Introduction

Dialogic Education is an important movement, considered by many as having the potential to lead to massive educational change (Alexander, 2020; Mercer & Howe, 2012; Wegerif et al., 2019). The philosophical and ideological tenets of Dialogic Education have been well articulated, yet the methodologies developed so far for analyzing processes involved in the implementation of programs designed to enhance Dialogic Education do not provide a broad enough picture. In particular, methods used to analyze classroom talk entailed by the implementation of dialogic pedagogies generally focus on isolated (rather than sequences of) communicative acts. We claim that other methods should be combined to delineate what is within and beyond educational dialogues in classroom talk. More generally, we will claim that the methods adopted do not handle well enough what we see as the tensions between dialogism and Dialogic Education. Our goal is to show that the actor-network theory (ANT) handles these tensions, and that it extends the scope of most of the methodologies used to analyze educational dialogues. To do so, we will compare the scheme for educational dialogue analysis (SEDA), which is currently a leading methodology for analyzing educational dialogues, and ANT which has not been used so far to analyze classroom talk in Dialogic Education. The comparison is undertaken through the analysis of some protocols collected in a program dedicated to Dialogic Education, and thereby stress the respective advantages and weaknesses of both methodologies. We show that the use of ANT enables observation of crucial phenomena for the analysis of classroom talk, especially for describing the deployment of educational dialogues: It enables observation of the formation (and dissolution) of networks that constitute the socio-material structures of classroom talk. It enables the inclusion of non-human actors in those networks, and the identification of what we call dialogic networks, in which there is no subordination of any human actor. We conclude that ANT is important in Dialogic Education alongside methodologies such as SEDA, for research and educational practice—especially for evaluating the social and material structures that maintain educational dialogues, and for helping teachers in the implementation of programs boosting Dialogic Education.

The tensions between dialogism and dialogic education

Dialogic Education is a pedagogical movement that relies on the theory of Dialogism (Skidmore, 2019). Dialogism is a social theory of language developed by Bakhtin, that inspired scholars from various traditions and disciplines (Holquist, 1990). According to Bakhtin, in conversations, every utterance is connected to another one, in order to constitute an organized stream in which people respond to one another. Without such an organization, conversations cannot occur. Moreover, often, once the present utterance has been heard—it is interrelated to a past conversation.

Bakhtin attacks the monologization of human experience reflected in the linguistic, literary, philosophical, and political theories of his time (Baxter, 2006). He opposes the anthropocentric humanist system of how we make sense of ourselves and of our material world: Classical humanism sees nature as a set of universal, and immutable patterns, that individuals can decipher. As a general social theory of language, Dialogism may focus on demagogic speeches delivered by politicians in front of mesmerized and silent audiences, or on angry disputes among couples, as relevant settings.

Dialogic Education favors specific forms of social talk. Dialogic Education, means ‘‘to be better at dialogue: learning to ask better questions, how to listen better, learning not only the words but also the implicit meanings, how to be open to new possibilities and new perspectives while learning how to think critically about new perspectives through comparing different points of view (Wegerif et al., 2019). Wegerif has advanced the concept of dialogic space (Wegerif et al., 2019) within which new meaning can be developed in educational settings, given an appropriate orientation towards learners on the part of the teacher. But how the orientation can be judged as appropriate? (Skidmore, 2019) has presented a theoretical framework that extends the concept of dialogic space by leveraging fundamental tenets of the social theory of language established by the Bakhtin circle. This framework outlines the various dimensions that could potentially structure such a space. This framework may provide some advancement on the issue of appropriate orientation. Skidmore displays this framework in a dialogic sphere that presents three dimensions. First, addressivity materializes the fact that in a speech there is always a kind of interlocutor, the quality of turning to someone (Bakhtin, 1986). The presence of the interlocutor can be integral to the way in which utterances are articulated, and at the opposite pole, a monologic mode of address, in which an authoritative speaker holds the floor and others are treated as an anonymous audience for his speech. In classroom talk, teachers frequently ‘lecture’ and do not address the students in their classes.

The second dimension is the voicing given to students. Plenary forms led by the teacher in a classroom, in which the teacher will call on many different students to answer his/her questions is dialogic but homophonic, in the sense that their contributions are orchestrated by the teacher’s management of the discussion. It is the teacher that controls the flow of the talk and allocates the right to talk to the next student contributor, often by name.

The third dimension—semantic permeability, varies between the poles of heteroglossic and orthoglossic forms of speech. The teacher is expected to provide a bridge between the heteroglot registers of everyday speech and the initially unfamiliar specialized languages of academic disciplines that require more orthoglossic (formal, sanctioned), linguistic practices. There needs to be enough permeability in the language of the classroom to enable students to cross the border between every-day and academic understanding of a subject.

The dialogic sphere makes salient tensions between the theory of dialogism and Dialogic Education: From the vantage point of dialogism, non-dialogic social talk is an oxymoron, but from the vantage point of Dialogic Education, it is possible. Skidmore’s dialogic sphere points at forms of communication that seem non-dialogic from an educational vantage point: this is the case when the talk is merely monologic/homophonic/orthoglossic. For example, when an authoritative speaker (generally the teacher but also a student in a small group) holds the floor without giving the floor to others, one might be inclined to say that the talk is not dialogic. We are inclined to say the same when in the case of plenary talk orchestrated for which the choice of voices the teacher permits to be expressed is selective. Also, we are inclined to claim that when there is no intercourse between heteroglossic and orthoglossic forms of speech, educational talk is not dialogic from an educational perspective. However, the situation is more complex. For example, selective voicing may be dialogic, when some students are bashful, or arrogant. Also, holding the floor might be dialogic, when the teacher intends to summarize previous activities and integrate ideas expressed previously by some students. Deciding on whether social talk is dialogic from an educational vantage point depends on the specific context in which the talk takes place. There is then a tension between Dialogism and Dialogic Education—between the dialogic by definition and the dialogic because appropriate educational aims are reached, and whose determination as dialogic is made in situ. In the deployment of classroom talk, the dialogic vision encounters teachers and students, and as well as a material context, that make it difficult to enable the emergence of educational dialogues: the fact that voices in educational talk should be aligned to what is recognized as correct, the centrality of the teacher as the authority controlling the emergence of official knowledge, the compliance with the finalizable character of school education, or the attainment of clear learning outcomes are a few of the phenomena that make it difficult to implement dialogic pedagogy (Segal & Lefstein, 2016). There is then a rich material context (curriculum, texts, instructions, etc.), and social context in which classroom talk evolves, in which deciding whether talk reaches appropriate goals according to Dialogic Education is complex. There is then a need to use tools that reflect the reliance of Dialogic Education on Dialogism, and that still enables tracing the emergence of the dialogic in its social and material intricacies, and delineating it from the non-dialogic (dialogic and non-dialogic from the vantage point of Dialogic Education).

In this paper, we will claim that the actor-network theory (ANT) helps in this endeavor. This methodology, which has not been applied to educational dialogues so far, will be presented and compared to the SEDA method of analysis—a well-established methodology for the analysis of educational dialogues. The comparison will help stressing what ANT (resp. SEDA) contributes to the analysis of classroom dialogues.

