Studying Discourse as Social Interaction: The Potential of Social Network Analysis for Discourse Studies

Abstract

Education researchers have extensively studied classroom discourse as a way to understand classroom structures and learning. This article proposes the use of social network analysis (SNA) as a method for discourse studies in education. SNA enables us to learn about the connections between persons and the patterns of relations within groups. This presents a novel approach to the study of discourse that may more accurately reflect current understandings of discourse as a social phenomenon. This article explains the theoretical links between SNA and the concept of discourse in education and then considers how SNA can be used to examine classroom discourse. A brief overview of promising methods is presented to provide examples of how SNA can be applied to discourse data. This article argues that continued exploration and applications of SNA could yield more complex understandings of the role of discourse in learning opportunities and outcomes.

Keywords

classroom research discourse analysis discourse processes peer interaction/friendship statistics

Education researchers have extensively studied classroom discourse as a way to understand classroom structures and learning (e.g., Cazden, 2001; Heath, 1983; Resnick, Asterhan, & Clarke, 2015). Approaches to measuring discourse have typically focused on words, utterances, and other measures of talk that can be assigned to individuals. Although such measures of talk can represent important dimensions of discourse, these measures treat discourse as an attribute of a single person. This view does not align with conceptual understandings of discourse as a social phenomenon in which talk is interactionally constructed and negotiated for particular purposes (Bakhtin, 1981, 1984; Gee, 2012; Goffman, 1981). This raises issues of validity in discourse research relying on attribute-based measures.

Social network analysis (SNA) is concerned with the connections between persons and the patterns of relations connecting persons and groups (Knoke & Yang, 2008; Scott, 2013; Wasserman & Faust, 1994). Because discourse entails social interaction, it can be conceptualized as social ties that can be mapped as a social network. We propose the use of SNA as a methodological approach that is theoretically consistent with understandings of discourse as social interaction. This article attends to key questions about the application of SNA to the study of discourse, including what methods of SNA are available, well suited, and feasible for the study of discourse in classrooms and other learning spaces. Additionally, we briefly describe a study conducted by the first author to illustrate the utility of SNA for research on discourse.

Conceptualizing Discourse as a Network

Social theories of discourse describe how language and speech are embedded in interactions between people (Bakhtin, 1981, 1984; Gee, 2012; Street, 1984). Participants in talk jointly construct and enforce social norms that tacitly define the expectations of the interaction, including what content and knowledge merits transmission, who is entitled to transmit it, and the modes through which transmissions may occur (Bourdieu, 1991; Bourdieu & Passeron, 1990; Goffman, 1959). This joint construction of the context and meaning of talk situates discourse as a distinctly dialogic process defined by the relationships and interactions between participants (Bakhtin, 1981, 1984; Cazden, 2001; Gee, 2012). By foregrounding these social interactions, social theories of discourse draw attention to the centrality of human interaction for talk and learning.

This emphasis on the speech relations connecting persons and their role in understanding teaching and learning suggests a broad philosophical alignment with SNA. SNA rejects “atomistic views of social and education phenomena” to instead preserve the “interconnected nature” of human interactions (de Lima, 2010, p. 243). Fundamental to a social network approach is the view that a phenomenon cannot be understood if it is segmented or isolated from social relations (Kadushin, 2012; Knoke & Yang, 2008). To attend to these social relations, connections between persons, rather than the persons themselves, become the unit of study. The attention of the researcher is thereby shifted from the person who acts to the nature and direction of the action.

This shift toward the interaction as the unit of study requires the use of relational data. In discourse studies, relational data includes the speaker, audience, and content and nature of the speech act. This focus on each interaction between speaker and audience mirrors what Bakhtin (1986) described as the “link in the chain of speech communication” (p. 93). Treating discourse as a series of interactions that can be mapped allows for the construction of networks, or visual representations of the interactions or links between persons (Scott, 2013; Wasserman & Faust, 1994). We use the term discourse networks to describe networks that map the talk of two or more persons. We additionally use the term dialogic ties to describe the links or interactions within a discourse network, where each dialogic tie is a speech utterance that connects two or more persons.

Conceptualizing and studying discourse networks shifts the focus of analysis from the unit of speech itself to the discursive interactions between persons. Patterns observed in relational data can then be analyzed to understand the interactional nature of discourse structures in schools, homes, and other learning contexts (Carolan, 2014; Goffman, 1959; Heller, 2003; Kadushin, 2012). This may include explorations of topics such as students’ and teachers’ roles in the classroom; the structures of group talk; correlations between speech patterns and individual attributes such as gender, language, or other characteristics; inclusion and exclusion in learning activities; and whether classrooms provide equitable opportunities for talk and learning. The example study we discuss later uses the concept of a discourse network to examine the positioning of an emergent bilingual student during a small group discussion.

