Modeling Group Discourse with Epistemic Network Analysis: Unpacking Connections, Perspectives, and Individual Contributions

Promoting students’ competencies to work with others, while engaging with diverging views, is an essential academic and civic aim of education (Dishon, 2018; Newton & Zeidler, 2020; Slakmon & Schwarz, 2019). However, operationalizing and modeling group talk is a difficult feat due to the complexity of this process and the need to account for situated and implicit aspects of talk. One key challenge is analyzing not only specific communicative acts but also their dynamic interplay (Goldfarb Cohen et al., 2023; Csanadi et al., 2018; Hennessey et al., 2024; Lefstein et al., 2015). Accordingly, this paper sets out to offer an approach for operationalizing dialogue that includes engagement with disagreement using epistemic network analysis (ENA), focusing on the different patterns of connections between communicative acts, individual engagement with others’ perspectives, and the need to attend to the whole-group and the individual-group interplay of such analyses.

We start by introducing the theoretical background of dialogue in educational contexts in general, and Mercer’s (2007) Exploratory Talk framework more specifically. We then offer a brief survey of literature relevant to our research context—online synchronous small group work in general, and the use of ENA to study group dialogue, attending to how it could bolster our understanding of shared knowledge construction. Overall, our aim in this paper is to explore the various ways in which epistemic network analysis can be utilized to model different talk orientations based on the Exploratory Talk framework. Additionally, we aimed to investigate the role of ENA in helping researchers identify the interplay of discourse patterns at both the individual and group level. Finally, we demonstrated how ENA could be used to model engagement with others’ perspectives, enhancing theoretical conceptualizations of such engagement.

Participants (n = 60, 67% female, 33% male) were first-year students in an undergraduate Introduction to Psychology course in a large research university. Participants consisted of 12 groups of five students that completed all course tasks and participated in all three group tasks. The course included three synchronous 90-min collaborative online tasks in groups of five students (see Dishon et al., 2023 for more details on the course design). Group discourse was conducted in the chat application on the Moodle platform that accompanied the course. In these tasks, groups were asked to evaluate accounts of psychological evidence from popular sources and to formulate more scientifically rigorous ways to study these phenomena. In each task, students discussed a popular information source (a newspaper piece, a TED talk, and a podcast) about psychological research. The information sources were published a week before the task, and the questions on these sources were uploaded to the course’s Moodle at the beginning of the class time. Students then had 90 min to discuss the evidence and to upload a one-page-long collaborative response to the task at the end of the class.

In a previous paper (Dishon et al., 2023), we examined students’ evolving epistemological practices for evaluating psychological evidence in pre- and post-assessments. Here, our focus is on the dynamics of group talk during group tasks. Types of group dialogue were operationalized according to the Exploratory Talk framework (Mercer, 2007), which distinguishes between three prototypical talk orientations: Cumulative Talk, characterized by building positively yet uncritically on others’ ideas; Disputational Talk, which is rife with disagreement and lacking integration of interlocutors’ contributions; and Exploratory Talk, where students engage critically yet constructively with group members’ ideas. To this end, we transcribed and coded student discourse across all groups and tasks according to a set of basic communicative acts (agree, disagree, develop-self, develop-other, explain; see Table 1). These acts were chosen to reflect the basic tenets of the various dynamics of interaction explored in the Exploratory Talk framework. Critically, we intentionally focused on basic communicative acts, which we expected to be common across groups, striving to distinguish between groups according to the connections between codes (Bouton & Asterhan, 2023). All data was coded by two different coders (the second and third authors), and differences were resolved using social moderation (Herrenkohl & Cornelius, 2013). Then, using ENA, we identified different groups’ patterns of discourse on the basis of the co-occurrence of these codes at the group and individual level. The moving stanza window method was used to assess and analyze the coordination of individuals within their groups and the broader groups themselves, providing insights into the discourse connections (Siebert-Evenstone et al., 2017). During coding, we identified to which previous turn each utterance responded throughout our complete dataset. On the basis of this analysis, we chose the stanza window size of 7, which reflected the common distance between a given utterance and its earliest referent (Ruis et al., 2019). After constructing and analyzing the epistemic networks, we returned to examine these models in light of our qualitative data, which allowed us to better understand the interplay between epistemic networks and students’ actual discourse.

