). H7 was not supported. However, in support of H8, extreme participants (M = 1.00%, SD = 1.80) used a higher percentage of “you” pronouns than non-extreme participants (M = 0.61%, SD = 0.95), t(234) = 2.15, Cohen’s d = .271, p = .016 (one-tailed).
In addition, in support of previous research, members who used extreme ends of the scale (M = 3.22, SD = 0.85) were more likely to perceive themselves as credible than more moderate members (M = 2.83, SD = 0.68), t(231) = 3.82, Cohen’s d = .507, p < .001. We examined whether there was a relationship between how extreme and non-extreme members perceived their credibility based on how much they talked. In a regression, word count, extremity (1 = yes, 0 = no), and their interaction were used to predict self-perception of credibility. See Table 4 for regression results. The model with the interaction was significant, R = .32, SE = .74, F(3, 226) = 8.60, p < .001. The interaction between extremity and word count was significant. To explore this interaction, we found that for non-extreme members, there was no relationship between word count and perceived credibility (r = −.037, p = .674), but the correlation was significant for extreme members (r = .299, p = .003). Thus, extreme members who spoke more perceived themselves as more credible.
Simple Linear Regression Analyses on Perception of Self-Credibility.
This model with the interaction effect added was significant, R = .32, SE = .74, F(3, 226) = 8.60, p < .001.
p < .01.
Perspective-Taking
We predicted that participants who reported engaging in more perspective-taking would use more “we” pronouns (H9) and would use more complex language (H10). There was a positive relationship between use of “we” pronouns and reported perspective-taking (r = .130, p = .048), in support of H9. There was a significant relationship between reported perspective-taking and use of more complex language (r = .134, p = .042), in support of H10. Perspective-taking was highly correlated to perception of credibility (r = .34, p < .001), suggesting that a third variable like intelligence might underlie perspective-taking results.
Polarization
We hypothesized (H11) that participants who polarized in their opinion would use more certainty language. There was a positive relationship between polarization and use of certainty language (r = .140, p = .034), in support of H11. In addition, although not hypothesized, we found a positive relationship between polarization and use of exclamation points (r = .132, p = .045) and use of negations (no, not, never) (r = .151, p = .021).
Discussion
Research and theory examining how granular-level language use in group discourse relates to individual inputs and perception of group processes is developing. This article added to the literature on group processes and language use by examining concepts such as perception of group similarity, self-credibility, and perspective-taking that have not been examined in relation to language use. In addition, the article partially replicated previous research (Van Swol et al., 2016) examining how extreme members use language in a group discussion differently than more moderate members. Results from all hypotheses are summarized in Table 5. Below we discuss these results in more detail.
Hypotheses Table.
Group Similarity
Perception of similarity to other group members can be reflected in language use. We found a positive relationship between perception of group similarity and use of assenting language. Although we cannot tease apart this relationship because we did not manipulate either variable, participants may use assenting language to show that they agree with the other group members or to validate information another member mentions (Wittenbaum et al., 1999), and hence the agreement may be driving both the feeling of group similarity and the use of assenting language. However, we found no relationship between perception of similarity and whether the participant’s opinion was leaning more toward the majority. Therefore, opinions may not underly feelings of similarity. Or, use of assenting language might engender feelings of similarity. Although not hypothesized, use of negations was negatively correlated to feelings of group similarity. Negations serve the opposite purpose of assenting language, so the negative relationship is not surprising, as participants who feel different than their group members may use more words such as no, not, or never.
In addition, we hypothesized that a group member using more cognitive processes language would perceive less similarity to other group members. We reasoned that participants thinking more complexly about the issue and using cognitive processes language would perceive more nuance in others’ thinking and recognize more differences from their own thoughts, which could reduce perceptions of similarity. Previous research has found that use of cognitive process language is related to having more integrative complex reasoning that emphasizes nuance, multiple perspectives, and differentiation (Van Swol & Carlson, 2017). Although our hypothesis was supported, our study did not explicitly test this reasoning. Future research could measure how use of cognitive process language affects group members’ perception of other members’ argument complexity or understanding of others’ positions and test whether this mediates the relationship between use of cognitive process language and perception of similarity in the group.
Finally, we hypothesized that use of “we” pronouns and perceptions of similarity would be related but did not find support for this. Previous research has found that “we” pronouns are used to integrate members within the group (Kane & Rink, 2015) and that “we” pronouns may signal a more collective focus, identity, and cohesion (Lieberman et al., 2005; Matthews et al., 2015). However, Gonzales et al. (2010) found that groups that had more cohesion used fewer “we” pronouns. We found no relationship. Pennebaker (2011) notes that “we” pronouns can both serve a collective function when they reference group members and also serve a distancing function when used as a royal “we” or a “we” and not you function. Thus, he notes that finding consistent relationships in use of “we” pronouns is difficult. In addition, groups in this study did not have to reach a collective goal or decision, which may have reduced a sense of collective identity.