Research on Discourse Networks

Despite strong connections between social network and discourse theories, SNA has had limited applications in research on discourse. Early investigations of discourse using SNA attempted to identify communication patterns that would optimize group performance (Bales, Strodtbeck, Mills, & Roseborough, 1951; Bavelas, 1950; Leavitt, 1951). These researchers conducted laboratory experiments in which they could restrict and manipulate the conditions of communication between participants. For example, Bavelas (1950) and Leavitt (1951) controlled who could send messages to whom within small groups to compare the functioning of centralized versus decentralized networks. Results of these experiments generally showed that centralization, or making a single person the hub of communication, improved performance on simple tasks but was detrimental for complex tasks. Though these experiments enabled researchers to test hypotheses about the dynamics of small groups in controlled contexts, the use of laboratory settings proved limiting for understanding talk and learning in more natural settings, including complex classroom interactions between students and teachers.

This line of research largely lost traction by the 1960s, partly due to the difficulties of collecting and analyzing speech data in naturalistic settings. These difficulties were exacerbated by a lack of portable and convenient methods for the collection, storage, and examination of audio and video recordings from the field. Though such technological challenges were reduced with the increased availability of video and audio recording devices, there remain challenges in collecting speech data in naturalistic settings. For example, it is often difficult to accurately hear all students in video recordings of classrooms and identify to whom particular utterances are being directed. More recent research has largely looked to bypass these challenges by focusing on the communications of individuals in online or digital spaces (Chai & Tan, 2009; de Laat, Lally, Lipponen, & Simons, 2007a; Oshima, Oshima, & Matsuzawa, 2012; Sharma & Tietjen, 2016; Shea et al., 2009; Thorpe, McCormick, Kubiak, & Carmichael, 2007). These studies consider discourse in digital forums, online learning environments, blogs, social networking sites, instant messaging, and other forms of e-communication where the discourse is discrete and written. This allows the researcher to rely on the structured nature of talk in these spaces to identify the speaker and recipient of each message. These studies have used SNA to answer questions about group and individual learning processes, including factors affecting participatory patterns, changes in discourse patterns over time, collaborative knowledge construction, and group problem solving.

New technologies and digital tools continue to make the collection and analysis of talk data from naturalistic settings more manageable. For instance, Yoon (2011) used handheld devices to track middle school students’ interactions during a debate on the ethics of genetic engineering. These handheld devices were used to record who each student spoke to during the discussion and their views in the debate. These data tracked both the order and directionality of students’ talk and students’ evolving views on genetic engineering. Visualizations of these data were then shown to students. Yoon considered both the students’ interactions during the debate and the ways in which viewing these visualizations impacted the students’ interactions during future discussions. Tools such as the handheld devices used by Yoon make the application of SNA to in-person talk in classrooms and other learning contexts more accessible and feasible for a broader range of education researchers. Despite the existence of tools such as these for collecting and reporting SNA data, there are currently few published applications of SNA to the study of in-person discourse networks.

In one such study, Ryu and Lombardi (2015) used SNA to map elementary students’ engagement in scientific discourse. The researchers mapped various types of “argumentative interaction” (p. 78) to explore individual and collective engagement in scientific argumentation, including how interactions and participation changed over time. Ryu and Lombardi also used SNA to examine how the teacher’s role during argumentation activities shifted during the school year as she became a less dominant participant. In another study, Mameli, Mazzoni, and Molinari (2015) examined discourse between students and their teacher in five Grade 3 classrooms totaling 106 students and 10 teachers. The authors used various quantitative measures (e.g., nodal degree, operationalized in their work as each individual’s level of connections during a classroom activity) to compare network structures and the effects of contextual factors on the number and strength of ties. Mameli and colleagues then used SNA to identify different discursive patterns in these classrooms, focusing on whether the discourse was led by the teacher or students and if talk was distributed equally across classroom members. Though both studies have limitations resulting from methodological choices that introduce possible biases or limit the generalizability of the findings, the authors point to the potential of SNA to explore how discourse structures may affect learning. Despite this potential, few researchers have utilized SNA for the study of discourse in educational contexts. Here, we propose ways of conceptualizing, creating, and analyzing data from talk with SNA so as to move past common hurdles that hinder the wider application of SNA to the study of discourse.

Creating Relational Data From Talk

Transforming discourse into the relational data sets that are needed for SNA has presented challenges for the analysis of data collected from the field and is one likely reason for the limited number of discourse studies using this methodology. While tools for the collection and analysis of data on discourse networks, including high-quality video recorders, data storage, and transcription and social network software, are more widely available and accessible to researchers, there have not been methods reported for conducting analyses on in-person talk, such as conversations between students during instructional activities or group discussions. This section reports on one approach to preparing discourse data from in-person, naturalistic settings, such as school classrooms, for analysis with SNA.

The most accurate assessment of dialogic ties can be obtained by relying on the direct observation of participants through video or audio recording (Carolan, 2014; Daly, 2010). This can include talk in various classroom contexts, such as turn and talk, small group discussions, or whole class discussions, and in other contexts, such as faculty meetings or working groups. To be analyzed, these recordings must be transcribed and coded to construct a relational data set (Knoke & Yang, 2008; Scott, 2013; Wasserman & Faust, 1994).