Our findings are structured according to our three research questions: (1) relying on the co-occurrence of the same set of talk moves to distinguish between different talk orientations; (2) refining the operationalization and conceptualization of Exploratory Talk by attending to the complexity of taking the perspectives of others, and distinguishing two variations of Exploratory Talk—self-oriented and other-oriented Exploratory Talk; (3) identifying different dynamics of interplay between individual contributions and group talk patterns, and their interplay with the above analyses.

To do so, we introduce the epistemic networks of four different groups. These groups were purposefully chosen as they are particularly salient examples of the four types of talk orientation we identified in our data (as detailed in findings 1 and 2). In this respect, these groups are not necessarily representative of the class as a whole but rather serve as polar cases (Yin, 2014). Hence, though these group dynamics are not generalizable, they allow us to bring into focus theoretical relationships that may apply to other contexts (Eisenhart, 2009).

To illustrate the differences between the various talk orientations, we complement the analysis of their epistemic networks with qualitative examples from the various groups’ work. To make this easier to follow, we have opted to offer excerpts from the same episode across the different groups. This episode took place during the third and last group task. In this assignment, participants were tasked with listening to a podcast (Spiegel & Miller, 2017) discussing various approaches to psychotherapy: psychodynamic, cognitive behavioral therapy (CBT), and mindfulness. Their objective was to evaluate a claim made by one of the podcast’s interviewees, who argued that psychodynamic therapy is “deeper” than the other two approaches. This task was designed to elicit engagement around a poorly defined concept, with the aim of emphasizing the necessity of establishing precise operational definitions for psychological concepts. As we will demonstrate, this prompted participants to grapple with the interpretation of “depth” and how it could be empirically investigated. The ensuing excerpts showcase the efforts of each group to reach a consensus on the definition of depth and how they reflected the diverging dynamics of group interaction.

Operationalizing and evaluating group talk is a complex yet vital endeavor (Hennessy et al., 2024; Mercer et al., 2019; Swiecki et al., 2020). This study joins existing work that uses ENA to model group discourse (e.g., Bressler et al., 2019; Csanadi et al., 2018; Nguyen, 2022; Vandenberg et al., 2021), while offering three distinct contributions: (1) a novel approach to modeling different talk orientations (Cumulative, Disputational, Exploratory) by relying on ENA to identify the frequency of co-occurrence of the same small set of basic communicative acts; (2) refining the theoretical conceptualization of Exploratory Talk by attending individuals tendency to engage with the perspective of others, suggesting two sub-variations (self-oriented and other-oriented Exploratory Talk); (3) using ENA to model the interplay between individual contributions and group discourse.

With respect to the three talk orientations, we illustrated that Cumulative Talk could be characterized by frequent connections between agreements and developing one’s and others’ ideas, while Disputational Talk is reflected in frequent connections between disagreeing and explaining (rather than developing) one’s ideas. Further, in its ideal version, Exploratory Talk implies disagreements among group members that co-occur with developing others’ ideas. However, the rarity of such instances in our own data, supported by the complexity of high levels of perspective taking (Goldfarb Cohen et al., 2023; Gehlbach & Mu, 2023), stresses the importance of attending to another variation of Exploratory Talk in which disagreements are used to develop one’s own thinking rather than others’.

These two models represent different levels of shared meaning making in Exploratory Talk, distinguishing between self-oriented Exploratory Talk and other-oriented Exploratory Talk. The former describes a more basic level in which disagreement led individuals to refine and reconsider their own position, while the latter is a more developed case in which individuals also develop others’ ideas. Critically, self-oriented Exploratory Talk might serve as a stepping stone towards the more complex process of developing others’ ideas in light of disagreements. Yet, our argument is that this kind of talk is invaluable in itself, as it could be the case that even when students struggle to develop their interlocutors’ ideas, they are able to enrich each other’s thinking in dialogue by exploring different ways through which they could invite their peers to view the matter at hand from different perspectives, all of which are intended to support their original line of argumentation (Goldfarb Cohen et al., 2023). Put differently, one key aspect of both self and other-oriented Exploratory Talk is the ability to engage in the process of perspective taking, either by explaining one’s own perspective or by developing others’ perspectives, by challenging or building on their ideas.

