Volume 10. Essays in Group-Cognitive Science
The eleven essays collected here constitute a program for a new science: the scientific study of small groups as the primary site where knowledge is built in collaborative learning and cooperative work. The collection moves through three interconnected stages—theoretical foundation, empirical demonstration, and methodological systematization—that together make the case for treating the small group not as a context for individual cognition but as a distinct unit of analysis with its own phenomena, methods, and theoretical vocabulary.
The theoretical essays that open the collection identify a gap in the scientific landscape. Existing theories of cognition cluster at two levels—the individual mind and the social community—while the small group, where most of the interesting work of collaborative learning and cooperative problem solving actually occurs, remains undertheorized. Chapter 1 names this gap and proposes filling it with a post-cognitive theory in which group discourse, not individual mental representation, is the primary locus of cognition. Chapter 5 situates this proposal in the full sweep of philosophy from Plato through Hegel and into the twentieth century, showing how the competing theoretical traditions of CSCL and CSCW—Vygotskian socio-cultural theory, activity theory, distributed cognition, ethnomethodology, conversation analysis—each approach but ultimately miss the small-group level. All of these traditions conduct case studies of groups and generate powerful analytical concepts, yet they theorize either the individual mind or the community of practice, sliding past the group itself. Chapter 2 extends this argument by tracing parallel historical progressions in education, theory, technology, and software design, and by arguing that the "objective" and "meaningful" research paradigms, rather than being incompatible, are complementary tools for investigating different aspects of collaborative learning. Chapter 3 widens the lens further, connecting the need for collaborative learning about complex interdependencies to the scientific demands of the Anthropocene epoch—arguing that the same post-individualist conception of agency that CSCL has developed for understanding small groups is also what the sciences of climate and ecology require.
The empirical case studies in the collection's middle section demonstrate the theoretical arguments in action. Chapters 4 and 6 analyze the same group—Team B in the VMT Spring Fest 2006—from complementary perspectives. Chapter 4 builds the multi-level analytical framework from the ground up, showing how indexical reference, recipient design, response pairs, discourse moves, conversational themes, sessions, and group events interlock to produce mathematical knowledge as a group achievement. Chapter 6 presents a simpler, unusually clear episode from the same group—ten discourse moves over ten minutes—demonstrating how the hierarchical structure plays out in a focused problem-solving sequence whose product, a mathematical formula, is literally co-constructed through discourse before it belongs to either individual. Chapter 7 extends the empirical work to biology education and to a different question: whether interaction analysis can serve as a rapid evaluation method within design-based-research cycles. The two- cycle structure of that essay—design, analysis, redesign, improved analysis— mirrors the iterative logic that runs throughout the collection and connects Chapter 7 to Chapter 2's advocacy of design-based research.
The methodological essays that close the collection turn to the question of how to analyze group cognition systematically at scale. Chapters 8 and 11 are reference documents—the coding schemes whose rationale is supplied by the surrounding chapters. They are included not as mere appendices but as evidence that the theoretical framework can be operationalized: the hierarchical structure of discourse described in Chapters 1, 4, and 6 can be coded reliably and systematically. Chapter 9 extends the sequential analysis framework to the transitions between activities, arguing that how groups manage these transitions reveals their orientation toward collaboration as clearly as anything they do within activities—and providing a frequency analysis that shows collaborative transition procedures outnumbering non-collaborative ones by a ratio of more than six to one. Chapter 10 is the most self-critical of the methodological essays, confronting the paradox that reliable, general coding may eliminate precisely the specific novelty that exploratory design-based research most needs to capture.
Taken together, the essays model the kind of multidisciplinary, design-iterative, methodologically pluralist research program that Chapter 2 recommends. Social scientists will find in them a rigorous application of conversation analysis to online text, extended to levels—discourse moves, conversational topics, group events—that conversation analysis has historically underexplored. Computer scientists will find a framework for evaluating collaborative software whose affordances are assessed not through individual pre- and post-tests but through the quality of the group interaction they support. Both audiences will find a coherent, if still developing, science of the small group: one that insists the group is not reducible to its individual members, and that the engines of knowledge building run on discourse.
* Download PDF free for reading online or printing: science.pdf
(This is the best version: up-to-date, complete, full color, hi-res.)