The adoption of ANT as a methodology for tracing the deployment of educational dialogues fits a novel reflection on Bakhtin’s theory of Dialogism. Dialogism is bound by its Humanist threads: it focuses on the human as the instrument of knowledge making. The centrality of alterity and dialogue across human differences means that knowledge-making is mutual among humans. The mutuality of knowledge-making among humans leaves the material world out of the dialogue. This is evidenced, for example, by the take-up of ‘the utterance’ as the basic unit of analysis (Bakhtin, 1935) in dialogical research. Following Deleuze and Guattari (1987), Davies and Renshaw (2019) explore the posthuman configuration of ‘the human’ that ‘‘stretches the notion of human becoming far beyond the unfinalisability of dialogism and its humanist legacies’’ (Davies & Renshaw, 2019, p. 42). They adopt a posthumanist political approach that ‘‘dislocates the foundations of the humanist worldview’’ (Braidotti, 2006, p. 203). For them, human is neither a biological nor sociological category, but an interface, a field of intersecting material and symbolic forces; a surface where multiple codes (sex, class, age, race, etc.) are inscribed. Their claim echoes Latour’s statement that sociologists should analyze society as a network that consists of interrelations between both human and non-human actors (Latour, 1985). What comes to the fore are new human-nonhuman linkages, which include complex media-technological interfaces of biological and nonbiological matter (Callon & Latour, 1981). Braidotti’s posthumanism allows her (and Davis & Renshaw, 2019) to state that matter is not organized in terms of dualistic mind/body oppositions, but rather as materially embedded and embodied subjects in-process. The ontological twist instantiated in Latour’s ANT methodology, which posits the primacy of relations over substances, provides fresh lenses to observe the dialogic as it deploys itself in the rich context of classrooms.

In the next section, we overview traditional methods for analyzing educational dialogues and discuss whether the methods uncover enough perspectives to evaluate them. Then we describe the ANT methodology, which seeks to account for social and material structures through an analysis of the processes and connections through which networks are assembled or dissolved.

Research methods for analyzing educational dialogues

The elaboration of methods for analyzing educational dialogues is a vivid research domain (Hennessy et al., 2020; Kershner et al., 2020; Lefstein & Snell, 2019; Nystrand et al., 2003). The various methodologies recently elaborated for assessing quality of educational dialogues point at a general trend to improve talk practices and, since this improvement implies big pedagogical efforts (dedicated PD programs, instructional design, etc.), these methodologies are sensitive to the passage from old to new practices. Accordingly, schemes were designed for analyzing all forms of discourse, not just dialogic forms (e.g. Boyd & Markarian, 2015; Wells & Arauz, 2006). Most of the schemes adopt a micro-level of granularity, and focus on specific moves. For example, Accountable Talk® relates to teacher discourse moves only (Michaels et al., 2008). Other schemes address student moves only (Hardman, 2019; Kumpulainen & Wray, 2002). Several researchers focus on the third turn in triadic dialogue or the ‘‘follow-up’’ move, widely considered to be significant in shaping (or obstructing) the course of dialogue (Park et al., 2017).

In the recent volume Research methods for educational dialogue, Kershner and colleagues (Kershner et al., 2020) provide a picture of several methodologies that include linguistic ethnography, sociolinguistics, and ethnomethodology. The common denominator of these methodologies is the categorization or ‘‘coding’’ of some aspects of interaction, which reduces the reality (Wegerif, 2020; Wegerif & Mercer, 1997), but is indispensable for research on educational dialogues. For example, Kershner and colleagues (Kershner et al., 2020) describe in detail the scheme for educational dialogue analysis (SEDA) (Hennessy et al., 2016), whose micro-level codes—open questions, extended contributions, reasoning with evidence etc., are broadly used in the Learning Sciences. The generality of these codes indeed reduces reality, but at the same time opens opportunities for answering research questions in quantitative studies in which comparisons are made possible. Indeed, inferential statistics applied to SEDA showed the association of certain types of dialogue or frequency of contribution with learner characteristics (e.g., socioeconomic status or linguistic capability) or achievement (Howe et al., 2019). In other studies, counts of the frequencies of patterns of language use predicted learning gains, in both group and individual reasoning (Mercer, 2004). A further advantage of micro-level coding is the clustering of categories to achieve further analyses. For instance, the sub-categories of Accountable Talk® were found to be related to teachers’ goals (or challenges) in orchestrating whole-class talk that supports reasoning (Park et al., 2017).

Despite its advantages, micro-level coding is not without its challenges. Hennessy and colleagues are aware of these challenges (Hennessy et al., 2020), as they claim that one of the most serious drawbacks of micro-level coding is that it identifies isolated moves rather than sequences of moves labeled as dialogic. Accordingly, Mercer (2010) states that ‘‘the most serious are the problems of dealing with ambiguity of meanings, the temporal development of meanings, and the fact that utterances with the same surface form can have quite different functions’’ (p. 4). The disrespect to the temporal dimension is especially consequential when micro-level coding is coupled with frequency counting, since any sense of how moves interweave within exchanges is lost (e.g., Lefstein et al., 2015; Wells, 2009; Wells & Arauz, 2006), or of how partakers in dialogues reply to each other’s contributions, which is essential to dialogic interactions (Bakhtin, 1981). Researchers do not consider more than adjacent precedents in educational dialogues. As summarized by (Baker et al., 2021), an approach that takes into consideration various timescales is imperative for analyzing educational dialogues.

The most serious drawback of SEDA hinges on the tension between Dialogic Education and Dialogism. It does not tell when the classroom discourse is an educational dialogue and when it is not. More so, it focuses on the teacher’s actions. It does not focus on relations among students. Focus on relations—among students, between students and the teacher, or with non-human actors, over time is necessary. We claim that the ANT methodology provides lenses for this focus.

The actor-network-theory and its relevance to research on educational dialogues

ANT analysis concentrates on socio-material interactions. It explores how actors assemble and disassemble, shape and are being shaped by the interaction, and exert force on other actors. Analyses proceed by tracing the ‘‘careers of mediators along chain[s] of action’’ (Latour, 2005). Latour and Woolgar’s (1986) ethnography of a biomedical laboratory draws on citational data and research reports from the lab, supplemented with field work and interviews. Their ethnographic observations turn the lab into a place where facts are ‘secreted’ through processes that involve researchers as actors that assemble and interpret data. Their ‘‘social constructionist views of scientific activity’’ (Lynch, 1994, p. 91) contributed to the development of a ‘new’ sociology of scientific knowledge’’ (p. 71).

ANT researchers oppose approaches that concentrate only on humans, while objects, machines, and materials remain often overlooked. According to the ANT approach, all things, human and non-human, can assemble, exert force, change, and be changed. Therefore, instead of asking what influences ‘the social’, scholars should ask how the social is created and what are its components (Latour, 1993).

The application of the ANT socio-material approach to education should be contrasted to the trialogical learning (Paavola et al., 2014) for which objects play a part in any social interaction as a related approach to analysis and design. Trialogical learning, however, is not sensitive to changes in roles, and to (changes in) power relations.

ANT does not confer a priori definitions of entities. Humans and non-humans change through the process of becoming part of a collective network (Latour 1987). ANT differentiates between an actor and actant—actant refers to the mere involvement of the entity in actions, and actor refers to a role taken by the entity. The title of an actor does not convey a pre-given role; the roles of actors are translated by their actions, connections, and the change they make in a specific network. Agency, power, identity, and knowledge are relational effects of a certain network and not pre-given attributes. Certain assemblages stabilize and move about from one network to another network, and extend its original network’s power. When these assemblages move to new networks, they can be taken for granted; they are Black Boxed.