The coding of dialogic ties requires attention to three specific features of relational data: relation, directionality, and degree. Methods for coding online discourse developed by Thorpe et al. (2007) provide a framework for coding these features of talk. This coding approach can be flexibly adapted to in-person talk and provides a starting point for identifying dialogic ties in in-person discourse. In this approach, the relation between persons in the network is a turn, defined as a change of the speaker (Bakhtin, 1986; Chinn, Anderson, & Waggoner, 2001). Transcript data can be divided into turns as shown in a sample of transcript data in Table 1. Turns can be coded for directionality by identifying the speaker and one or more intended recipients of each utterance. A turn and its direction together form a dialogic tie. The ties among each set of participants can be counted to determine the degree, or the frequency of the tie in the discourse network.

Table 1

Sample of Transcript Data Divided Into Turns and Coded for Directionality

Turn	Speaker	Speech	Directionality
1	Teacher	So are you, Jessica, thinking that she should speak up or she should not speak up?	Question to Jessica
2	Jessica	She should speak up cause no teacher has the right to give you a new name.	Response to Teacher
3	Martín	I agree with her because what if someone, like, what if they call you a different name, like, every day? Like, hi Maria, hi Mary. And then what if they said you wouldn’t be Mary?	Response to Jessica
4	Jessica	Yea, I, I couldn’t.	Response to Martín
5	Martín	Yea, you wouldn’t like that.	Response to Jessica
6	Jessica	No.	Response to Martín
7	Carlos	[Simultaneously] I	Response to Martín
8	Carlos	I agree with Martín because she doesn’t feel comfortable with the name, and like, imagine, like, she thinks, like, the other girls stole her name.	Response to Martín

The data can then be assembled into an adjacency matrix that provides a complete accounting of the dialogic ties in the network (Knoke & Yang, 2008; Scott, 2013; Wasserman & Faust, 1994). An example of an adjacency matrix is shown in Table 2. This data matrix captures information about the relation, directionality, and degree of dialogic ties, and once input into SNA software, such as UCINET (Borgatti, Everett, & Freeman, 2006), it provides the basis for generating visualizations of the network and conducting analyses of the network structure.

Table 2

Adjacency Matrix Showing Dialogic Ties per Minute Between Participants in a Small Group

	Martín	Carlos	Helena	Nelson	Jessica	Teacher	Total Out
Martín	—	0.00	0.10	0.06	0.13	0.23	0.52
Carlos	0.06	—	0.06	0.07	0.03	0.15	0.37
Helena	0.08	0.03	—	0.06	0.30	0.35	0.82
Nelson	0.12	0.00	0.12	—	0.45	0.27	0.96
Jessica	0.33	0.12	0.38	0.60	—	0.56	1.98
Teacher	0.08	0.12	0.23	0.27	0.43	—	1.12
Total In	0.66	0.27	0.89	1.06	1.34	1.56	5.77

Note. The first column identifies the speaker. The first row identifies the recipient.

Though the process has been simplified in this explanation, and researchers will have to make decisions about how to handle defining turns and directionality in ambiguous or complex exchanges, this procedure offers a general framework for constructing relational data sets from transcribed talk. Given the accessibility of equipment for capturing high-quality video and audio of classroom talk, the construction of relational data from discourse should no longer be viewed as an obstacle to the use of SNA to study in-person discourse. In the authors’ experiences, when appropriate tools and software for the collection and analysis of data are used, the approach described here is not more time consuming than other approaches to the analysis of discourse that rely on coding and conducting analyses of transcript data.

Approaches to the Analysis of Discourse Networks

In this section, we overview methods and network features that may be particularly well suited for the study of discourse in educational contexts. Because networks are comprised of data that link persons, the assumption of independence on which inferential statistics depend is violated (Knoke & Yang, 2008). This has required the development of analytic methods specific to relational data sets (Carolan, 2014; Scott, 2013; Wasserman & Faust, 1994). Researchers must be aware that not all network methods are appropriate to all data sets and that school and classroom contexts may present particular challenges and limitations, such as the small sample size of many classrooms and groups. A few approaches and network features are overviewed here to provide examples of how SNA may enable new or deeper understandings of classroom structures and learning. Throughout, when possible, we highlight studies that have used these SNA approaches to examine discourse, both in naturalistic settings as well as online or digital spaces.

Network Features

Researchers using SNA may study a whole network or a single person within a network. Whole network analysis involves looking at the dialogic ties between all persons in a group. For example, a researcher wanting to explore correlations between speech patterns in a classroom and student attributes, such as first language or gender, might use whole network data. Though strict data requirements exist for whole network analyses, such as minimum group sizes that may exceed the sizes of some groups used in classroom instruction, they are often the most informative and can provide insights about discourse patterns in groups or classrooms (de Lima, 2010). Egocentric analysis focuses on the ego, or a focal person within the network, and the people who talk with the ego. For instance, researchers interested in examining the teacher’s use of talk during classroom activities might undertake an egocentric analysis of the teacher. This approach has more modest data requirements as it does not require the collection of data among all members of a group or classroom (de Lima, 2010). The type of network studied should be chosen based on the research purpose and the availability and quality of data.