* Download iBook version free for iPad, Mac, etc.: science.epub
* Download MOBI version free for Kindle: science.mobi
* Order paperback from Lulu at printing cost: Lulu page for Gerry Stahl
* Order paperback from Amazon at printing cost: Amazon page for Gerry Stahl
Group Cognition as a Foundation for the New Science of Learning
This opening essay makes the foundational argument for the entire collection: there is a scientific gap between the science of individual cognition and the social science of communities, and that gap can be filled by a new science of the small group. Drawing on a long tradition of post-cognitive theory—including distributed cognition, situated action, and activity theory—the essay observes that while these traditions have conducted case studies of small groups, they have theorized either the individual mind or the social community, never the small group itself as a distinct unit of analysis.
The essay proposes a three-level framework: the individual level (mind, mental representations, constructivist theory), the small-group level (discourse, shared understanding, intersubjective meaning, post-cognitive theory), and the community level (culture, mediating artifacts, socio-cultural theory). A table distinguishes appropriate terminology at each level to prevent the conceptual slippage that has compromised prior theories.
The Virtual Math Teams (VMT) Project serves as the model for building this new science. The project was designed around three interlocking elements: a corpus of naturalistic data (text-based online math chats among middle-school students), a set of interaction-analysis methods including collaborative data sessions of six to twelve researchers, and a developing theoretical framework centered on group cognition. Key design choices—text-only chat, automatic logs, the VMT Replayer—serve specific methodological purposes: ensuring objectivity (logs capture everything visible to the participants), reliability (collaborative data sessions bring multiple analytical perspectives), and validity (non-laboratory, naturalistic settings preserve the character of actual group interaction).
The central thesis, repeated throughout the collection, is that small groups are the engines of knowledge building. Thinking does not happen only in individual neurons; it happens in the discourse of groups. Learning is not the storage of facts by individual minds but the collaborative refinement of knowledge artifacts. The essay calls for a post-cognitivist shift in which the group's shared discourse—not the individual's mental representations—is the primary object of scientific inquiry.
A View of Computer-Supported Collaborative Learning Research and Its Lessons
Originally delivered as guidance for doctoral students and next-generation researchers, this essay surveys the historical development of CSCL and extracts six factors and six lessons for the field's future.
Four historical progressions are reviewed: in education, the dominant model shifted from the transfer of factual knowledge to collaborative exploration of meaning; in theory, the individual-mind orientation of behaviorism and cognitivism gave way to post-cognitive approaches after the philosophical watershed represented by Hegel; in computer technology, mainframes gave way to networked, mobile, and social media; and in software design, a techno-centric model yielded to socio-technical approaches that incorporate human and institutional factors into the design process.
A central theoretical distinction organizes the analysis: the "objective paradigm" (positivism, behaviorism, cognitive science—measuring pre- and post-test outcomes, controlling variables, using statistical methods) versus the "meaningful paradigm" (interpretive sociology and anthropology—analyzing case studies, attending to meaning-making, taking holistic units of analysis). The essay argues that CSCL requires both, and that researchers must recognize their complementarity rather than treating them as incompatible.
One finding highlighted as particularly important is "productive failure": groups that appear to perform less well on short-term tests sometimes demonstrate deeper long-term learning, which challenges the validity of traditional individual testing as an evaluation tool for collaborative learning.
The concept of intersubjectivity is reframed: rather than a mental state shared by individuals, intersubjectivity is an interactional achievement, built through discourse and grounded in a shared world that CSCL software must support.
The essay concludes with six lessons for future researchers: collaborate in multidisciplinary laboratories; study a wide range of theoretical approaches; conduct design-based research that co-evolves software design and educational inquiry; engage in socio-technical design; leverage new technological affordances; and pursue international collaboration. The VMT Project is offered as an example that attempts to embody all six factors.
CSCL for the Era of Climate Change
This shorter essay connects the agenda of CSCL to the challenges of the "Anthropocene" epoch—the geologically recent era defined by humanity's large-scale influence on the natural environment. The essay argues that understanding climate change and related problems requires a foundational shift in how causation, agency, and interdependency are conceived. The sciences that bear on the Anthropocene—from ecology to climate science—no longer support methodological individualism; agents are defined by their interactions and dependencies rather than by isolated properties. CSCL, which studies and supports collaborative learning, is well aligned with this post-individualist scientific perspective.
