Journal of Computer Assisted Learning

2.1 Computer‐supported knowledge building discourse

Knowledge building is a computer‐supported educational model emphasizing students taking collective cognitive responsibility for community knowledge advancement (Scardamalia & Bereiter, 2006, 2014). Ideas are treated as improvable conceptual artefacts (Bereiter, 2002), and students take responsibility and agency to pursue the creation and improvement of ideas that add value to the community (Zhang et al., 2007). To support these knowledge‐building processes, KF has been developed. It provides a multimedia community knowledge space wherein students can pursue ideas collaboratively to advance community knowledge in much the same way as mature research communities do (Scardamalia, 2004). KF scaffolds students to develop and publish their ideas (as an author or co‐author), and link, revise, and reorganize these ideas. KF is primarily constituted of views and notes which serve as collaborative inquiry spaces for organizing notes and improving ideas. In Figure 1, the small square icons represent students' notes, and the lines between icons are the build‐on relationship between notes. Figure 2 shows an example of a note. The scaffolds (e.g., I need to understand; my theory; new information; and a better theory) are prompts students can choose to support their theory‐building process.

Examining and analyzing knowledge building discourse on KF is central to knowledge building research (Scardamalia & Bereiter, 2006; Zhang et al., 2007). Some researchers have used social network analysis to visualize the quantitative interaction in knowledge‐building discourse (Chen, Scardamalia, & Bereiter, 2015; Oshima, Oshima, & Matsuzawa, 2012), some used qualitative content analysis to examine discourse quality and patterns (Lipponen, 2000; Zhang et al., 2007), and some used mixed methods to quantify qualitative discourse analyses (Lee, Chan, & Aalst, 2006; Resendes, Scardamalia, Bereiter, Chen, & Halewood, 2015). Hakkarainen's (2003) “questions and explanations” framework is widely used to analyse knowledge‐building discourse (e.g., Lee et al., 2006) and allows researchers to code and count the depth of questions and explanations in the inquiry and relate them to students' learning (Hakkarainen & Sintonen, 2002; Lee et al., 2006; Zhang et al., 2007). However, this framework falls short in unpacking discourse interactions and in identifying conceptual connections among ideas (e.g., how coherent later ideas are linked to previous ideas). As CSCL trends emphasize connected discourse, knowledge‐building discourse research has shifted to analysing KF inquiry threads and notes for emergent theory‐building and knowledge advances informed by knowledge‐building principles (Zhang et al., 2007).

Premised on Bereiter and Scardamalia's knowledge‐building model, van Aalst (2009) distinguished between three modes of discourse: knowledge sharing, knowledge construction, and knowledge creation. Knowledge sharing involves knowledge transmission, and people share information and opinions without much processing. Knowledge construction involves cognitive processing (e.g., metacognition, interpreting, and evaluating new information), and students interact to process information and ideas deeply. In knowledge creation, new understandings and ideas emerge from collective inquiry. Ideas are treated as improvable conceptual artefacts, and the community works together, continually identifying gaps and engaging in inquiry. These different discourse modes reflect different degrees of knowledge building and advancement.

Despite the growing interests in understanding the interaction and collective aspects of the discourse, less is known about how the discourse moves are related to the collective knowledge advancement of the community. In this study, we characterized students' discourse moves from an epistemic and theory‐building perspective as well as examined the relationship between discourse moves and collective knowledge advancement. Specifically, building on van Aalst's (2009) work, we coded discourse threads (threads of notes that are physically connected) constructed by a group of students to depict different degrees of conceptual quality and collective knowledge advancement and examined how the proportion of discourse moves might differ in these discourse threads.

2.2 Epistemic cognition in science and knowledge building discourse

Epistemic cognition in science is concerned with how people think about the nature of knowledge and knowing in science (Elby, Macrander, & Hammer, 2016; Greene, Sandoval, & Bråten, 2016). Sophisticated epistemic cognition in science is an important goal of science education and is related to students' learning and reasoning (Chinn, Buckland, & Samarapungavan, 2011). Elby et al. (2016) discussed how epistemic cognition in science stems largely from two traditions of research: personal epistemology and the nature of science (NOS). The former originates from the psychological tradition (Hofer & Pintrich, 1997; Schommer, 1990) and examines students' epistemological beliefs including such dimensions as source, certainty, development, and justification of knowledge (Conley, Pintrich, Vekiri, & Harrison, 2004; Elder, 2002). The second tradition, the NOS, stems from discussions about the goals of science education and is informed by philosophy and sociology of science. The examined NOS aspects include the empirical, creative, and imaginative nature of science along with the theory‐laden and social‐cultural aspects of science (Lederman, Abd‐El‐Khalick, Bell, & Schwartz, 2002; Tsai, 1999).