For both whole networks and ego networks, descriptive measures of the network can provide rich information about the nature of discourse between persons. Among key network measures that are relevant to discourse networks are density, reciprocity, and centrality.

Density describes the general level of linkage in a network. Density is a comparison of the actual number of dialogic ties present in a network to the total number of dialogic ties possible if all persons are connected (Scott, 2013). Density provides a relative measure of how many people within a group are talking to each other. A higher density indicates more members of a network are talking to each other. Lipponen, Rahikainen, Lallimo, and Hakkarainen (2003) used this measure to analyze the extent to which upper elementary students were exchanging ideas and responding to one another’s comments in a computer-supported learning environment. They found that although all students interacted through the online learning environment, which resulted in a high density for the class network, there was variation in the amount of students’ participation. Though the digital environment appeared conducive to enabling broad participation, the authors pointed to the need to develop ways to better support sustained participation.

Reciprocity measures the proportion of dialogic ties that are returned (Scott, 2013). If a person speaks to another member of a network but that person does not respond, a tie goes unreciprocated. When used to describe a network as a whole, reciprocity provides an indication of whether discourse is conversational or consists of disjointed statements. When used to assess the ties of individuals in a network, reciprocity provides a relative measure of who is included or excluded by other members of the network. For example, Mameli and colleagues (2015) considered the reciprocity of teacher and student exchanges to augment their understandings of the distribution of student and teacher talk in the classroom described previously. Reciprocity draws attention not only to the outgoing talk of the students and teachers but to incoming talk and the creation of dialogic or conversational discourse.

Reciprocity also provides insights into the power dynamics of learning contexts by considering whose ideas and speech are recognized, taken up, or co-opted within a group and how this occurs in relation to the conditions, structures, and rules in which the discourse occurs (Foucault, 1972; Weedon, 1987). Persons perceived to have low status, capital, or power are more likely to be ignored or rejected and therefore have lower reciprocity. Persons perceived to have high status, capital, or power are more likely to have their talk recognized, responded to, and taken up by the group and therefore are likely to have higher reciprocity. Patterns of reciprocity within a group can be used to consider the social structures and history that are perpetuated within learning contexts and shape the talk of teachers and children in schools (Bourdieu, 1991). This can include ways that relational and structural power can create opportunities and constraints for the participation of persons from specific groups within classrooms and other learning contexts.

Centrality applies only to individual persons and provides a measure of how well connected a person is within a network (Carolan, 2014). Centrality is calculated by summing the ties of a single person. Comparisons of the amount of talk by members of a classroom or group can be made by comparing these values. Because heavy talkers will have high centrality values regardless of how many people in a network they talk with, relative centrality is used to measure how many people within a group a given person is engaging in talk. Relative centrality is calculated by dividing the number of people a person is connected to by the total number of people in the network (Scott, 2013). These two measures together provide an indication of a person’s level of participation in group discourse. For example, Hakkarainen and Palonen (2003) used measures of centrality to examine discourse between female and male students in a computer-supported learning environment. These measures enabled the researchers to examine the extent to which students engaged in discourse with their peers on the digital platform, including differences in connectedness by gender.

Like with reciprocity, centrality can indicate status within a group by identifying those with or without capital or power (Bourdieu, 1991). Persons with high capital or status tend to have more resources and opportunities in which to establish contacts and interactions during discourse. When considered through a critical lens, centrality can function as a measure of access, opportunity, or privilege in a learning context.

Visualizing Networks

Visual representations of networks, called sociograms, can enable the identification of basic structural features of networks. These can include discourse patterns, groups leaders and isolates, and reciprocity (Scott, 2013). If charted over time, sociograms can also show changes in group talk structures. Sociograms enable broad exploratory analyses of network structures and can guide more robust analyses of specific network features. Sociograms can be constructed by hand or using various software programs that are now widely available, such as UCINET (Borgatti et al., 2006) or Gephi (Gephi Consortium, 2017). These programs enable the quick construction of sociograms for whole networks and egocentric networks and allow various network features to be easily highlighted or explored. Moody, McFarland, and Bender-deMoll (2005) provide a discussion of how sociograms can be made dynamic, enabling researchers to see how the structure of a discourse network changes across a single conversation (e.g., how a discourse network might become more dense or how a person might shift from the center to the periphery of a network as their centrality declines). Later in this article, we share and describe an egocentric sociogram in the example study.