The concept of dependency takes center stage. Dynamic geometry environments (such as GeoGebra and Geometer's Sketchpad) allow students to construct geometric figures and then drag elements, observing which relationships persist and which change. This practice of exploring what depends on what provides a conceptual primer for the reasoning about complex interdependencies required by Anthropocene science.
The VMT Project extended its collaborative multiuser GeoGebra environment to support this kind of dependency-focused exploration. A curriculum of fifty challenges at increasing levels of difficulty was developed. A case study of one group working through these challenges identified approximately sixty distinct group practices—ways the group coordinated its activity that moved from explicit negotiation to tacit, taken-for-granted procedure.
Five hypotheses guide the essay's vision for future CSCL: that understanding the Anthropocene requires the ability to reason about dependency; that dynamic geometry provides an effective setting for developing that reasoning; that CSCL technology can scaffold collaborative learning of dependency concepts; that collaborative groups can develop shared practices around dependency exploration; and that the VMT Project represents an early model of what this approach to CSCL might look like.
How I View Learning and Thinking in CSCL Groups
The longest and most analytically detailed of the theoretical chapters, this essay presents a multi-level framework for analyzing how learning and thinking occur in CSCL groups, illustrated through a close reading of a single session by Team B in the VMT Spring Fest 2006.
The framework moves from fine-grained to large-scale: at the level of indexical reference, groups must establish a shared network of meaning—a joint problem space—in which words like "it" or "that" point reliably to shared objects; at the level of the utterance, each posting is designed for specific recipients who will do the "reading's work" of interpreting it; at the level of the response pair, postings project expectations and elicit uptake—a question calls for an answer, a mathematical proposal calls for ratification or rejection; at the level of the discourse move, sequences of embedded response pairs accomplish a coherent interactional task; at the level of the conversational theme, an extended sequence of moves develops and closes a topic; at the level of the session, participants manage temporal structure and re-constitute shared history from previous sessions; and at the level of the group event, participants form a collaborative group and co-construct durable knowledge artifacts.
The key interactional unit specific to mathematical discourse is the "math proposal response pair": one student bids a mathematical move, and the group accepts or rejects it, driving the inquiry forward.
The post-cognitive argument is stated most forcefully here: thinking does not happen in individual neurons; it happens in text postings, drawings, and symbolic expressions that refer to each other within the group's shared discourse. Learning is not the storage of information in individual memories but the collaborative refinement of knowledge artifacts that exist as group products in the log. The essay insists that this is not a metaphor but a literal description of how mathematical understanding is built in online collaborative groups.
Theories of Group Cognition: Foundations for CSCL and CSCW
This chapter provides the broadest theoretical sweep of the collection, tracing the philosophical history of theories of cognition from Plato through Kant and the decisive turn represented by Hegel, through the three post-Hegelian traditions (social, situated, and linguistic, associated with Marx, Heidegger, and Wittgenstein respectively), and into the seminal texts these traditions generated for CSCL and CSCW: Vygotsky's work on mind and society, Lave and Wenger's situated learning, Sacks' lectures on conversation, and Winograd and Flores' critique of the rationalist foundations of artificial intelligence.
The essay organizes existing theories into three types according to their unit of analysis: theories of individual cognition (constructivist, socio-cognitive, and even most Vygotskian approaches, which ultimately focus on the individual mind despite acknowledging social context); theories of community cognition (activity theory, distributed cognition, situated learning, ethnomethodology, and conversation analysis—all of which, the essay argues, either slide to the community level or remain ambiguous about the small group); and theories of small-group cognition (dialogical theory in the tradition of Bakhtin, and the theory of group cognition developed in the VMT Project).
The essay argues that neither individual-level nor community-level theories are adequate for the central concerns of CSCL and CSCW, which are explicitly about supporting small groups. A terminological table distinguishing phenomena at the individual, small-group, and community levels is provided to support more precise theoretical discourse.
The chapter concludes by acknowledging that CSCL and CSCW likely require multiple complementary theories operating at different levels and time scales, and that the goal is not a single grand unifying theory but a coordinated set of analytical approaches sensitive to the irreducibly distinct phenomena at each level of description.