Related to these two traditions, another particular strand examines children's understanding of science from a constructivist and role‐of‐idea perspective (Carey et al., 1989; Smith, Maclin, Houghton, & Hennessey, 2000), that is, variously seeing science as concrete, tentative, or constructing explanations for their understanding of the world. Some researchers (Chuy et al., 2010) from this perspective also emphasize the theory‐building nature of science and identify four dimensions of epistemic views: nature of scientific progress, theory–fact understanding, role‐of‐idea in scientific progress, and invention. This study followed this line of research and examined students' epistemic cognition in science from a role‐of‐idea and theory‐building perspective.

One of the key goals of this study is to examine the relationship between knowledge‐building discourse and epistemic and conceptual understanding. Substantial research in science has shown how more‐sophisticated epistemic cognition is related to conceptual understanding (Mason, 2010; Stathopoulou & Vosniadou, 2007); students who believe knowledge is changeable are more likely to engage in deeper processing, leading to better conceptual understanding. Many studies have also examined the role of knowledge‐building discourse in conceptual understanding, and research has shown that students' online knowledge‐building discourse which was characterized by level of inquiry and explanations predicted their science understanding (Lee et al., 2006). However, specific investigation on the relations of epistemic cognition from a theory‐building perspective with student's conceptual understanding and systematic examination on the relationship between knowledge‐building discourse, epistemic cognition, and conceptual understanding are still lacking.

There are now increasing research interests in examining students' epistemic cognition through the lens of discourse (Knight & Mercer, 2017; Scherr & Hammer, 2009). For example, Knight and Mercer (2017) examined the relationship between epistemic cognition and information‐seeking in the context of collaborative discourse and identified epistemic moves associated with learning outcomes. Rosenberg, Hammer, and Phelan (2006) analysed students' discourse, to examine the nature of epistemological understanding and coherence. Scherr and Hammer (2009) explored how students' behaviour and discourse reflected their epistemic framing. Although various studies examined epistemic cognition from the lens of classroom discourse, there have been few studies investigating the relations between online collaborative discourse moves and students' explicit epistemic cognition.

In the knowledge building research tradition, as discussed above, there have been major research efforts toward analysing discourse patterns and their relationship with conceptual understanding. There has also been some interest in examining discourse as epistemological inquiry (Hakkarainen, 2003) and the influence of knowledge building environment on students' epistemic cognition (Chen, 2017; Hong & Lin, 2010; Lin & Chan, 2018). Hong, Chen, and Chai (2016) examined how college students with more or less sophisticated epistemic cognition differed in their KF discourse, indicating that KF discourse impacts epistemic cognition. However, few studies have specifically investigated how KF discourse moves might contribute to students' explicit epistemic cognition, particularly among elementary school students.

Therefore, in this study, we will examine the relationship between KF discourse, explicit epistemic cognition, and conceptual understanding. To this end, we will first characterize students' KF discourse from an epistemic and theory‐building perspective and then examine whether these characterized KF discourses are related to students' epistemic and conceptual understanding. We hypothesize that students' engagement in theory‐building discourse will contribute to students' epistemic understanding of science as a collective theory‐building process and their conceptual understanding in science.

Three research questions were addressed in this study: (a) What characterizes students' discourse moves in KF, and how do they differ in discourse threads of high and low knowledge advancement? (b) How do discourse moves predict epistemic understanding of science? And (c) How do discourse moves predict conceptual understanding in science?

3.1 Participants and context

The participants in this knowledge‐building environment were 52 fifth graders (aged 10–11) from two classes in the same school (Class 1, 13 boys, 20 girls; Class 2, 10 boys, 9 girls) located in a low‐income area of Hong Kong. Two different teachers taught the same science topic on electricity and conductors; they used the same pedagogy and material and jointly designed the program with the researchers.