Network Structures

More robust analyses of discourse networks can reveal trends in group talk patterns or features of discourse that can be difficult to identify from the visual inspection of sociograms. Though many methods exist to explore networks and most methods may be applicable to discourse networks given the appropriate context or question, four analytic approaches are suggested here, with a brief description of each and what purpose it might serve in the analysis of discourse. These are: (a) position analysis, (b) cluster analysis, (c) attribute analysis, and (d) longitudinal analysis.

Position analysis

Position analysis provides specific information about the roles of persons in a network, including which persons hold similar roles (Borgatti & Everett, 1992; Knoke & Yang, 2008). Position analysis considers the direction and degree of a person’s dialogic ties, along with the characteristics of the group members to whom a person is tied, to characterize their role in a network. For example, a person may be identified as a leader of a clique or subgroup within a network; an outsider who is excluded on the basis of age, gender, or another characteristic; or a bridge between two cliques or subgroups of differing languages or genders. This enables researchers to identify and explore the ways talk is enacted or used by different persons in a network. This can reveal norms and structures that guide group discourse and point to rules and roles that are adopted by students and teachers as they negotiate interactions in learning contexts. For example, exploring the degree of students’ dialogic ties enabled Yoon (2008) to identify rules that students were following, such as “do as the smart students do” (p. 908) when making decisions of which peers to talk to during debates of socio-scientific issues.

Cluster analysis

Cluster analysis provides insights into the patterns and structures that are created within a network (Katz, Lazer, Arrow, & Contractor, 2004). Clusters are subgroups within a network and defined by their contiguity and separation from other clusters. The identification of clusters within a discourse network can point to groups of persons who are engaged in regular or sustained discourse. This may suggest either productive sites of discourse or the formation of groups that point to the inclusion or exclusion of some persons. This could include discourse among students or discourse among adults at faculty meetings, PTA meetings, or other school- and learning-related contexts.

Attribute analysis

Attribute analysis explores associations between attributes of persons in a network, such as first language, race, or gender, and network measures, such as centrality or position (Scott, 2013). Attribute analyses can provide statistical measures of the correlation between characteristics that have been shown to affect school outcomes and discourse patterns in a classroom or group (Carolan, 2014). For example, Ryu and Lombardi (2015) compared an English language learning student’s position across discourse networks from a whole class discussion, playground group, and science class group. These analyses have the potential to identify patterns of discourse that may be associated with the inclusion or systematic exclusion of subgroups from learning activities in classrooms and aide in the identification of teaching approaches that can more effectively include marginalized students.

Longitudinal analysis

If applied longitudinally, SNA can measure changes in the structure and patterns of discourse in networks over time (Carolan, 2014; Snijders, van de Bunt, & Steglich, 2010). This can include changes in the participation of individual students; the inclusion of specific subgroups in discourse, such as students with special needs or emergent bilingual students; or teacher talk patterns, such as attempts to move away from teacher-centered discourse structures. De Laat and colleagues (2007b) used this type of analysis to explore changes in the teaching styles of two instructors in an online learning community, specifically looking at the instructors’ roles in facilitating discussions over time. Longitudinal analysis can also provide new approaches to assessing the effectiveness of instructional interventions by using discourse networks to assess changes in participation, engagement, and communication structures during learning activities.

Combining Social Network Analysis With Other Methods

Though there are many analytic advantages to using SNA to investigate discourse, network data can overlook both the social contexts of talk and individual aspects of decision making. The combination of SNA with other forms of data, such as interviews or naturalistic observations, or with other forms of analysis can strengthen research and provide broader evidence from which to draw conclusions about both the persons and interactions within a discourse network. Sequential designs that use SNA to identify and guide other analyses of discourse, including selections from larger data sets, may help researchers to study features of talk in ways that are more complex and holistic. Simultaneous designs that use SNA alongside other analyses of discourse also hold promise as a form of mixed methods research that enable each analytic approach to inform decision making and understandings of the data concurrently.

In particular, the coupling of SNA with various forms of conversation or discourse analysis (Florio-Ruane & Morrell, 2011) presents a promising approach to exploring both the structure and content of talk. Conversation and discourse analysis have been used to examine the social organization, structure, and content of discourse and share similar aims with SNA (Heller, 2003). For example, González-Howard and McNeill (2017) used SNA and discourse analysis to examine large group discussions in three middle school science classrooms. These researchers first used SNA to create sociograms illustrating the discourse patterns around critique between the teachers and students. These sociograms enabled the researchers to identify groups in which students frequently critiqued their peers’ ideas. The researchers then applied discourse analysis to these focal groups’ conversations to characterize interactional moves that encouraged students to engage in critique. Sequential applications of discourse analysis such as this one enable researchers to consider both the broad structural aspects of discourse within a learning context and how language and other expressions are used by participants to engage in learning and enact various practices, relationships, and identities. That SNA has different functions and capabilities than current methods used to analyze discourse allows it to complement these methods.

An Example

To show how the ideas suggested would work on research in a classroom setting, we draw on an example from the first author’s research (Wagner & Proctor, 2015). Because of space limitations, we provide only a brief outline of the research context. This example serves as an illustration of the investigative potential of SNA and how it can be applied to the study of discourse in learning contexts.