ANT has been the object of criticism as a social theory (e.g., Law, 1999). In particular, ANT has been attacked for its simplicity—for example the fact that behind the relation actor-network, there is a tension that elides the distinction between structure and agency, or that is behind the term translation, there is an oversimplification—the becoming of entities being substituted in their mere replacement. The thorny issue of the nature of the agency of non-human actors is also the object of criticism. We adopt ANT as a methodology, rather than a theory, though. We argue that its simplicity is an advantage for describing the deployment of educational dialogues. The fact that we look a priori for delineating dialogic and non-dialogic networks, would have been unacceptable for ANT as a theory, which describes networks without looking for theory-laden categories. Our methodologic approach is then inspired by ANT as a theory.

The use of ANT as a methodology is not new in educational research: For instance, Law (1992) presents an example of how material objects like overhead projectors (OHP) shape interactions in class. Teachers can control interactions by using an OHP to present and explain graphs and texts. In other cases, students may have access to the OHP to share, present and discuss ideas—meaning OHP has an agency in designing interactions and understanding. McGregor (2004) shows how a science teacher becomes a knowing location through the interaction of materials, routines, texts, and people. She describes how laboratory artifacts, instructional practices, and teacher’s traits are assembled into a network and create knowledge and agency. She shows that through this agency, the actant is influencing the interactions, and its presence is felt for a long time.

Fenwick and Edwards (2010) bring an example of Black Boxing in education and describe how curricula become Black Boxes. A curriculum is an effect of a heterogeneous network of people and things, and historical and political networks, which often solidify and become durable. Then it spreads and affects many teachers’ pedagogical choices and shapes the learning in classes. The teachers (and the stsudents) might take the curriculum for granted, leaving behind the ongoing discussion and interaction that led to its creation.

This review suggests that ANT may promote research on educational dialogues. Its focus on interactions over time among various actors refers to tensions between Dialogic Education and Dialogism. Cazden (2001) already described classroom discourse with the metaphor of a group of actors playing together: ‘‘Teacher and students may have different visions of how the performance should be performed, so the teacher assumes the dual role of stage director and chief actor.’’ (p. 40). ANT expands this metaphor beyond human actors and emphasizes the role of non-human actors shaping what happens in the classroom. Moreover, it perceives classroom interaction as an effect of complex networks of actors, and not necessarily as a direct outcome of the director or chief actor’s hand.

The educational context

In September 2019, the DIALOGOS Center—a Center dedicated to the promotion of dialogic argumentation, initiated a program in four schools in a mid-size town in Israel. A school professional development (PD) program was implemented in which teachers taught in Grade 7 in mathematics, humanities, and science. The school principals encouraged them to participate in the program in order to enact dialogic practices in their classes. We negotiated the program with all the stakeholders involved. The teachers and superintendents who were responsible for schools of the chosen municipality were approached. The superintendents pointed at the need for a program that would raise strong interest and engagement. The teachers, who were aware that dialogic practices in their classes involve major changes in their teaching practices, pointed out that the program should be focused on several out-of-regular-schedule learning days during a school year rather than on educational change in regular in-school teaching routines. We instigated the enactment of a dialogic pedagogy for enabling interdisciplinary dialogic argumentation (Koichu et al., 2022; Schwarz & Baker, 2016) among students, in a series of interdisciplinary activities—so-called ‘focus days’.

The general structure of each ‘focus day’ remained the same: it turned around a different societal issue; disciplinary sessions (in mathematics, science and philosophy) during which students learned about ideas relevant to the societal issue, preceded an interdisciplinary session in which a joint product (e.g., a presentation, a clip, a written story, a petition) was produced (Schwarz et al., 2024). During the day students participated in small groups or all-class activities, aimed at encouraging inquiry and dialogic argumentation. Deliberations around texts that invited inquiry and dialogic argumentation were conducted in all disciplines and in all sessions. These activities were prepared in collaboration with the DIALOGOS team (Koichu et al., 2022), but were conducted by the regular school teachers of the participating schools.

Each focus day was opened by an introductory talk given by an expert knowledgeable about the chosen issue to be discussed. Then disciplinary (science, mathematics, philosophy) 90-min preparatory sessions were conducted in parallel, for 2 times each. Each student took part in two out of three disciplinary sessions while arranged in groups of about 20 students from the four schools during the first part of the day (i.e., there were students who attended mathematics and science sessions, or mathematics and philosophy sessions, or philosophy and science sessions). This was followed by a 90-min interdisciplinary session, in which the students discussed the interdisciplinary issue of the day while capitalizing on what they had learned in the disciplinary sessions. Our study focused on the interdisciplinary session.

About 130 students from four seventh grades from different schools participated in the study. The students are all from ‘‘excellent classes’’ in their schools. The study also includes the teachers participating in the project. Each class has three teachers, one from each discipline. In the interdisciplinary session, in which we focus, 3 teachers were present in each classroom, guided the activities together and randomly joined the different groups.

Checking the dialogicity of two lessons of classroom discourse with ANT and SEDA

The aim of this study is to contrast between ANT and SEDA for the description of classroom talk. The research question on which we focus, is to what degree does each methodology analyze whether classroom talk is dialogic. The comparison between the methodologies will help understand their advantages and drawbacks. We chose to focus on two lessons of the program described above. Any choice is questionable as it raises the issue of representativity. More so, as the strength of SEDA resides in its power to compare kinds of interventions, based on inferential statistics, we confronted a dilemma. On the one hand, applying SEDA in two lessons misses its designation, which is to identify general trends through many lessons. On the other hand, ANT is an ethnographic methodology, hence helps to describe each lesson without pretending to provide generalizable processes. Therefore, the comparison between SEDA and ANT in the analysis of a few lessons is essentially problematic. However, our comparison is rather a juxtaposition, and we will claim that this juxtaposition provides insights for analyzing classroom discourse.

The first lesson is taken from the third focus day which was around the issue: ‘‘Would it be right for the government to mandate parents to vaccinate their children?’’. The students had a lesson in science about the biology of vaccination and its history, a lesson in mathematics about the math behind ‘‘the herd immunity phenomenon’’, and a lesson in philosophy about the different assumptions underlying western medicine supporting vaccination and eastern medicine which opposes them. In the last lesson, on which we focus, the students were arranged into small groups, and engaged in an activity designed to bring them to discuss the issue about the coercion of vaccination by the state.

The second lesson is taken from the fourth focus day which was around the issue: ‘‘Should the government allow parents to genetically engineer their future babies?’’. The students had a lesson in science about DNA and genetic healing, a lesson in mathematics about probability, relative frequency and the relationship between probability and genetics, and a lesson in philosophy about Science Fiction and Futurism. In the last lesson (on which we focus), the students engaged in a ‘court’ activity, in order to have an argumentative and dialogical discussion on the issue about genetic engineering. They were organized into three groups: judges, supporters or opponents of genetic modifications of embryos. To put together their arguments, each group was given a text which describes a future scenario of genetic modification of embryos and its possible health, psychological and sociological consequences, and cards containing opinions and positions that support or oppose genetic engineering of embryos.

We will first present the analysis of episodes from both lessons according to the ANT methodology. The ANT analysis consisted of several steps, which derive from the theoretical and methodological considerations we presented in the previous section. They were adapted for the context of analyzing classroom discourses. These steps are not fixed. Rather, the analysis processes are inter-connected and recursive:

  1. a.

    Recognizing actors: Identifying human actors, and non-human actors (physical or abstract).

  2. b.

    Tracing connections: the analysis involves portraying the interactions between the actors and how they converge into assemblages while exposing the dynamics, connections, cooperations, and tensions between them.

  3. c.

    Identifying translations: defining the actors’ influence on each other and the roles that are translated through this process. The roles of actors such as teachers, are defined through the network dynamics and do not have a priori role. In the episodes we analyzed, the teacher played alternatively the role of a source of knowledge, an activity facilitator, and a referee, according to the specific dynamic in the network.