Perception of Self-Credibility
Based on past research, we hypothesized that participants who viewed themselves as credible would talk more in the group. Our hypothesis was supported, replicating past research finding that people talk more when they perceive themselves as more credible (Bonito, 2006; Ridgeway & Correll, 2006; Sniezek, 1992; Van Swol, 2009), although, in the regression, the relationship between credibility and word count is driven by just extreme members. Only for extreme members did we find a relationship between word count and perception of credibility. In addition, previous research on status has found that people with perceived higher status use more “we” pronouns, and those with lower status use more “I” pronouns. We replicated these results with credibility as participants with higher perceived credibility used more “we” pronouns, and those with lower perceived credibility used more “I” pronouns. Again, we did not directly measure status, so whether those who felt more credible also felt higher status should be tested in future research. Given that those with higher perceived credibility also were more likely to be in the majority, it is possible that their greater use of “we” pronouns related to inclusion in the majority group more than a relationship with status.
Extremity
This study sought to replicate results from Van Swol et al. (2016) about the language that members with extreme opinions use in a group discussion in comparison to more moderate members. First, participants with extreme opinions did perceive themselves as more credible than other members, which is in line with previous research (Van Swol, 2009; Van Swol et al., 2016). We did not replicate the result that participants with more extreme opinions would have a higher word count than other participants, although the results were in the predicted direction. We did find, however, that for extreme members word count was positively related to their perceived self-credibility, but that this relationship was not significant for moderate participants. Extreme members with higher confidence and perception of their own credibility may be more likely than moderate members to act on their views by discussing them. Some research has found that people with extreme political views have an illusion of understanding (Fernbach et al., 2013) which causes unwarranted confidence, compared with moderate members. This may drive extreme members’ greater willingness to talk more when they have higher perceived credibility.
Finally, participants with more extreme opinions used a higher percentage of “you” pronouns than moderate participants, replicating previous research (Van Swol et al., 2016). Extreme members are likely to perceive their views as more superior, for example, due to an illusion of understanding (Fernbach et al., 2013), than more moderate members. This may lead to more use of “you” pronouns when extreme members, based on their perceived differences in views, differentiate themselves from other members (Pennebaker, 2011; Simmons et al., 2005). Moreover, extreme members’ more frequent use of “you” pronouns may signal a focus of attention on others in comparison with oneself as they focus on others to explain their superior views.
Perspective-Taking
Much previous research on perspective-taking has found that it is related to increased complexity of thinking. Research has found that complexity of thinking can increase perspective-taking, but it has also found that perspective-taking can increase complexity of thought (Alcorn & Torney, 1982; Clark & Delia, 1977; Hale & Delia, 1976; Ku et al., 2015; Lutwak & Hennessy, 1982; Suedfeld et al., 1992). Our study is the first to find the linguistic link between perspective-taking and cognitive complexity. Participants who stated they engaged in more perspective-taking during group discussion used more cognitive processes language. In addition, given the outward focus of attention with perspective-taking, we hypothesized that participants with higher perspective-taking would use more “we” pronouns; this was supported. Future research should explore the link between complex language and perspective-taking as they function in group discussion. Perspective-taking was highly correlated with how credible participants perceived themselves on the issue. This suggests that a third variable could be accounting for the results. Participants with higher fluid intelligence may perceive themselves as more credible on the issue of climate change, use more cognitive processes language, and take the perspective of others more frequently.
Polarization
Finally, because attitude polarization after group discussion is often accompanied by increases in confidence (Heath & Gonzalez, 1995; Sunstein, 2009), we hypothesized that participants who had polarized in their opinion after the online chat would use more certainty language; this was supported. In addition, although not hypothesized, participants who polarized in their opinion after the discussion used more exclamation marks during the discussion and more negations. Use of negations could indicate disagreement or entrenchment in one’s opinion and has been found to be an indicator of dogmatic thinking (Fast & Horvitz, 2016).
Implications
Pennebaker (2011) notes that small words like pronouns occur at the highest frequency in our discourse, often go unnoticed, yet provide a window into our focus of attention and psychological states. By carefully documenting relationships between word use and psychological processes, we develop unobtrusive tools to understand the dynamics in groups and psychological states of members. In essence, language can help us take the temperature of a group, understanding when group members may have more cohesion (e.g., assenting language) or more differentiation (e.g., “you” pronouns). Especially with online groups, where language can be easily monitored by algorithms in real time, monitoring of language may indicate when a group might need more help dealing with conflict or can help assess status relationships among members.
Van Swol and Kane (2019) created a model to describe how group inputs, group processes and emergent states, and group outcomes are exhibited in the language used during a group discussion. This study adds to their model. Similarity and perspective-taking can be emergent states in a group discussion that can both be influenced by the language used (Kane & Rink, 2015; Stout & Dasgupta, 2011) or reflected in language use. Language use then becomes a tool to understand these group processes. In addition, perception of credibility and extremity of opinion may be inputs that individuals bring to the group discussion that then influence language, such as pronoun use and amount of contributions as measured by word count. In essence, a group discussion can reify these individual inputs through language. Finally, group discussion has both group and individual outputs that result from the discussion and that are influenced by the language used during the discussion. This study examined an individual output of opinion polarization. Although much research has examined the topic of polarization in groups (e.g., Sunstein, 2009), little research has examined how language used is related to amount that individuals polarize after a discussion.
