Introduction

The four papers in this issue all bear on the driving force of CSCL: tool-mediated dialogue. Here, dialogue is not only a process of knowledge co-elaboration, going hand in hand with maintaining a collaborative working relation, but is also considered as an object for further reflection, whether individual or dialogical. This forms the basis of our comments on the first pair of papers (Slakmon & Abdu, 2024; Beal & Steir, 2024), under the rubric ‘dialogues on dialogues’. A second theme that we want to pick up on is temporal analysis. Two papers in this issue include sequence analysis as part of their data analysis (Paneth et al., 2024; Yang et al., 2024). These are discussed in the third section below. Our concluding remarks explore emerging directions for CSCL research, instigated by the four papers in this issue.

Dialogues on dialogues

Slakmon and Abdu (2024), in their paper “Learning to notice collaboration: examining the impact of professional development on mathematics teachers’ enhanced awareness in CSCL settings”, revitalise the concept of teachers’ monitoring of students’ activities. This was a research topic from the beginning of the use of AI techniques in generating educational dialogues. Thus, the “coaching” system produced by Burton and Brown (1982), extending earlier work on “Socratic tutoring” (Collins, 1977), formulated the problem of coaching as determining when to intervene and what to say (choice of content and pedagogical strategy). The ‘when’ question concerned not only identifying (noticing?) mismatches between a normative model of the teaching domain and the students’ problem-solving activity (inefficient, incorrect procedures, for example), but also the timing of the intervention. The idea was to avoid interrupting ongoing and possibly productive processes: the system, the teacher, should wait for the appropriate moment when the students can be attentive and receptive. This is still a problem for teachers who have to manage several small groups in the classroom, given that when they arrive at a group in order to intervene, they usually lack the group’s dialogue history. We believe this to be still an open and interesting research question.

How is noticing different from monitoring, from identifying potential moments for providing guidance to students? Slakmon and Abdu point out that “[t]he concept of noticing sits at the intersection of three theoretical frameworks: collaborative learning, computer-supported collaborative learning (CSCL), and dialogic pedagogy.” They state that “[n]oticing contributes to “monitoring” by delving into its challenges in greater detail, making sense of learning interactions”, involving trying “to understand how teachers attend to learning events, how they interpret what they see, what their interpretation is based on, and how the interpretations turn into responses.”.

Teachers’ noticing in well-defined knowledge domains, such as mathematics and sciences, is one thing, since a clear epistemic norm is available in the background. But noticing with respect to processes in learning interactions is somewhat different: fundamentally, it must be interpretative. To our knowledge, there is no generally accepted theory and model of dialogue, including in education. Levinson (1983) introduced the classic distinction between DA (Discourse Analysis) and CA (Conversation Analysis), that still obtains today. Different models of dialogue (verbal interaction, talk, conversation) would lead to noticing somewhat different phenomena; and what is crucial is to show how they relate to knowledge elaboration, meaning-making, and learning. An interesting multidimensional analytical model for evaluating the quality of collaboration, that could be used to decide what to notice, has been proposed in the present journal, ijCSCL, by Meier et al. (2007). The choice of Slakmon and Abdu is to provide teachers with a Bakhtinian model to frame the teachers’ activity of noticing, and that is of course a perfectly valid and acceptable choice, grounded in dialogic education (Mercer et al., 2020). Notwithstanding, given the diversity of theoretical viewpoints on dialogue, we propose that the interpretative process of noticing in dialogue should itself be dialogical, involving teachers, certainly, but also researchers and students themselves in certain situations. Finally, with respect to this paper, it is important to note that a clear conception and evaluation of the teachers’ learning, qua professional development, is described, in terms of an improvement of the ability to notice specific dialogic aspects.

The paper of Beal and Steier (2024), entitled “Dialogues across time and space in a video-based collaborative learning environment” takes the ‘dialogue as object of reflection’ concept further, in that the reflective activity is itself a dialogue and a dialogue on dialogue. There is no a priori reason why such an iterative process of dialogues on dialogues should end. In a sense, this is a form of iterative vicarious learning, learning from observing the learning of others in dialogue (see Stenning et al., 1999), but where observation is an active and collaborative activity. Beal and Steier’s study (ibid.) investigated the emergence of pre-service teachers’ dialogues oriented towards preparing for exams in a pedagogy course. The dialogues in question were video-recorded, which provides a rich source of dialogic experience, providing the teacher-learners with not only verbal information, as in typewritten CHAT, but also all the verbal, paraverbal, non-verbal and affective dimensions required for deeper understanding. Without attempting to summarise the details of the study, we highlight two main points: tool mediation and temporalities.

Two technological and semiotic tools are used here: videos of dialogues, that were exchanged between groups (whose dialogues on the dialogue videos were further exchanged), and graphical mind-maps that recorded key topics in the curriculum. Qualitative analyses illustrate how these tools effectively facilitate and shape the interactions within and across small groups. Secondly, the intertwined networks of dialogues, mediated by the video-based mind-maps open up the hic et nunc dialogue situation to other temporalities (see Lemke, 2000; Ludvigsen et al., 2011), past and future. This needs to be understood in a Bakhtinian sense, in terms of the circulation of other voices in a present dialogue, from dialogues that have already taken place in the past, and those that are ‘pointed to’ in the future.

We may also ask: what is learning, in this case? Could it be found in an enhanced degree of dialogicity, the increased multiplicity of voices that circulate along different timescales? In what Beal and Steier describe as “nonlinear meaning making”? These are intriguing possibilities that will require further investigation.