How a Virtual Math Team Structured Its Problem Solving
This empirical case study focuses on the final segment of Team B's fourth session in the VMT Spring Fest 2006, when two students—Aznx and Bwang—work together to solve the diamond-sticks problem after their third teammate has left. The analysis identifies ten successive discourse moves that together constitute the entire problem-solving sequence: opening the topic, deciding to start, picking an approach, identifying a pattern, seeking an equation, negotiating the solution, checking cases, confirming the solution, presenting a formal solution on a wiki, and closing the topic.
Each of the ten moves is driven by a base adjacency pair—typically a question posed by one student and a response from the other. The essay demonstrates that the larger problem-solving sequence is not imposed by the analyst but is oriented to by the participants themselves: Aznx and Bwang open the topic and close it, and each step follows logically from the previous.
The analysis shows how collaborative mathematical meaning is co-constructed through the intertwining of individual contributions and group negotiation. When Aznx and Bwang independently arrive at proposed formulas, they reconcile them algebraically rather than one simply overriding the other, establishing a jointly owned solution whose derivation and meaning are likely understood by both.
The case study is presented as a paradigmatic example of group cognition: a mathematical problem that had eluded both students individually (and a third team as well) is solved through a structured sequence of collaborative discourse moves. The mathematical formula exists in the shared discourse of the group before it exists as an individual possession of either student. The essay uses this case to propose a multi-layered hierarchical structure of discourse in virtual math teams, corresponding to the levels analyzed in detail in Chapter 4.
Interaction Analysis of a Biology Chat
This essay extends VMT methods—originally developed for online mathematics—to a biology education setting, examining what happens when the VMT collaboration environment is combined with software helping agents and accountable-talk training to support student discussion of biology experiments. The essay is explicitly framed as a methodological experiment within a design-based-research experiment: can sequential interaction analysis serve as a quick-and-dirty evaluation tool for the analysis phase of a design-based-research cycle?
The method is applied to sixteen chat logs from two cycles of classroom intervention. Working with a representative case from each cycle, the author constructs a visual representation of the response structure—drawing an arrow from each posting to the previous posting it responds to—and reads this representation alongside the content of the postings.
In the first cycle, the analysis reveals a fundamental problem: the software agent dominates the early part of the chat, posting over two hundred and sixty words while students respond with a total of nine words mostly stating their names. The agent fails to respond to students in ways that make sense to them—it treats a sarcastic joke as a genuine scientific prediction and asks students to build on it. Students eventually ignore the agent and work together to fill in their worksheets, but the productive student-to-student interaction is largely disconnected from the accountable-talk goals of the intervention.
The analysis findings are used to guide redesign for the second cycle. The redesigned intervention shows clear improvement in the response structure analysis: the agent's behavior is more responsive and less authoritative, and the student group's interaction more closely resembles the accountable-talk model of reciprocal uptake and collaborative knowledge building. The essay argues that quick sequential analysis of a single case study—without requiring pre-specified hypotheses, statistical testing, or extensive data collection—can provide grounded, actionable recommendations for redesign in design-based-research cycles.
Coding Scheme for Sequential Discourse
This brief chapter presents the coding scheme for sequential discourse developed within the VMT Project and applied to the chat log of Team B in the VMT Spring Fest 2006. Rather than an argumentative essay, it is a reference document: two tables listing the codes for the hierarchical structure of discourse and the four turn-constructional units analyzed in detail in the following chapter.
The hierarchical structure moves from (g) indexical references within (f) textual utterances, which contribute to (e) adjacency pairs, which combine to form (d) discourse moves, which are organized into (c) conversational topics, embedded within (b) group events, and constituting (a) the temporal session. Standard symbols for first pair parts, second pair parts, pre-sequences, insert sequences, post-sequences, and sequence-closing thirds are listed alongside their common instances: questions and answers, requests and grants, proposals and ratifications, greetings and farewells, and so on.
The four turn-constructional units are named: the proposal-ratification-uptake (PRU) sequence, the suggestion-ratification-uptake (SRU) sequence, the request-acknowledgment-uptake (RAU) sequence, and the directive-compliance-report (DCR) sequence. An excerpt from the coded log of Team B is reproduced, illustrating how the coding scheme is applied to actual chat data. The chapter functions as a methodological supplement to the surrounding analytical chapters, providing the symbolic vocabulary through which the multi-level structure of group cognition in chat can be systematically documented.