3.2 Procedure

The program was conducted in 11 sessions (40 min each) over 3 months. Both classes were designed based on knowledge‐building pedagogy and epistemic reflection using KF. The students posed problems, questions, and explanations on KF to advance community knowledge. The design encouraged the students to reflect on their inquiry processes and to compare their inquiry with mature scientific inquiry. The 11 sessions included classroom inquiry in related to science concepts, the teachers' instruction on using KF features, students' experimentation and discussion in the classroom, and students writing on KF.

The knowledge‐building environment design supported by KF included several features: The first was concerned with activating prior understanding. Students discussed their initial ideas about electricity and inquired into what scientists do. The second feature was about authentic problems and inquiry. The students watched experiments about materials conducting electricity which triggered their wonderment; then, they posed ideas and questions on KF. The students could use KF features when posing ideas and writing computer notes on KF (Figure 2). For example, when they start a note, they could select relevant scaffolds (e.g., I need to understand, my theory, and new information) to help structure their writing and inquiry. These scaffolds were explained by the teacher and discussed in class to facilitate students' theory‐building discourse. The third feature involved deepening their inquiry through experiments and classroom epistemic talk based on KF writing. The students designed and conducted group experiments and continued working on KF to improve their ideas. The teachers scaffolded students to reflect on their forum discourse. They started with initial “views” (discussion area), and then deepening views on KF were created for emerging questions. KF discussions were asynchronous. The students worked both in class and after class on KF, and online and offline discourse were integrated. Both classes worked in their own classroom and respective KF discussion views. The materials and media of communication were in Chinese.

3.3 Data source

3.3.4 KF Writing: Discourse moves

Analyses of individual notes were conducted to examine discourse moves within inquiry threads, from a theory‐building perspective. Although they were examined individually, they were considered in situ as part of the entire thread. The unit of analysis was individual note within the physical thread, meaning that notes were coded individually, but their physical level in the inquiry threads was considered. Figure 3 shows how the physical levels of a complete KF thread were defined. The original question represents Level 1, a response to this note is Level 2, a response to a Level 2 note is Level 3, and so on. This allowed us to track the sequences of ideas in physical threads and to assess the coherence of idea interaction. During the analytic process, we sorted all notes in an Excel file and labeled their physical levels. Two key discourse move types were identified, and all KF notes were double coded for both discourse types.

Problem‐centred uptake moves

This discourse move is about how ideas are progressively and coherently developed to problematize and to address previous ideas. Scientific and epistemic inquiry is problem‐focused; problem‐centred discourse moves have been identified as important in students' dialogue for conceptual change (Chan, 2001). A key characteristic of knowledge‐building is students working on progressive problem solving (Scardamalia & Bereiter, 2014). We adopted the interactional analysis term “uptake” (Suthers, Dwyer, Medina, & Vatrapu, 2007) to describe the “act of a participant taking reifications of prior or ongoing participation … as having certain relevance for further participation” (p. 696). Uptake focuses on social dimensions, and we elaborate it to include epistemic aspects—problem‐centred uptake—to reflect how students problematize collective knowledge, track problems, and coherently address ideas in problems.

Four levels of problem‐centred uptake notes were identified. Level 1 notes made no connection or a weak connection to the previous note (e.g., irrelevant response and simply saying yes/no). Level 2 notes showed a vague connection, which was not essential to solving the problem (e.g., asking for elaboration and offering simple explanation). Level 3 notes had clear and reasonable connections to previous notes (e.g., building connections and asking for explanation). Level 4 notes showed coherent connection to previous notes and were important for tackling and solving the problem (e.g., avoiding closure; deepening; and refocusing the discussion). Notes were contextualized and coded with consideration to their physical levels.

Knowledge building emphasizes problem‐centred discourse and often discourse threads begin with a problem. Our analysis on problem‐centred uptake could help tap into whether students can make connections to the previous notes and reflect whether they attempt to address community needs for solving problems. Students may or may not be able to solve the problems, just as scientists may not solve complex problems after years; but the attempts to connect to previous notes (connect to others' work) is an important part of problem solving and epistemic practice.

Two raters independently coded 30% of the problem‐centred uptake notes, obtaining an interrater reliability of K = .84 (Cohen's Kappa).