The research was grounded in a real classroom problem: Small discussion groups were being implemented by teachers in classrooms with significant numbers of students who were emergent bilinguals. We wanted to know whether emergent bilingual students were being included in talk in these groups to know whether these instructional events were providing real opportunities for these particular students to engage in language practice. We posed two research questions: (1) What positions do emergent bilingual students occupy in small discussion groups? and (2) How are these emergent bilingual students enabled or disabled as participants?

This study was part of a multiyear research collaborative that included one elementary school in an urban Northeastern district and two area universities (Ossa Parra et al., 2016). The school was culturally and linguistically diverse, with 66% of students speaking a first language other than English. One discussion group in a Grade 4 classroom was the focus of this case study. Martín and Carlos were Spanish-English emergent bilingual students. They were placed in a discussion group with three monolingual English students, Jessica, Nelson, and Helena. The group met over an 8-week period during the reading of a novel. Six discussion sessions were video recorded, totaling 39:35 minutes, 253 turns, and 5,568 words. These students and this teacher were just beginning to learn to participate in this form of discussion, resulting in relatively short discussions during the period of study described here.

Because of our interest in the emergent bilingual students’ roles during these discussions, the approach we applied to this network was position analysis (Borgatti, Everett, & Johnson, 2013), which allowed us to map interactions and students’ positions within the group. We also used discourse analysis (Bucholtz & Hall, 2005; Gee, 2014) to examine how the content and nature of talk was used to construct or enforce positions and structures in the group. Both approaches treat discursive interactions as the unit of analysis. These complementary techniques provide different functions and capabilities in the analysis of discourse and together allowed us to consider both relationships and content in the group discourse. The sequential use of SNA followed by discourse analysis allowed the former to inform and guide our use of the latter, which directed our attention to aspects of the discourse that might otherwise have been overlooked.

The egocentric sociograms in Figure 1 provide a visual display of the patterns of outgoing and incoming dialogic ties for Carlos, whose findings will be discussed here. Though both Carlos and Martín provide compelling examples of how these methods can be applied to the study of classroom interactions, we have selected to focus only on Carlos so that we might speak in more detail and provide a clearer and more fully explained example of how these methods were applied to an analysis of his case.

Figure 1.

Egocentric sociograms showing the (left) outgoing and (right) incoming dialogic ties for Carlos.

Carlos’s outgoing dialogic ties were characterized by infrequent, though even, participation with other group members. Carlos extended ties to all of his peers, though favored talk directed to the teacher. Though Carlos positioned himself centrally, he made few strong dialogic ties within the group, as shown by the low weights of his outgoing dialogic ties. Only two of four peers and the teacher reciprocated these ties by talking to Carlos. As seen in the sociogram on the right, the students were less responsive than the teacher to his participation and directed talk to Carlos at a lower rate than he did to others. They often ignored his contributions, though returned comments made by other students. Only Jessica, the most talkative child in the group, directed talk to Carlos more than he directed talk toward her. However, the application of discourse analysis to the data enabled more nuanced understandings of the nature of these interactions (Bucholtz & Hall, 2005; Gee, 2014). This attention to the content of Jessica’s comments showed that even she often dismissed his claims or attributed them to other students in the group. These patterns show how Carlos was largely unable to gain acceptance from his peers despite his attempts to establish a central position within the group.

In the investigation of the positions the students constructed in this group, the first author found that students did not consistently reciprocate talk to Carlos and peers, rather than the teacher, played a central role in enabling and disabling Carlos’s participation. This analysis provides insights into the problem of “recognition and being recognized” (Gee, 2014, p. 49), or the attempts of emergent bilingual students to be recognized as participants and able students in school. In this case, the use of SNA provided insights into the interactions that affected Carlos’s inclusion in the group. By pairing this analysis with discourse analysis, we were also able to gain deeper insights into how language was employed by the participants to include and exclude specific participants in these interactions. Though the example here is brief, combining SNA with discourse analysis ameliorated analytic limitations of both methods and facilitated a more holistic understanding of discourse processes in a naturalistic learning environment.

Considerations in Social Network Research on Discourse

Conceptualizing discourse as relational data that can be studied through the analysis of dialogic ties in discourse networks offers a new method for studying classroom discourse that is consistent with widely used theories that regard discourse as a social phenomenon. SNA redirects attention to the connected nature of discourse and the complexity of social interactions in learning events. This methodological approach presents opportunities for the examination of new or under-researched aspects of classroom talk, including discourse patterns and structures, what positions are available and who holds them, whether positions are correlated with attributes, who is included or excluded, and whether participation is equitable.