  4. d.

    Identifying shifts and changes in the network: Highlight moments when new actors are introduced to the network, existing actors leave, and new dynamics are formed.

  5. e.

    Delineating the interactions’ outcomes: new knowledge, reduction or promotion of students’ agency, production of abstract or material outcomes, reinforcement, stabilization, or change of former norms and roles.

  6. f.

    Examining the connection between other networks and the current network: for example, understanding the circumstances and educational goals that lead researchers when designing educational activities and how these instructions were manifested in students’ activities.

  7. g.

    Revealing black boxes: sometimes, things created in one network move to other networks, are taken for granted, and shape the interaction. ANT analysis unpacks black boxes, such as tools, technologies, curriculum, and educational goals, whose origins are often forgotten, yet they spread to numerous networks and shape their interactions.

We collected around 150 discussions from the interdisciplinary sessions of the five focus days, in which students were arranged in different settings ((un-)guided small group discussions, teacher-led whole-class discussion, whole-class unguided debate). We audio-or video-taped these discussions, in order to identify episodes in which. All discussions were recorded. For all of them, the Ministry of Education and the authority for ethics at the Hebrew University authorized the recording (audio-or video-, depending on the consent of parents and students). All stakeholders agreed upon at least audio-recording of all sessions. We watched and listened to all recordings. We followed the guidance Derry and colleagues (Derry et al., 2010) provided for selecting, and analyzing the video and audio dataset of the discussions, in addition to the technologies and the ethics that helped us share the dataset and interpret them. While watching and listening to the discussions, we directed our attention to the three dimensions—addressivity, voicing, and semantic permeability of the social theory of language that Skidmore identified in the Bakhtin theory of dialogism. While we looked at the sessions, we focused on the issue of appropriate orientation, which is central in Dialogic Education, and which we conjectured would delineate episodes that are beyond the scope of Dialogic Education. Among the 150 sessions, we chose two sessions that exemplify (in-)appropriate orientation in classroom talk, which we defined as the creation and dissolution of dialogic networks.

First example: revealing the role of design in the creation of dialogic networks with ANT

In the last lesson of the third focus day, the students are arranged in groups of four or five around different tables across the classroom. The teacher distributes to each group an axis drawn on a paper and cards with written arguments regarding the issue at stake, collected in the former disciplinary lessons. The teacher also distributes to each small group a worksheet in which students are asked to organize together the arguments on the axis between supporting mandating vaccination by the government and opposing it. They are thereafter invited to choose where to locate on this axis their shared opinion.

At the beginning of the lesson, the teacher gives the following instructions:

‘‘This activity summarizes everything you have been through today in philosophy, science, and mathematics. I will distribute to each group a board with an axis, which represents the degree to which arguments support or oppose the issue we discuss today—Would it be right for the government to mandate parents to vaccinate their children? You will be provided the cards you already produced in the previous sessions in the different disciplines. We will discuss this question in this summary section using all the information you gathered along the different lessons. You should already have two cards of reasons and arguments pro or against vaccination. So, you should pinpoint your card somewhere on this axis. In addition, I will give you cards with new arguments, and you should decide as a group where to place them on the axis.’’

The instructions presented by the teacher are an outcome of a previous activity—a preparatory meeting between researchers for designing the educational activity (see Koichu et al., 2022). The researchers’ goal was to facilitate dialogic argumentation among students by asking them to map written arguments (text cards) using a Bristol board diagram. They aimed at fostering dialogues by providing students with different arguments and asking them to analyze them. Moreover, the instructions refer to another former network: the teacher refers to previous activities done on the same day in mathematics, science and philosophy on the issue at stake—Would it be right for the government to mandate parents to vaccinate their children? This context invites discussion between students. In previous activities, students actually participated in animated discussions. In the present episode, the teacher refers to the topic of discussion in a whole-class setting. These instructions are given by the teacher, to the students, refer to a board with an axis, to cards to be pinpointed on the axis.

The instructions direct students’ activity on the production of a material product: The teacher asks the students to produce a diagram on a board and locate different text cards on it. The board and the cards are actants, meaning non-human entities that shape the interactions in the network. Later on, we will see that the actants facilitate the interaction in unexpected ways, which differ from the initial educational goal of promoting dialogue. We present an episode from one group of five girls, after the teacher arranged the students into different groups. It is presented in Table 1. The last column refers to categories of moves according to SEDA that will be referred to in a later section. The excerpt shows that after clarifying the instructions, the group discussed how to design the board:

Table 1 The Bristol board subordinates human actions
Full size table

In this episode, the human actors are the teacher and the students. They explicitly refer to the Bristol board or point at it in all turns. It is an actant, as being the source of their actions. The students refer to the teacher’s instructions (turn 55), that is another actant, as it shapes the interaction upon the students the ‘‘we need to arrange the board’’ in turn 55 and the ‘‘it is on purpose’’ in turn 57. All turns in this excerpt point at stable social and material structures—all students collaborate to ‘‘arrange the board’’ (turn 55). This stability confers to the socio-material structure the label of network. It displays the relationships between the actors in the network. The Bristol board functions as an actor, which almost ‘‘gives orders’’ to the human actors, like in line 56 (‘‘Let’s split it into halves and then we will know’’) as if the board ‘says’ ‘‘split me in two halves and then you will know’’, or like S4’s ‘‘Don’t make dots, make lines. This way’’ in (turn 61) and ‘‘should I write it this way’’ (turn 62), which expresses a lack of own voice and a willingness to comply with the Bristol board’s orders—a non-human actor at the center of the network, which addresses an anonymous audience. It exerts force upon them and subordinates them. We define such a network in which some human actors are subordinated as a non-dialogic network: coercively complying with instructions is not an appropriate orientation according to Dialogic Education. The students do so probably because the instructions—which function as a non-human actor in this network, define the board as the unescapable activity’s outcome.

After 5 minutes during which students went on focusing on filling the board, a student suddenly asked her friends' opinion on mandatory vaccination, as shown in Table 2:

Table 2 A dialogic network in which a group engages in dialogic argumentation
Full size table

This new episode is different from the previous one, in which interactions were ruled by the Bristol board and the instructions which subordinated the students' actions. In contrast, students here express their own opinion–their own voice about the issue: at turn 76, S5 decides to articulate an open question about compulsory vaccination in turn 76. This question lead students not to refer to the board and to express their opinions in turns 77–80 and 82–84. In turn 81, S4 tries to re-focus on the Bristol board diagram and on the classification of text cards. Despite her call, the discussion continues. and the group engages in collaborative argumentation in 85–88: at turn 85, S4 expressed a full argument in favor of non-compulsory vaccination; in turn 86, S5 challenges this argument, and in turn 87, S3 expands this challenge; at turn 88, S4 responds to S5 and S6’s challenges by showing the unacceptable implication of mandatory vaccination. She refers to video clip (‘‘they showed a video-clip from the US and they gave a ten thousand fine for people who did not get vaccinated’’), and by do so she incorporates an actant from a previous lesson into the current small group discussion, using this knowledge to enrich the discussion. Like for the first episode, the social structure is stable: it shows a network. However, in this episode, the network is very different: the 5 human actors show clear agency, refer to non-human actors but exert their power upon them. We call such a network, in which all the human actors have agency, and no actor, human or non-human subordinates them, as dialogic, as the orientation of the learners is appropriate from the vantage point of Dialogic Education—nurtured by collaborative argumentation. The previous (non-dialogic) network was dissolved as one student (S5) decided to end compliance with given instruction, and triggered autonomous dialogic argumentation. This network is maintained for 6 more minutes. In the previous episode, the Bristol and the cards were at the center of the network, whereas now, the students function as the center while creating argumentative discourse. The Bristol board and cards are present in the classroom, yet they are left alone and do not play any role in the network. As will be shown in Table 3, it dissolves itself, again in favor of the fulfillment of a task:

Table 3 The dissolution of a dialogic network and the formation of a non-dialogic network
Full size table

This excerpt demonstrates how easily a network can be dissolved, as interactions are constantly transformed, reshaped and reassembled. Up to turn 157, students discuss the issue at stake. S3’s try in turn 149, to bring her friends back to the fulfillment of the task, remains unnoticed. In the dialogic network in which they evolve, the agency of the students is salient. They exert power upon each other in a distributed way, and the group exerts power upon the non-human actors (the Bristol board, the cards). In turn 157, S3 calls her peers to ‘‘start working’’, namely classifying text cards on the axis on the board. At that stage, the board, and the text cards exert their power upon the students. Turns 158–180, that are not transcribed, are moments of confusion, between the drive to continue the discussion and the appeal for compliance to what needs to be done. Turn 184 definitely ends the argumentative dialogue, as S4 suggests, ‘‘we have to start working and don’t say our opinions otherwise, we will not work.’’ This utterance stresses a move away from voicing, and that the students’ work should address the text cards as their rulers. The students reorganize themselves and allocate roles and responsibilities for reading the cards, and writing on the Bristol board. Consequently, the students obediently read the reasons on text cards and decide whether they support or oppose compulsory vaccination (turns 187–204). Their decisions are immediate and do not lead to further discussion (turns 209–210). Turn 157 is then the precursor of the shift of networks. The different and sometimes contradicting instructions, the board, and the text cards exert power and lead the students to end their collaborative argumentation. There is a shift of networks as the dialogic network is replaced by a less dialogic one. In this shift, students’ agency abruptly changes from strong in collaborative argumentation as students elaborate arguments and challenge each other, to weak agency when the making of an artifact dictated by non-human actors makes students’ opinions and judgments irrelevant. It is quite telling that in this shift towards the mere classification of cards, one of the students sarcastically says ‘‘our opinions do not matter’’ (turn 185). These two episodes exemplify that human agency can subordinate artifacts, and can be subordinated by artifacts in different contexts, especially by educational designs.

At a first glance, the shift from a non-dialogic to a dialogic network and again to a non-dialogic network may look like a failure of educators dedicated to Dialogic Education. However, a more precise ANT-based analysis of what happened during the focus day gives some cues about this surprising alternance. We cannot show this analysis here but the observation of dozens of classroom hours with these students, evidenced that the teachers worked very hard in instilling dialogic norms by fulfilling the instructions. The teachers traveled between the working groups and often stressed the importance of fulfilling all instructions (that were designed by the researchers). On the other hand, students complied with these requirements in order to satisfy the teachers and to be recognized as good students. Although we did not encourage the establishment of this norm, we confess that in our program, students prioritized the teachers' demands. The Bristol board subordinated the students because their main goal was to fulfill the teacher’s instructions, instead of carrying out a dialogic discussion.

We suggest that there is another reason for the alternance between dialogic and non-dialogic networks, that points at one failure of the researchers and of the designers. They envisioned that through the classification of cards, students would initiate discussions around possible positions on the issue of mandatory vaccinations. They assumed that reading the text cards would enrich the students’ dialogue and open it into new directions. They also wanted students to produce artifacts to be used in further discussions. They perhaps wanted too much. Too many instructions, too many tasks to do, impair dialogue. Indeed, turns 147–210 exemplify an inexorable phenomenon that Latour termed as ‘‘Black-Boxing’’—a process at the end of which dynamic events and negotiations solidify and become durable. The durability is possible because the black-boxing conceals the dynamics and circumstances in which it was produced. The dialogic network that resulted in collaborative argumentation solidified into compliance to a set of instructions that dissolved the small group network. The classification of the text cards black-boxed the network into the production of a material product to be used in further activities without reference to the network that preceded its production. Therefore, instead of promoting a richer dialogue, the students’ interpretation of instructions led to a different network where instructions and boards ruled interactions. We will return to the issue of black-boxing in Dialogic Education in the discussion.

After the small groups ended the filling of the Bristol boards, the teacher initiated a whole class discussion, in which she asked each group to present their position, locate it on the Bristol board’s support-oppose axis, and explain their position. Table 4 shows how the group of students whose interactions were previously described stands in front of the class to present their arguments:

Table 4 A network in which the teacher subordinates students' actions, but students deviate from the instructions
Full size table

The episode in Table 4 shows that the teacher is a central actor that exerts her power upon the other actors-students: she sets the goals of the presentations and invites groups to present their positions (at turn 1766), organizes groups (in turn 1779), recalls students to mark their positions on the Bristol board (at turns 1792, 1787), rephrases the arguments of the students (at turn 1817), and asks for classifications (turn 1821). Every time the teacher calls another group to present. The teacher decides who will speak and when. Unlike the previous network, although the students present their arguments, and challenges that arose during their previous collaborative argumentation, the presentation is a mere juxtaposition of different arguments under the baton of the teacher. As appears for example in turns 1785 (when S2 claims ‘‘S5 has another opinion’’) and 1786, students are not totally subordinated to the teacher, but the behavior of the teacher suggests that she is subordinated to the instructions! The episode shows a network with actors that interact according to the same pattern, but the network is hardly dialogic.

The Bristol board continues to be an actor in this network, yet its role is very different from in the previous network, where we demonstrated how the Bristol board black-boxed the network’s dynamics by subordinating the students’ actions to the mere placement of cards on an axis. This black-boxing shifted the previous dialogic network. In the present episode too, the Bristol Board is a central actant in a network which is first non-dialogical, but here it is the teacher who subordinates the students through the use of the Bristol board: The students open by presenting their position (1779); the teacher interrupts them and asks them to mark their position on the Bristol board (1782). Nevertheless, the control of the teacher through the Bristol board does not lead to a collapse into a totally non-dialogic network: in turns 1785–1989 students illustrate with the board how their positions differ from each other. Somehow, the students ‘‘reopen the Black Box’’, by deviating from the instructions and from the subordination to the Bristol board. In the small group discussion, the Bristol Board was treated as part of to-do task instructions that contradict the researchers' educational goal of facilitating dialogue: it appears that the designers figured out that positioning opinions on a Bristol board would facilitate dialogue, but the teacher turned this instruction into an ‘‘order’’. During the whole class presentations, it returned to fulfill its original aim as a tool for promoting dialogue and, in spite of the control of the teacher, enabled students to express their respective positions. This process of the Bristol board being Black Boxed and taken for granted and then reopened, demonstrates how an actant may exert power, promote, or hinder dialogue in the deployment of different networks.