Theory‐building moves

Theory‐building moves refer to the different roles the notes play in theory‐building processes. Scientific inquiry and theory‐building involves initiating a problem, asking questions, constructing explanation, deepening explanation, theorizing, and sustaining emerging inquiry. Qualitative analyses, using bottom‐up and top‐down interactive approaches, identified eight key theory‐building moves (see Table 5): (a) initiate inquiry (low), fact‐seeking questions/statement; (b) initiate inquiry (high), explanation‐seeking questions; (c) explanation (low), general and intuitive explanation; (d) explanation (high), theorizing and searching for mechanism; (e) sustain inquiry (low), asking elaborate questions; (f) sustain inquiry (high), asking emerging and deepening questions; (g) cognitive conflict, disagreement, and refutation; and, (h) nonbuild‐on. Table 5 shows the definition, and Table 6 provides specific examples for these different types of theory‐building discourse moves in a discourse thread. Discourse moves based on KF notes were coded from the 28 discourse threads. The first author coded all data and identified eight categories of theory‐building moves. A second‐rater independently coded 30% of the notes. Cohen's Kappa was .97, indicating very good interrater agreement.

Table 5. Coding scheme for theory building moves

Code	Description
Initiate inquiry (low)	Initiates a thread with fact‐seeking question or a statement
Initiate inquiry (high)	Initiates a thread with explanation‐seeking question
Explanation (low)	Provides general and intuitive reasons for the phenomena using simple statements or paraphrases information
Explanation (high)	Constructs elaborated responses to theorize and to explain; conjecturing mechanisms for phenomena, incorporating new information
Sustain inquiry (low)	Asks simple or superficial questions to continue the discussion
Sustain inquiry (high)	Sustains the inquiry with deepening questions; engages in progressive problem solving
Cognitive conflict	Shows disagreement to other ideas; refutes different viewpoints
Nonbuild on	Scattered notes without any build‐on; or other irrelevant notes

Table 6. An example of coding for theory‐building discourse moves

Note #	Student #	Note content	Physical level	Theory‐building moves
#1	5a28	[I need to understand]: why can wet stuff get people an electric shock?	1	Initiate inquiry (high)
#2	5a06	[My theory]: because of water	2	Explanation (low)
#3	5a25	[Your theory can not explain]: if it is distilled water, then it cannot conduct electricity, because it is distilled.	3	Cognitive conflict
#4	5a22	[I need to understand]: why?	4	Sustain inquiry (low)
#5	5a31	[My theory]: because only impure water can conduct electricity	5	Explanation (low)
#6	5a29	[I need to understand]: which kind of impure water can conduct electricity?	6	Sustain inquiry (high)
#7	5a33	Salt water	7	Explanation (low)
#8	5a21	Salt	7	Explanation (low)
#9	5a27	[My theory]: because water is a conductor	2	Explanation (low)
#10	5a09	[My theory]: water is not a conductor. It is semiconductor.	3	Cognitive conflict
#11	5a01	[I need to understand]: why is water a semiconductor? What is semiconductor?	4	Sustain inquiry (low)
#12	5a34	[my theory] a semiconductor has resistivity that is between an insulator and metals …	5	Explanation (high)
#13	5a07	[I need to understand] what is resistivity?	6	Sustain inquiry (low)
#14	5a29	[My theory] a material that has a low resistivity is called conductor, the common conductor is metal, but the best conductor is silver. ….	7	Explanation (low)
#15	5a14	[I need to understand] how do we know which material has a higher or a lower resistance?	8	Sustain inquiry (high)
#16	5a20	[A better theory]: the water we use in our daily life is not pure water. It has some minerals in it, and contains positive and negative ions. Therefore, it can conduct electricity. The negative ions come from the neutral atoms that gain electrons due to the chemical reaction outside.	2	Explanation (high)

4.1 Background information

The general effects of the designed learning environment on epistemic and conceptual understanding are reported to provide contextual information. Paired‐sample T‐tests showed Classes 1 and 2 students improved in conceptual understanding at statistically significant levels (pretest, M = 0.25, SD = 0.08; posttest, M = 0.56, SD = 0.14; t = −11.602, p < .001, Cohen's d = −2.72) and (pretest M = 0.25, SD = 0.10; posttest M = 0.51, SD = 0.14; t = −6.49, p < .001, Cohen's d = −2.14), respectively. Students' improvement in epistemic cognition was also statistically significant (pretest, M = 6.99, SD = 1.15; posttest M = 9.92, SD = 1.75; t = −9.84, Cohen's d = −1.98) and (pretest M = 6.22, SD = 1.77; posttest M = 8.14, SD = 2.08; t = −3.157, p = .005, Cohen's d = −.99), respectively. These results indicate that both classes experienced changes in epistemic and conceptual understanding after the knowledge‐building instruction.