Social network methods will require continued development to reach their full potential for the study of discourse. Current unanswered questions about this methodology include considerations of sample selection, such as processes to select a sample whose discourse patterns may be representative of a larger population, and samples that are of sufficient size to power more complex social network methods. SNA alone is often not sufficient to understand critical details about discourse networks, including the nature of discursive interactions. For example, SNA can miss some key details about what occurs in discourse, such as the ways language is used to exclude certain participants. This points to the limitations of quantifying talk to explain discursive interactions. Despite these challenges, applications of SNA will likely play a critical role in expanding the reach and value of research on discourse. Further explorations of the theoretical and methodological alignments between social networks and discourse will extend understandings of both these phenomena and the methods used to explore them.

Footnotes

Authors’ Note

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

ORCID iD

Christopher J. Wagner

Authors

CHRISTOPHER J. WAGNER, PhD, is an assistant professor of Early Childhood Education at Queens College of the City University of New York, Powdermaker Hall 054, 65-30 Kissena Blvd., Queens, NY 11367-1597; christopher.wagner@qc.cuny.edu . His research focuses on early literacy and biliteracy, the development of reading and writing identities in young children, and the teaching of multilingual children in early childhood.

MARÍA GONZÁLEZ-HOWARD, PhD, is an assistant professor in STEM Education at the University of Texas, Austin; mgonzalez-howard@austin.utexas.edu . Her research explores the intersections of teaching and learning science with bilingualism development, with a focus on supporting culturally and linguistically diverse students’ engagement in science practices.

References

Bakhtin

M. M.

(1981). The dialogic imagination ( Emerson

Holquist

, Trans.). Austin, TX: University of Texas Press.

Bakhtin

M. M.

(1984). Rabelais and his world ( Iswolsky

, Trans). Bloomington, IN: Indiana University Press.

Bakhtin

M. M.

(1986). The problem of speech genres. In Emerson

Holquist

(Eds.) & McGee

V. W.

(Trans.), Speech genres & other late essays (pp. 60–102). Austin, TX: University of Texas Press.

Bales

R. F.

Strodtbeck

F. L.

Mills

T. M.

Roseborough

M. E.

(1951). Channels of communication in small groups. American Sociological Review, 16(4), 461–468.

Bavelas

(1950). Communication patterns in task oriented groups. Journal of the Acoustical Society of America, 22, 723–730.

Borgatti

S. P.

Everett

M. G.

(1992). Notions of position in social network analysis. Sociological Methodology, 22, 1–35.

Borgatti

S. P.

Everett

M. G.

Freeman

(2006). UCINET 6 for Windows (Version 6.286) [Computer software]. Lexington, KY: Analytic Technologies.

Borgatti

S. B.

Everett

M. G.

Johnson

J. C.

(2013). Analyzing social networks. Los Angeles, CA: Sage.

Bourdieu

(1991). Language and symbolic power. Cambridge, UK: Polity.

10.

Bourdieu

Passeron

J. C.

(1990). Reproduction in education, society and culture (2nd ed.). ( Nice

Trans.). Los Angeles, CA: Sage.

11.

Bucholtz

Hall

(2005). Identity and interaction: A sociocultural linguistic approach. Discourse Studies, 7(4–5), 585–614.

12.

Carolan

B. V.

(2014). Social network analysis and education: Theory, methods & applications. Los Angeles, CA: Sage.

13.

Cazden

(2001). Classroom discourse: The language of teaching and learning (2nd ed.). Portsmouth, NH: Heinemann.

14.

Chai

Tan

(2009). Professional development of teachers for computer-supported collaborative learning: A knowledge building approach. Teachers College Record, 111(5), 1296–1327.

15.

Chinn

C. A.

Anderson

R. C.

Waggoner

M. A.

(2001). Patterns of discourse in two kinds of literature discussion. Reading Research Quarterly, 36(4), 378–411.

16.

Daly

A. J.

(2010). Social network theory and educational change. Cambridge, MA: Harvard Education Press.

17.

de Laat

Lally

Lipponen

Simons

R-J

. (2007a). Investigating the patterns of interactions in networked learning and computer-supported collaborative learning: A role for social network analysis. Computer-Supported Collaborative Learning, 2(1), 87–103.

18.

de Laat

Lally

Lipponen

Simons

R-J

. (2007b). Online teaching in networked communities: A multi-method approach to studying the role of the teacher. Instructional Science, 35, 257–286.

19.

de Lima

J. A

. (2010). Studies of networks in education: Methods for collecting and managing high-quality data. In Daly

A. J.

(Ed.), Social network theory and educational change (pp. 243–258). Cambridge, MA: Harvard Education Press.

20.

Florio-Ruane

Morrell

(2011). Discourse analysis: Conversation. In Duke

N. K.

Mallette

M. H.

(Eds.), Literacy research methodologies (2nd ed.. pp. 87–103). New York, NY: Guilford.

21.

Foucault

(1972). The archaeology of knowledge ( Sheridan Smith

A. M.

, Trans.). New York, NY: Pantheon Books.

22.

Gee

J. P.

(2012). Social linguistics and literacies: Ideology in discourses (4th ed.). New York, NY: Routledge.

23.

Gee

J. P.

(2014). An introduction to discourse analysis: Theory and method (4th ed.). New York, NY: Routledge.