Second example: translations and ANT’s ability to identify transformations in classroom talk

In one of the classes, the teacher began the last lesson of the fourth focus day by giving instructions. She first asked one of the students to read a text which describes a future scenario of genetic modification of embryos and its consequences in the medical, psychological and sociological realm, and then said: ‘‘What I ask you right now is to split into two groups: supporting and opposing [genetic modifications of embryos]. You can use the futurist scenario [to support your position]. Think about what kinds of reasons, supporting and opposing it provides. There are things here that can help you. What I ask you to do is to split into two different groups and write arguments supporting and opposing [genetic modification of embryos]’’. Table 5 will show that the beginning of the discussion in the group assigned to support genetic modifications deploys itself in a dialogic network, in which the students present different reasons, weigh them, and challenge each other:

Table 5 The stability of a non-dialogic network
Full size table

After the teacher (T1) gave the instructions above, she arranged two groups of students. Table 5 shows the discussion of one group of four children (S1-4). The teacher, as well as another teacher (T2), who acted as the helper of T1, followed the discussions of the 2 groups. Several actors appear in this protocol—not only the 4 students, and the teachers, but also the instructions. In this episode also, the relations between actors show some stability—the maintenance of a network: Students constantly refer to the instructions, like at turn 4, when S1 asks ‘‘Wait, [do we] support or oppose? or at turn, 6, as S11’s wonders ‘‘Can we claim against (genetic modifications of embryos)?’’. T2’s answer ‘‘No’’ at turn 12 shows that the instructions are a central actor, and that the voice of the students is limited. From turn 4 to turn 14, the students try to understand the written instructions (lines 4–14). The instructions exert power upon the students, who understand that they should bring supporting reasons for genetic engineering (turn 5). S1’s claim in turn 13 ‘‘But I want to claim against’’ which is followed by T2’s answer ‘‘We think one thing, but we should (justify the case supporting genetic modifications)’’ shows two things: First, the students comply with the instructions, although these instructions force them not to express their own opinions. Secondly, the teachers are the gatekeepers of these instructions. At Turn 14, T2 is compassionate with students’ difficulties in refraining from expressing their opinions. She marks the dialogue’s borders and shapes the students’ agency in this network—they are obligated to support a certain position. They are agents for building the reasons by themselves, although they cannot decide or change the activity's instructions. Their agency is not neglectable. This network can then be considered dialogic, although the instructions and the teacher exert considerable power upon the students.

The subsequent turns show that from this network emerges a collective argumentation, in which S2 presents a supportive reason for genetic engineering (people will be healthier and save money—turn 15). Then, S3, S2, and T2 support and challenge this reason (turns 16–23). In this protocol, T2 acts as a participant, not as a gatekeeper of the instructions (lines 16 and 26). After 12 min, one of the students stops the discussion, as shown in Table 6.

Table 6 A new dialogc network in which students participate in dialogic argumentation
Full size table

S4’s invitation in turn 102 ‘‘Girls, let’s her the arguments we made so far, because what we say is pretty much different versions of the same arguments. Let’s look for new arguments ’’, delineates a new social structure, a new network. Cards on which arguments are written are actants (in turn 103), which refer to previous outcomes (‘‘I will say the things I wrote’’ in turn 102) and future expected networks (‘‘they will challenge us’’ in turn 105). All students show agency as dialogic argumentation is deployed in this network: S4 suggests a new reason (genetic engineering improves children’s mental health). S2 challenges S4’s statement while supporting her refutation of the argument written on a card (Turn 103). S1’s move incorporates the text card into the network. S4 answers the challenge by saying that they are supposed to support genetically modified embryos. By doing so, she echoes T2’s instruction (at Turn 14). S2 responds that they should consider counterarguments that they will face from other groups in classroom discussion (‘‘they will challenge us’’) (Turn 105)—she incorporates future possible networks (whole classroom discussion) into consideration. The students went on examining and challenging each other’s ideas, and their agency was salient. After 15 min of dialogic argumentation in the same network, T1 joins the conversation, as described in Table 7:

Table 7 The translation of teachers as interlocutors whose role can be challenged
Full size table

This excerpt demonstrates the unexpected dynamics of dialogic networks—networks in which all human actors are not subordinated to one actor, human or non-human. T1’s question in turn 129 is intended to refute critiques about genetic engineering based on a comparison of human and food genetic engineering, ineluctably leading the students to the conclusion that people should not fear genetic engineering (as alluded to in T1’s elaboration in 131, and expanded by T2 in 144). One might expect that this move would reduce the students’ agency. However, S1’s interventions changed the power relations that T1 and T2 tried to exert upon the students (Turns 145, 154 and 156). She challenges T2’s position according to which genetic modifications were never experienced on humans’ embryos because humans have a different genetic code than vegetables. By doing so, she put herself as an equal contributor as the teacher. S1’s challenge of T2’s position evidences semantic permeability as T2’s position refers to a session in biology. To T2’s counterargument ‘‘You know that one of the opponents’ arguments against genetically modified embryos is the fear of technology’’ at turn 153, S1 counterargues ‘‘ = but we never tried it on humans ’’. In this short series of turns, the role of the teacher is translated and she no longer tries to dictate her view; the students argue with the teachers as equals. Yet this social structure is fragile. A few seconds later, following her friends’ request (turn 158), S1 writes the teacher’s assertion on the board (turn 159), despite her reservation. We can see the tension between authentic dialogue, where the student argues against the teacher, and the following students’ actions to complete the task according to the teacher’s satisfaction, thereby reducing their agency.

This example shows that roles cannot be prescribed in advance in networks in which educational dialogues deploy themselves. Entities, human and non-human, are translated in specific contexts, e.g., changed through to the process of becoming part of a collective network (Latour 1987; Fenwick & Edwards, 2010). The role of the teacher, of the students, and actants emerge from their interactions and assemble in specific networks. The title ‘teacher’ or ‘student’ does not convey a pre-given role; it is translated by actions, connections, and changes made in a specific network. The second example demonstrates the tension in the process of translation. The teacher acted as an authority that wished to transfer knowledge (by such to limit semantic permeability); however, as she joined an existing discussion where agency was in the hands of students, S1 did not see her as a source of knowledge but as an interlocutor whose position could be challenged.

The comparison between T1 and T2’s actions and interactions, enables deeper exploration of the idea of translation. We already saw how T2 clarified the activity’s instruction for the students (Turns 9, 12). Further on, her actions become similar to the students’ actions: she supports and challenges other students' statements (Turns 16, 25). She is one of many voices and does not function as the center of the network. She is not in charge of the classroom and functions as a helper. In contrast, when T1 joins the conversation, she immediately positions herself at the center (Turns 129, 144, 153): she controls turn taking and pushes the students towards a certain answer. She wants to comply with the instructions, and to finish the lesson.

The role of T1 remains the same—she tries to be the center of the network, but the students partly succeed in opposing her will to control. In contrast with the first example in which the educational design (the Bristol board, the cards, the instructions) dissolved at some point a dialogic network, here the dialogic network is maintained with the help of the educational design and of a human actor (T2), who confers more agency to the students. This example shows how ANT enables describing transformations in classroom talk by looking at the translations in the roles of the actors in networks.

The analysis of the two discussions according to SEDA

ANT helped identify the deployment of intervals in the form of networks that were dialogic or non-dialogic. As mentioned before, SEDA explores a different timescale as it identifies dialogic moves. The scheme consists of 33 dialogic moves that are organized in eight clusters. This analysis relies on the eight general categories of the SEDA methodology: Invite elaboration or reasoning (I), Make reasoning explicit (R), Build on ideas (B), Express or invite ideas (E), Positioning and coordination (P), Reflect on dialogue or activity (RD), (G) guide direction of dialogue or activity and Connect (C). We decided to use the eight clusters for coding, and not the full 33 codes scheme because it increases practicality and affords a higher inter-rater reliability. This method is consistent with other studies that applied SEDA for analyzing educational dialogues (Firer et al., 2021; Vrikki et al., 2019). Table 8 displays the SEDA analysis of the two discussions described in the previous section.