Class 1 students wrote 449 notes on KF. On average, students wrote 13.2 notes each, read 30.4% of their peer's notes, used 11.7 thinking prompts, and made 1.6 revisions; a total of 66.7% of notes were linked (connected) to other notes. Class 2 wrote 202 notes. On average, students wrote 7.5 notes, read 32%, used 9.9 thinking prompts, and made 1.5 revisions. A total of 52.7% of notes were linked. With the exception of productivity, the two classes are comparable on major KF participation indices, particularly those pertaining to collaboration (e.g., notes read and linked). These participation levels are also generally comparable with those reported in similar studies (Lai & Law, 2006). Both classes used similar knowledge‐building designs, the teachers codesigned all lessons with the researchers, and the students all had the same instructional time. Together with the information on the patterns of preposttest change, these two classes are considered comparable in providing data for discourse analysis in relation to epistemic and conceptual change. To increase the power of analysis, the two classes were combined for analysis. Of the 52 students, one was excluded from the analysis as an outlier based on initial data analysis related to prior conceptual understanding scores.

Both classes in total wrote 651 KF notes. Data checking was conducted to identify the notes relevant for discourse analysis. For all the KF notes written, some are unrelated notes (not related to electricity discussion), and some are individual portfolio (summary) notes on what they learned. Because the current study mainly focuses on knowledge‐building discourse, we did not include those unrelated and portfolio notes and only analysed students' discussion about electricity (475 notes).

4.2 Research Question 1: Discourse moves and collective knowledge advancement

Question one examined the characterization of discourse moves and their relations with collective knowledge quality and advancement. KF discourse threads were analysed using a 4‐point coding scheme to identify the extent of knowledge advancement. We identified 28 discourse threads, ranging from low‐ to high‐knowledge quality, including 2 fragmented discussion, 9 knowledge‐sharing, 13 knowledge‐construction, and 4 knowledge‐building discussions threads (episodes). Table 7 shows the descriptive information on frequencies and percentages of different discourse move types for the different knowledge‐building thread levels.

Analyses were conducted to investigate how the occurrence of discourse moves varied in discourse threads that depicted different levels of knowledge quality and advancement. To control for frequencies in discourse threads of different lengths (i.e., loner threads have more occurrence of discourse moves), we examined the proportions rather than frequency counts. We categorize the four types of threads into two major groups (fragmented and sharing as low versus knowledge‐construction and knowledge creation as high) to illustrate students' collective work in discourse of high‐ versus low‐conceptual quality. We then performed Chi‐squared tests of homogeneity of proportions using RStudio to compare the proportion of high‐level problem‐centred uptake between high‐level (n = 17) and low‐level (n = 11) threads. The results showed that high‐level discourse thread differed from the low‐level ones on proportion of high‐level problem‐centred uptake moves (χ² = 51.952, df = 1, p < .001). The p values were adjusted using the Bonferroni–Holm method to account for cumulative error.

Similarly, we also conducted Chi‐squared tests of homogeneity of proportions to compare the proportion of different kinds of theory‐building moves between high‐ and low‐level discourse patterns. The results showed that there were significant differences in proportions of sustain inquiry (low)(χ² = 5.351, df = 1, p = .02), sustain inquiry (high)(χ² = 8.592, df = 1, p = .003), and explanation (high)(χ² = 14.097, df = 1, p < .001). There were no significant differences on initiate inquiry (low), initiate inquiry (high), explanation (low), or cognitive conflict. The p values were adjusted using the Bonferroni–Holm method.

These results indicate that in high‐level discourse threads (more collective knowledge advancement), ideas were more coherently built upon each other to address the inquiry problem, and there were more efforts to sustain the inquiry and develop elaborated explanations.