Lastly, we argue that ENA could operationalize individual contribution to group dialogue. This is based on the assumption that individual engagement leads to changes in collective engagement, and conversely, changes in collective engagement shape individual engagement (Mercer, 2008; Ryu & Lombardi, 2015; Swiecki et al., 2020). Using ENA, we were able to identify how individuals’ patterns could explain the emergence of Exploratory Talk, or lack thereof. First, with respect to the two groups that engaged in Exploratory Talk, we illustrated how Exploratory Talk emerged in light of two dominant group members, an “instigator,” who engaged mainly in Disputational Talk—challenging other group members’ ideas and requiring them to hone their ideas—and a “connector,” whose contributions were more cumulative in nature—clarifying and developing her own and other’s ideas. Yet, the combination of these two networks served to support the emergence of Exploratory Talk at the group level. Comparing this to a group that engaged in Cumulative Talk, we can see that the main distinction was that we could not find such contrasting dynamics in which a group member was disputational enough to elicit engagement with disagreement at the group level that was concurrently connected to the development of ideas (either one’s own or others’).

To briefly summarize, this study offers three key findings. First, we found that ENA could model engagement with disagreement by identifying diverging patterns of co-occurrence of communicative acts—Cumulative Talk was associated with connections between agreements and developing ideas, whereas Disputational Talk exhibited more frequent connections between disagreements and explanations. Second, ENA could be used to refine theoretical conceptualizations of engagement with disagreement. Specifically, we distinguished between self-oriented Exploratory Talk, in which disagreement led individuals to refine and reconsider their own position, and other-oriented Exploratory Talk, which includes more in-depth engagement with the ideas of others. Third, ENA allows examining the interplay between individual and group networks: we found that groups engaging in Exploratory Talk included an “instigator”—an individual that challenged others without dominating the group’s work—and a “connector”—a group member that clarified and developed ideas, thus allowing disagreement to be connected to the development of new ideas.

Our small data set and theoretically informed choice of groups imply that this is intended to be an initial “proof of concept” for the use of ENA to explore this interplay, which could serve as a basis for future inquiries into different types of group talk, how they can be operationalized via ENA or other methods, and how they relate to individual contributions. In future research, it may be insightful to explore whether these observed patterns persist when applying ordered network analysis (ONA). ONA is a method that models the sequential aspect of collaborative problem-solving, by attending not only to connections between utterances but also to their order (Tan et al., 2022). With respect to our study, employing ONA could reveal distinct trajectories through which the sequence of certain communicative acts shapes group discourse.

While these suggest that ENA could model engagement with disagreement in dialogue, we wish to conclude by highlighting a key challenge in the process of developing the use of ENA to larger bodies of data. Namely, attending to the different patterns of relationships we described above requires coding and modeling the overall patterns or groupwork, the interplay between various levels of epistemic networks (individual and group), and closing the interpretive loop by examining both of these with respect to qualitative data. While we hope that this study could serve as a first step towards more robust comparisons between groups across large data sets, we highlight the importance of such time intensive and exploratory investigations as an initial step prior to their application to larger data sets. Specifically, the above analyses require attention to, and careful coordination and interpretation of, three simultaneous levels of connections: between the data and qualitative coding, between the codes themselves in each network, and between individual and group level networks. As such, they could be theoretically rich, but are in danger of being particularly context-specific, and more easily shaped by idiosyncratic characteristics of a given task, setting, or individuals. Thus, ENA’s capacity to facilitate both in-depth quantitative ethnographies and large-scale comparisons that rely, at least partially, on automatic coding and analysis demands particular vigilance with respect to the contextual factors shaping epistemic networks and careful interpretive alignment.