Table 8 The distribution of dialogic moves according to the SEDA categories
Full size table

The SEDA analysis reveals the distribution of dialogic moves according to eight communicative categories. Table 8 shows that the second group discussion included more dialogic moves than the first group discussion, although it was shorter. 29% of the communicative acts were not coded as dialogic. In the first discussion, 52% of the communicative acts were not coded as dialogic. The first discussion, which lasted for almost 29 min, included few expressions or invitation of ideas (E, 8%), building on ideas, giving examples (B), reasoning—presenting justifications and explanations (R) (B + R = 11%),Footnote 1 and positioning and relating to other positions (P, 14%). In contrast, the second discussion included much more moves in these categories (E, 21%; B and R, 18%); P, 16%). Moreover, the discussion also contained dialectic (as well as dialogic) moves as the students based on other students’ utterances and challenged them. In addition, the dialogic moves were uniformly distributed in the second discussion, in contrast with the first discussion in which the dialogic moves concentrated in a few minutes (3:30–10:00). The large difference between the moves that were not coded as dialogic in the first discussion (52%) and in the second discussion (29%) intuitively suggests that the second discussion was ‘‘more dialogic’’ than the first—a fact which sounds reasonable in the light of the description of both discussions. However, the relatively large percentage of moves that were left uncoded with respect to dialogic moves invites a reflection on the SEDA methodology, which will be developed in the next section. More generally, we will undertake a comparison between the SEDA and the ANT methodologies, from which we will draw insights about the analysis of educational dialogues.

Comparing the SEDA and the ANT analyses towards a better comprehension of the deployment of educational dialogues

We used the SEDA and ANT analyses to inquire about the dialogicity of the episodes we presented. We could see that ANT enables describing these episodes as networks that form and dissolve themselves, some of them being dialogic—in which no actor subordinates human actors, and some non-dialogic, for example when to-do task instructions are actors that dissolve dialogic networks. SEDA focuses on dialogic moves and shows their frequency in each episode. The temporal dimension of ANT is salient, but it is not totally absent in the SEDA analysis. For example, in Table 7, we see that T2 starts with a build-up question for the discussion (Turn 129). However, this question does not yield to any reasoning: the following utterance was coded as E (making other relevant contributions, turn 130). Then, T2 provides information she thought to be helpful, since it suggests that people should not fear genetic modifications of embryos (Turn 131). We coded this utterance as R (Make reasoning explicit) accordingly. However, this invitation was not taken up until turn 144 (S2’s utterance ‘‘I don’t understand how it is related’’ at turn 137 is most telling in this matter). Most of the utterances until Turn 144 were labeled as uncoded (UN). The fact that these utterances are labeled as UN although the students are quite responsive to the teacher and the task, provides evidence that SEDA provides codes for talk in which something is learned. T2’s turn in 144 makes explicit the role of this information and in turn 146, S4 is very responsive (‘‘It really does. If genetically modified fruits come out so tasty. Why shouldn’t we?’’. This turn was coded as P + E, accordingly. The subsequent turns were coded as B, R, P + R, P, P, G, E and reflect what looks like a productive collective argumentation. T2’s turn in 154 (‘‘But we never tried it on humans! Vegetables, fruits, and humans have a different genetic structure.’’) as P + E (positioning, challenging viewpoint) when S1 challenged the teacher. Turn 155 was coded as G (guide directions for the dialogue) since T2 reminded S1 of the activity’s instructions (Yet, we had doubts as to whether we should code it as P—positioning, since the teacher relates to Noa’s position). Therefore, the distribution of SEDA codes gives a fine-grained description of relations and ideas among participants. It delineates dialogic chains and discerns slight shifts from dialogic to non-dialogic episodes (in what the ANT analysis described as a dialogic network). In contrast with the ANT analysis, the SEDA analysis is blind to power relations and indifferent to the identity of the speaker, it does not capture the tensions, agency, and the roles students and teachers play in a dialogue. It does not capture, like ANT, translation processes. It does not discern between teachers acting as an authority and teachers that act as participants with the same status as students. Despite its advantages, SEDA is blind to agency, power relations and non-human actors. It also does not provide detailed data of how and why dialogues emerge, persist and cease in classroom discourse.

Table 3 in the section on the ANT analysis provides other details about the differences between the two methodologies. The ANT analysis described how the discussion was laid in a dialogic network and shifted to a discussion laid in a non-dialogic network in which students completed the task of organizing text cards on the Bristol board. The SEDA analysis detects this change by displaying turns according to dialogic and non-dialogic categories. The ‘‘Uncoded’’ category is the most frequent category in this episode. At the beginning of the discussion, the students are busy presenting their arguments, and their turns are coded as R, R, E and E (turns 147, 148, 150, and 151). Turn 149 is a precursor of a non-dialogic turn (‘‘Do you want to begin with that (organizing the cards)?’’), followed by a long series of ‘‘uncoded’’ turns around technical issues about writing and designing the Bristol board (e.g., ‘‘please copy it’’ in turn 214) or instructions that inhibit dialogue (e.g., ‘‘We have to start working’’ at turn 158). The code G (guide direction of dialogue or activity) which deals with the reading and analysis of arguments (‘‘let’s organize it [the cards on the board]’’ turn 181, and similarly turns 184–186) shows the peremptory character of the discussion.

SEDA seems to help describe shifts from less dialogic to more dialogic episodes but does not explain the mechanisms that lead to these shifts. In contrast, ANT provides an explanatory framework. Shifts from one network to other convey the change of power relations between the different actors, and the determining role of non-human actors. Indeed, in the example displayed in Table 3, we showed how the dialogic network was dissolved: various educational designs subordinated the students, by requiring them to focus on producing material products—organizing the text cards on the Bristol board. SEDA shows how ideas are constructed. ANT does not. SEDA is blind to the role of non-human actors in educational activities. ANT emphasizes how these actants shape its dynamics of networks in which educational dialogues lay.

We claim that an important additional insight can be drawn from the comparison between ANT and SEDA. In the episode described in Table 4, we identified a less dialogic network because of the absolute control of the turns of the students by T1 (e.g., ‘‘First of all, S3, sorry I am stopping you. Where do you locate (your position on the axis)?’’ at turn 1766 or a similar interjection at turn 1782). However, the SEDA coded categories are ‘‘dialogic’’: E for S4 at turn 1789, P + R for S5 at turn 1797, C + R for S2 at turn 1798, etc. There is an opposition here between the two methodologies. This opposition conveys an idea we already mentioned: SEDA, when applied on one or few dialogues, focuses on the circulation and development of ideas among participants. ANT focuses on power relations. SEDA points at an ideational dimension—the ways ideas are communicated and co-constructed. It is a central tenet of Dialogic Education. ANT handles the tension between the theory of dialogism and Dialogic Education. It identifies dialogic and non-dialogic networks through the observation of the subordination of human actors. It provides an interesting response to the post-humanist criticism of the anthropocentric character of Dialogic pedagogies that (Davies & Renshaw, 2019) expressed. ANT gives a central place to relations (in the forms of networks), and which involves materiality. Its application to classroom talk helps discern two kinds of networks, dialogic and non-dialogic. The subordination to non-human actors does not fit the scope of Dialogic Education. The application of ANT to Dialogic Education recognizes that the power that non-humans exert on humans can be tolerated, if it does not reach subordination.