24.

Gephi Consortium. (2017). Gephi [Computer software]. Retrieved from https://gephi.org/

25.

Goffman

(1959). The presentation of self in everyday life. Garden City, NY: Anchor Books.

26.

Goffman

(1981). Forms of talk. Philadelphia, PA: University of Pennsylvania Press.

27.

González-Howard

McNeill

K. L.

(2017, April). Variation in how teachers support student critique in argumentation discussions. Paper presented at the annual meeting of the National Association for Research in Science Teaching, San Antonio, TX.

28.

Hakkarainen

Palonen

(2003). Patterns of female and male students’ participation in peer interaction in computer supported learning. Computers & Education, 40(4), 327–342.

29.

Heath

S. B.

(1983). Ways with words: Language, life, and work in communities and classrooms. Cambridge, UK: Cambridge University Press.

30.

Heller

(2003). Discourse and interaction. In Schiffrin

Tannen

Hamilton

H. E.

(Eds.), The handbook of discourse analysis (pp. 250–264). Malden, MA: Blackwell.

31.

Kadushin

(2012). Understanding social networks. New York, NY: Oxford University Press.

32.

Katz

Lazer

Arrow

Contractor

(2004). Network theory and small groups. Small Group Research, 35(3), 307–332.

33.

Knoke

Yang

(2008). Social network analysis (2nd ed.). Los Angeles, CA: Sage.

34.

Leavitt

H. J.

(1951). Some effects of certain communication patterns on group performance. Journal of Abnormal and Social Psychology, 46, 38–50.

35.

Lipponen

Rahikainen

Lallimo

Hakkarainen

(2003). Patterns of participation and discourse in elementary students’ computer-supported collaborative learning. Learning and Instruction, 13, 487–509.

36.

Mameli

Mazzoni

Molinari

(2015). Patterns of discursive interactions in primary classrooms: An application of social network analysis. Research Papers in Education, 30(5), 546–566.

37.

Moody

McFarland

Bender-deMoll

(2005). Dynamic network visualization. American Journal of Sociology, 110(4), 1206–1241.

38.

Oshima

Matsuzawa

(2012). Knowledge building discourse explorer: A social network analysis application for knowledge building discourse. Educational Technology Research and Development, 60, 903–921.

39.

Ossa Parra

Wagner

C. J.

Proctor

C. P.

Leighton

C. M.

Robertson

D. A.

Paratore

J. R.

Ford-Connors

. (2016). Dialogic reasoning: Supporting emergent bilingual students’ language and literacy development. In Proctor

C. P.

Boardman

Hiebert

E. H.

(Eds.), Teaching emergent bilingual students: Flexible approaches in an era of new standards (pp. 119–137). New York, NY: Guilford Press.

40.

Resnick

L. B.

Asterhan

C. S. C.

Clarke

S. N.

(2015). Socializing intelligence through academic talk and dialogue. Washington, DC: American Educational Research Association.

41.

Ryu

Lombardi

(2015). Coding classroom interactions for collective and individual engagement. Educational Psychologist, 50(1), 70–83.

42.

Scott

(2013). Social network analysis (3rd ed.). Los Angeles, CA: Sage.

43.

Sharma

Tietjen

(2016). Examining patterns of participation and meaning making in student blogs: A case study in higher education. American Journal of Distance Education, 30(1), 2–13.

44.

Shea

Hayes

Vickers

Gozza-Cohen

Uzuner

Mehta

. . . Rangan

. (2009). A re-examination of the community of inquiry framework: Social network and content analysis. Internet and Higher Education, 13, 10–21.

45.

Snijders

A. B.

van de Bunt

G. G.

Steglich

C. E. G.

(2010). Introduction to stochastic actor-based models for network dynamics. Social Networks, 32, 44–60.

46.

Street

(1984). Literacy in theory and practice. Cambridge, UK: Cambridge University Press.

47.

Thorpe

McCormick

Kubiak

Carmichael

(2007). Talk in virtual contexts: Reflecting on participation and online learning models. Pedagogy, Culture & Society, 15(3), 349–366.

48.

Wagner

C. J.

Proctor

C. P.

(2015, April). Bilingual learners talking in school: Identity and inclusion in small discussion groups. Poster session presented at the American Educational Research Association Annual Meeting, Chicago, IL.

49.

Wasserman

Faust

(1994). Social network analysis: Methods and applications. New York, NY: Cambridge University Press.

50.

Weedon

(1987). Feminist practice and poststructuralist theory. Oxford, UK: Basil Blackwell.

51.

Yoon

(2008). Using memes and memetic processes to explain social and conceptual influences on student understanding about complex socio-scientific issues. Journal of Research in Science Teaching, 45(8), 900–921.

52.

Yoon

(2011). Using social network graphs as visualization tools to influence peer selection decision-making strategies to access information about complex socioscientific issues. Journal of the Learning Sciences, 20(4), 549–588.