(Con-)Sequential data analysis

The contribution by Paneth and co-authors entitled “Zooming in: The role of nonverbal behaviour in sensing the quality of collaborative group engagement” provides an in-depth analysis of the role of nonverbal behaviours in collaborative group engagement (CGE) within computer-supported collaborative learning. Using a multi-method approach, including quantitative analysis and qualitative case studies, the study examines how nonverbal cues such as nodding, laughing, and eye contact correlate with different dimensions of CGE—behavioural, social, cognitive, and conceptual-to-consequential engagement.

Amongst the methods employed in this paper is (narrative) sequence analysis, whose use, with other forms of temporal analysis, is of growing importance in CSCL (Chiu & Reimann, 2021; Lämsä et al., 2021). Process mining, for instance, is frequently employed (Saqr & López-Pernas, 2023; Sobocinski et al., 2017; Song et al., 2022). Lag-sequential analysis is used in (Kapur, 2011) and (Yang et al., 2024), for instance; an example of statistical discourse analysis is (Molenaar & Chiu, 2014).

In Paneth et al., the events that follow each other are demarcated by turns in verbal communication, accompanied by non-verbal behaviour (coded into five categories). Whilst temporal analysis fulfils only a secondary purpose in this study — the focus is on non-verbal behaviour and its relation to the quality of CGE — it is nevertheless conducted rigorously, based on a thoughtful selection of cases and conversation snippets from a much larger corpus. Technically speaking, the sequence analysis method is qualitative-narrative, characterised by a high degree of attention to detail and only ‘mild’ compressions into patterns and generalisations.

From a methodological perspective — and paraphrasing the question raised in Lämä et al. (2021) — what are we doing when we provide a narrative account of an event sequence occurring in a single case or a small number of cases? Firstly, and importantly, we are demarcating the beginning and end of events that can be tied together, such as critical events in a group discussion. Secondly, we are providing a causal explanation for the event sequence. A well-formed narrative description is formally equivalent to a directed graph in mathematics, with the following components: (1) a finite set of descriptive states of the world (W); (2) a weak order in time on W (the chronology of states); (3) a finite set of actors A, which may be individuals or collectives; (4) a binary causal relation between some pairs in W, running from earlier to later states (each such pair can be referred to as an event); (5) a finite set of actions that transform some elements of W (Abell, 2004, p. 289). And since the observed actions performed by human actors obey the laws of physics, an action can be treated as the cause of an event (singular causation; Psillos, 2002). The problem lies in distinguishing consequence from mere sequence: how do we know that condition C affects E in a particular case, rather than C preceding E? (Abell, 2004, p. 294). Since a narrative account cannot rely on frequencies to discern sequence from cause — more precisely, from reason — it needs to rely on human agency and intentionality: the activities in the chain of events need to be described as means to an end, i.e. what free agents would do to bring about one event as the means to bring about another event (Menzies & Price, 1993).

On the other hand, the sequence analysis reported in the paper by Yung et al. with the title “Reflective assessment using analytics and artifacts for scaffolding knowledge building competencies among undergraduate students” does make use of frequencies — of ‘large numbers’ — to discern systematic from potentially ‘random’ sequences. This study explores reflective assessment, analytics, and artifacts to scaffold knowledge building competencies among undergraduate students. The study defines knowledge building competencies as epistemic, metacognitive, and conceptual competencies essential for the sustained pursuit of knowledge and inquiry. It introduces a framework for analysing KB competencies using discourse moves such as creating epistemic inquiry, alleviating lack of knowledge, negotiating fit, and synthesising community ideas. The discourse moves are, amongst other forms of analysis, subjected to a lag-sequential analysis. This is because it was expected that the pedagogical and technical elements that, in combination, constitute the main intervention (reflective assessment) would influence the frequency distribution and the temporal order of discourse moves. The method used is lag-sequential analysis; the analysis is confined to lag-1, i.e. to transitions between two discourse moves.

A lag-sequential analysis’s contribution is epistemological; it provides the conventionally accepted answer to the thorny question of causation: if the number of times that C precedes E is statistically more significant than E preceding C, then — everything else being equal — C is said to cause E. The lag-sequential technique provides the statistical test for distinguishing systematic from chance transitions (Abbott, 1990). Regrettably, like most statistical methods, lag-sequential analysis depends on several assumptions that are hardly met outside of highly controlled (“laboratory”) environments (Abbott, 2001). Also, a staggering amount of data points that meet those assumptions is required for analysing longer chains (Faraone & Dorfman, 1987). And unlike qualitative-narrative methods, it is difficult to account for context, feedback, and emergence with lag-sequential analysis (and with other stochastic methods such as Markov Chains and Hidden Markov Models, and most variants of the General Linear Model). But as we know, context matters. Paneth et al.’s study, for instance, emphasises the dynamic and contextual nature of CGE, arguing that nonverbal behaviours cannot be universally interpreted without considering the specific context, including cultural and relational factors. And theorising collaborative learning by necessity requires consideration of social structures (e.g., schools and the policy frameworks they operate in) and emergence (e.g., the coupling of cognition and metacognition; of individual and group).

Closing remarks

Cooperative/collaborative learning research, out of which CSCL emerged, had at first to defend the possibility of learners being able to work autonomously, against a dominant backdrop of teacher-controlled classrooms. Over the past two decades, the emphasis on small group work as an isolated phenomenon has been redressed, by situating learners’ interactions within broader educational and social contexts. This has led to extending the research focus to more fully embrace teachers’ roles in organising and scaffolding groups, the emergence of the concept of classroom orchestration, the study of dialogues between groups, the use of dialogues as objects for other dialogues, and even to the taking into account of the roles of families, outside the classroom. With group learning now a mainstream pedagogy and digital technology omnipresent in and outside the classroom, CSCL research has become more widespread and diverse. No longer a niche area, our theories must keep up with the development of methods and research practices.