Discussion

In this paper, we have contrasted the use of two methodologies for tracing the implementation of a dialogic pedagogy in classroom talk. The term ‘‘educational dialogues’’ concatenates dialogic intentions that are often vaguely articulated, and a design that is planned to afford the deployment of dialogues. While intention and design are important, the actual deployment of the talk in classrooms is a very complex process. We used two methodologies to observe this talk and specifically to discern the dialogic from the non-dialogic (or the more dialogic from the less dialogic, to remain faithful to the theory of dialogism). We claim that this discernment handles the tension we mentioned between Dialogism and Dialogic Education: ANT relies on the identification of networks that constantly assemble, dissolve and reassemble, in which human and non-human actors exert force and agency on each other. The application of this classical feature of the ANT methodology in the realm of classroom talk provides a socio-material structure to classroom talk, and by such a tool that makes salient fundamental elements of the Theory of Dialogism. The symmetry between actors fits the idea of reciprocity in the theory of dialogism. The network structure fits the fact that in the theory of Dialogism, every utterance is connected to another one, in order to constitute an organized stream in which people respond to one another, and is interrelated to a past conversation. The importance of non-human actors (instructions, diagrams) that exert power upon other actors in networks fits the post-modern extension of Dialogism. Our delineation of dialogic and non-dialogic networks situates Dialogic Education within Dialogism: Our definition of non-dialogic networks as networks in which some human actors are subordinated to others (human or non-human) points at an inappropriate orientation from the vantage point of Dialogic Education, yet indicates that these are still networks, a fact which positions Dialogic Education within Dialogism. The maintenance of dialogic networks appeared to be fragile: They may assemble, due to instructional designs, and paradoxically dissolve because of these very designs. The ANT methodology stressed the importance of artifacts, instructions, texts, or various tools in inhibiting or catalyzing the deployment of dialogues. These artifacts were designed to facilitate the deployment of educational dialogues. But they may play as actors whose agency interrupts this deployment. Although it is impossible to foresee exactly this deployment, the ANT methodology suggests that designers should plan settings meticulously in order not to impair the flow of educational dialogues.

The translation process enabled us to show that actors have not a fixed predetermined role. Rather, this role emerges in specific networks that assemble or dissolve: The role of the teacher or of the students varied in the protocols we presented. The teacher who was a peremptory figure became a facilitator of dialogues; students alternatively complied with instructions or opposed the injections of the teacher. The Bristol, that was designed as actant to facilitate dialogue, became an actor that mandated from the students to locate cards on an axis and inhibited the formation of a dialogic network. We claim that the context of educational dialogues, which is a context of educational change, is characterized by frequent passages, back and forth, between traditional and novel roles. At the time of educational changes during which teachers modify traditional practices, a methodology that is sensitive to the role that teachers play is crucial. We confess, however, that the hic et nunc observation of roles by the ANT methodology misses the fact that roles are informed by experience.

The black-boxing of networks—their solidification and use in further activities whose reliability is absolute seemed a priori a sign of non-dialogic network. Such black-boxing moments were not infrequent in our analyses. Indeed, black-boxing indicated a non-dialogic moment, but, in the context of consecutive dialogues, black-boxing appeared to help in the mobility of learned ideas from one activity to another. Moreover, the black-boxing was sometimes reversed in further activities and the products of the black-boxing turned to cues for the emergence of a new dialogue—necessary nodes for the creation of a new network. We claim that this is an essential characteristic of educational talk: the networks in which they lay may be black-boxed to comply with classroom constraints (time, complexity) but may be reopened in further interactions.

Although the analysis of several protocols is not sufficient to validate the usefulness of the ANT methodology in research on educational dialogues, the several phenomena that recurred in the DIALOGOS project open promising avenues in the analysis of classroom talk. At any rate, it is obvious that the ANT methodology enables discernment of the more dialogic from the less dialogic in classroom talk, by identifying networks in which some relations are less or more subordinated.

The application of SEDA on the classroom talks we analyzed with ANT shows the strengths and the limits of both methodologies. First, the fact that moves can be identified as dialogic, although they do not belong to what ANT identifies as a dialogic network. This apparent misfit suggests that micro- and meso- levels of analysis show different pictures. We conjecture that these two pictures are not orthogonal, though: Impoverished dialogic moves when learning is at stake points to the absence of dialogic networks. Anyway, the misfit between the methodologies originates from the fact that SEDA traces the circulation and development of ideas, in contrast with ANT, which observes the distribution of power among actors.

Another important finding of our SEDA-based analysis is that even in a piece of talk in which a dialogic network deploys itself, dialogic moves that belong to the SEDA categories are scattered. These findings do not diminish the many advantages of SEDA, among them the possibility to compare corpuses of talk from different classes, schools, or programs of intervention. ANT is a method which provides, in the context of classroom talk, ad hoc descriptions of formations and dissolutions of (dialogic) networks.

Pedagogical implications of the adoption of ANT for analyzing classroom talk

In this paper, we have mentioned the advantages and the weaknesses of the SEDA methodology for analyzing classroom talk. In the pedagogical realm, the SEDA methodology provides a clear advantage: In-service teacher programs that foster the moves that SEDA identifies as constituents of Dialogic Teaching (such as Invitation to reasoning), are effective from two perspectives. First, promoting the moves that SEDA identifies as constituent of Dialogic Teaching in PD programs increases the quantity of such moves in classroom talk (Vrikki et al., 2019). Secondly, the presence of such moves predicts learning outcomes (Howe et al., 2019). The insertion of SEDA constituents in PD programs dedicated to Dialogic Education is then highly productive.

In this section, we conjecture that like SEDA, the ANT methodology provides profound ideas for PD programs. First, the PD program may convey the general idea that what happens in classrooms can be figured out as a network of actors that interact with each other in complex ways. The second idea is that there are more dialogic and less dialogic networks. Networks are more dialogic when human actors are not subordinated to other actors. Another idea is that the emergence of dialogic networks can be designed but dialogic networks are very brittle and their maintenance demands a high level of effort, both in the design of tasks and in the interventions of the teacher. Otherwise, the dialogic network can be dissolved at any moment. Teachers should be aware of the fact that design, meticulous it may be, risks leading them and their students to obey instructions, instead of enabling the deployment of educational dialogues. We conjecture that conveying these ideas may invite the teacher to give liberty to other actors to interact without dominating their interaction. Another idea drawn from the ANT methodology, and that can be conveyed to teachers, is that dialogues involve not only human actors but non-human actors. These non-human actants may facilitate or inhibit the maintenance of dialogic networks. The technologies whose use is progressively growing in classrooms should bring a great variety of actants in dialogic networks. An additional idea drawn from the ANT methodology is the idea of translation—the fact that actors may change the role they play, and this translation may generate/maintain or dissolve a dialogue as a network. The idea of black-boxing whose character is generally irreversible, is replaced in a dialogic context by a dynamic idea in which black-boxing allows further provision of resources through boundary crossing between successive dialogues. Also, we saw how dialogues were black-boxed in simplistic conclusions or diagrams, can be re-opened later on for the regeneration of the dialogue.

In a nutshell, both SEDA and ANT methodologies provide ‘‘grammars’’ to talk about Educational Dialogues. Instead of designating and evaluating them in general and vague terms, these methodologies provide tools for teachers for identifying moves (with SEDA) and episodes (with ANT) during which the (non-)dialogic can be delineated. SEDA enables tracing the chaining of ideas in talk. The ANT methodology brings to the front stage the role of instructional design in facilitating or inhibiting dialogic networks. ANT problematizes interactions in classrooms. It explains to teachers and researchers that interactions encompass multiple levels of talk and of agency. Therefore, it invites designers to reflect on what facilitates or inhibits educational dialogues, beyond the role of authentic questions or of uptake.