This study used an online chat. A large percentage of research on language use in groups has been conducted on online groups—mostly because of ease of access without the costs of transcription (Van Swol & Kane, 2019). However, there are differences between face-to-face and online group communication that bear mentioning. First, nonverbals are more available with face-to-face communication and often replaced with a substitute in online communication. Huffaker et al. (2011) note that nonverbals often provide backchannel communication that have to be reflected more with language in online communication. They note that “assents (e.g., ‘mmhmm’, ‘yes’, ‘right’) let a dialogue partner know that one is in agreement listening and attentive . . . This implies that backchannel cues such as assents can also be conveyed through online conversations and should have similar effects” (as nonverbal cues; p. 69). Walther (1992) theorized and received empirical support that, even at the relational or affective level, people adapt to their medium to utilize communication cue systems at their disposal (Walther et al., 2005). Hence, even lacking nonverbal cues, people adapt to the medium to form ties and develop impressions comparable with those of face-to-face communication. Thus, language may take on an additional significance in online chats, as it replaces nonverbal cues to understand emotions. As more and more groups interact online due to COVID-19, how language can be used to replace nonverbal cues, especially as it relates to group constructs like similarity, is important to research. Thus, assenting language or use of negations may be especially important toward taking the emotional temperature of online groups.
Finally, the online chat used in this study did not require participants to complete a task or come to consensus. While this may have reduced the sense that participants were a group with a common goal, previous research examining language use in groups has often studied online groups in which members engage with each other to post information and opinions and respond to each other, but are not completing a designated or explicit task or coming to agreement. These include fan forums, opinion forums, or hobby forums. For example, Huffaker (2010) examined language use in Google groups with topics such as politics, health, hobbies, and technology. Similarly, Burke et al. (2010) examined hobby groups, such as Grateful Dead fans and vegetarian cooking, to analyze language use in online groups. Dino et al. (2009) examined language use and status in fan forums. Results from these studies have been replicated in groups with an explicit goal. For example, findings from Dino et al. (2009) have been replicated with other types of groups (Kacewicz et al., 2014). Thus, although groups in our study did not have a goal to come to consensus, the results likely generalize to groups reaching consensus or having another common goal.
Limitations and Future Research
While adding to the literature on language use in groups, this study has several limitations. As mentioned above, groups were not completing a task, so whether results generalize to teams working on a common goal should be examined. The variables analyzed were not manipulated or controlled in this study, which excludes any causal claims about the use of language and group processes. In future research, either use of certain language features could be manipulated, possibly through confederates, to examine the effect of language on group processes, or group processes could be manipulated to examine changes in language used, for example, a researcher could ask, “Would encouraging more perspective-taking among group members increase the complexity of language used in group discourse?” Given the widespread use of naturally occurring online groups in the study of language and group processes, lack of designs to make causal claims is a widespread problem in this area of study. Finally, with extreme members, there was a ceiling effect on polarization, as extreme members, by definition, could not polarize more in their rated opinion after discussion. Given the limited research on language use and group processes (Van Swol & Kane, 2019), it is important to replicate these results just as this study replicated some effects from past research and observed established effects (i.e., perspective-taking) in a new domain of language use.
This study only examined perception of the group processes for a short online discussion. Van Swol and Kane (2019) note that the effects of language on group dynamics can be dynamic and longitudinal. A longitudinal design could follow discussion groups and their members over a longer period of time to understand processes, like how increases or decreases in status affect pronoun use. For example, Danescu-Niculescu-Mizil et al. (2012) followed online Wikipedia editing groups and examined what happened to the language of members when they were elected to leadership positions or lost status in the group.
Conclusion
In the group interaction model created by Van Swol and Kane (2019), members in a small group enter discussions with certain inputs—such as perceived self-credibility and differing opinions—that affect language use. As people participate in groups, processes emerge—such as feelings of group similarity—that are reflected in language use. Language serves the function to both reflect and help create these perceptions and processes in the group. This study highlighted the importance of language as a means to understand where group members are directing their attention, how group members perceive the group, and how the group discussion can reflect individual inputs brought the group. As groups continue to interact more online, language becomes especially important to take the emotional temperature of the group as nonverbals become less salient.
Footnotes
Appendix
In every chat, Participant_04 (italicized) was the human subject and the other four participants were bots. Throughout the group chat, Participant_01, Participant_02, and Participant_05 were in support of the opinion that global warming is really happening, whereas Participant_03 was not or ambivalent at best. Toward the end of the chat, Participant_02 tries to derive a consensus, asking if all members agree that global warming is really happening.
Acknowledgements
The authors thank Annie Hwang and Riley Roehl for help running this experiment and Peter Sengstock for programming the chat software.
Author’s Note
The research was approved by the University of Wisconsin–Madison Institutional review board (IRB) 2014-1370 and followed IRB guidelines for the treatment of human participants.
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 funded by research from the Hamel Family foundation and National Natural Science Foundation of China (71801120).