Coordinating Collaborative Action in Online Math Problem Solving
Co-authored with a colleague, this essay investigates how actors in a CSCL setting transition from one activity to the next—a question that has received surprisingly little attention in conversation analysis despite the centrality of sequential organization to that tradition.
Using data from Team B's sessions in the VMT Spring Fest 2006, the essay identifies four types of next-sequence selection sequences. Proposal-ratification- uptake (PRU) sequences occur when one actor proposes a next activity and the group ratifies it before taking it up. Yes-no query request sequences solicit a binary confirmation before proceeding. Suggestion-initiated selection sequences are somewhat less binding than proposals but still invite group ratification. Directive-compliance-report (DCR) sequences occur when one actor tells another what to do next without soliciting ratification.
The first three sequence types are analyzed as collaborative: they treat the collectivity rather than any individual as the actor responsible for choosing and initiating the next activity. The DCR sequence is analyzed as non-collaborative: it constitutes an individual actor, rather than the group, as the agent of selection. A frequency count of Team B's session data shows that collaborative transition procedures far outnumber non-collaborative ones—sixty combined occurrences versus nine for DCR.
The essay's broader claim is that collaboration is as much a matter of how groups manage transitions between activities as it is a matter of how they conduct activities. By examining these transitions, analysts gain a particularly clear window onto the group's orientation toward collaborative action—because at the moment of transition, the group's implicit agreement or disagreement about what to do next becomes visible in the discourse itself.
Methodological Issues in Developing a Multi-Dimensional Coding Procedure for Small Group Chat Communication
Co-authored with a visiting researcher, this essay is explicitly a methodological report on what went wrong—and right—when the VMT Project attempted to apply a multi-dimensional coding scheme to small-group chat data. The essay identifies three issues that arose during development and addresses each in turn.
The first issue is unit of analysis and unit fragmentation. The chat line was chosen as the unit because it is user-defined and typically contains a single communicative act, avoiding arbitrary segmentation. However, students frequently spread a single communicative act across multiple successive chat lines, creating "unit fragmentation." The solution is to use "extend" and "setup" codes that link fragments to each other, treating them as a compound utterance and assigning the conversational code only to the first fragment.
The second issue is the reconstruction of response structure. Unlike face-to-face conversation, where turn-taking conventions make the sequence of responses clear, small-group chat allows multiple students to type simultaneously, producing postings that may respond to earlier messages rather than the immediately preceding one. The essay introduces "threading"—the reconstruction of who is responding to whom—as a preliminary step that must be completed before any coding of communicative content can proceed reliably.
The third issue is the risk of reliability overestimation. The essay demonstrates that when categories are sparsely used (many cells in the coding matrix receive no code), standard reliability statistics can produce misleadingly high values that do not reflect genuine inter-rater agreement on the meaningful codes.
The essay ends with a pointed observation: the paradox of coding in exploratory CSCL research is that the reliable application of general codes can eliminate precisely the specific, novel content that is most important for design-based research.
A Multi-Dimensional Coding Scheme for Mathematical Collaboration
The final chapter presents the complete multi-dimensional coding scheme for mathematical collaboration developed within the VMT Project in 2004 and subsequently refined through application to actual chat data. Like Chapter 8, this chapter is primarily a reference document—a set of carefully defined codes with examples and rules for application—rather than an argument. It is the practical instrument whose development is narrated in Chapter 10.
The scheme is organized into dimensions. The conversation dimension distinguishes: State (an unprompted assertion), Offer (introduction of new problem-solving content), Request (a question or solicitation of any kind), Regulate (coordination of turns, activities, or timing), Repair (correction of a typing error), Respond (a response not otherwise specified), Follow (confirmation of understanding without explicit agreement or disagreement), Elaborate (an utterance that builds new content on a previous one by the same actor), Extend (a fragment that continues a split utterance), Setup (a fragment that precedes the main utterance), Agree, Disagree, Critique, and Explain. Additional dimensions—social, problem-solving, and support—capture other aspects of the interaction.
Rules for applying the scheme emphasize: establishing threading before assigning conversational codes; distinguishing extension (part of a fragmented single act) from elaboration (a new contribution building on a previous act); and coding only completed utterances rather than fragments as independent acts. The scheme is designed to treat the group as the unit of agency, analyzing how jointly produced sequences of coded acts constitute the collaborative cognitive work of the group rather than simply cataloguing the contributions of individuals.