4.3 Research Question 2: Discourse moves and epistemic understanding of science

The second and third research questions examine how the identified discourse moves predict students' epistemic understanding of science and conceptual understanding. We first conducted correlation analysis to examine what discourse moves were related to epistemic understanding. Based on this information, we then conducted hierarchical regression analysis to examine if these discourse moves were significant predictors of epistemic and conceptual understanding.

4.4 Research Question 3: Prediction of discourse moves on conceptual understanding

Hierarchical regression analyses were conducted to examine how the identified discourse moves and other variables predicted posttest conceptual understanding. Only the variables that were statistically significantly correlated with conceptual understanding were included. For order of entry, we followed causality‐relevance and chronological sequence. Prior science understanding (including both pretest and achievement scores) is key to determining posttest conceptual understanding and so was entered first. After prior science understanding, KF discourse moves were entered as they took place during instruction that preceded posttest epistemic cognition (see Table 10).

Prior science understanding explained 21.3% of variance (R² = .213, p = .004). Adding high‐level problem‐centred uptake explained no significant additional variance. Adding explanation (high) explained an additional 6.7% of variance (R² = .323, p = .04). When post epistemic cognition was added, 5.4% of variance was added, but it was not significant (R² = .378, p = .056). These results suggest that, over and above prior science understanding, high‐level explanation discourse moves contributed to students' conceptual understanding. It indicates that the more students contributed to the collaborative discourse through elaborated explanations, the more likely they scored higher on conceptual understanding.

5.1 Characterization of discourse moves

Discourse analyses were conducted to examine the epistemic aspects of collaborative knowledge‐building discourse. We first analysed KF discourse threads depicting different levels of knowledge building and advancement: fragmented discussion, knowledge sharing, knowledge construction, and knowledge creation (van Aalst, 2009). We then analysed the discourse moves within inquiry threads and identified two major kinds of discourse moves, that is, problem‐centred uptake and theory‐building moves. Problem‐centred uptake moves were coded based on the notes' relation to previous notes and the problem under discussion, whereas theory‐building moves were coded based on their different roles in the thread. The patterns overlap somewhat but are distinctive, as the former reflects students' epistemic goals of problematizing knowledge and coherently addressing the problem for community progress, and the latter reflects the epistemic processes of theory building and scientific inquiry.

Argumentation and theory building are important science practices. Many previous studies have examined online discourse from an argumentation perspective (Chiu, 2008; Felton & Kuhn, 2001; Lu et al., 2011; Nussbaum, 2008; Weinberger & Fischer, 2006), whereas few have conceptualized collaborative discourse as a theory‐building process. We based our work on knowledge‐building theories (Scardamalia & Bereiter, 2006) that emphasize continual idea improvement and theory building and explored ways of analysing collaborative discourse from a theory‐building and epistemic perspective (Bereiter, 2012). Although different CSCL and knowledge‐building studies have used questions and explanations to examine discourse (Hakkarainen, 2003; Lee et al., 2006), this study examined them as a system of theory‐building processes, including initiating inquiry, asking questions, giving explanations, posing conflict, and sustaining inquiry. The analysis reflected the epistemic features of knowledge building (e.g., progressive explanation and sustained inquiry) and allowed us to understand students' epistemic inquiry in more connected ways.

This approach extends previous studies on the epistemic nature of individual notes (e.g., question and explanation; Hakkarainen & Sintonen, 2002) by emphasizing the connection among notes and ideas and enriches the literature on approaches to examining interaction in CSCL (Puntambekar, Erkens, & Hmelo‐Silver, 2011). Rather than just coding and counting discourse moves (Lee et al., 2006) or qualitatively examining inquiry threads (Zhang et al., 2007), we adopted a code‐and‐count approach, while also considering the interactive and situated nature of discourse, using physical‐level analysis and tracing the development of ideas. Such an approach has implications for analysing online discourse, as it both captures the interaction of ideas and allows for examining the relationship between identified discourse moves and other measures.

5.2 Relations between discourse moves and collective knowledge advancement

Current CSCL and KB research focuses on collective knowledge advancement; and our analysis included both collective and individual aspects. Specifically, we examined whether and in what ways the identified discourse moves are associated with more or less collective knowledge advancement. The results showed that the high‐level discourse threads (more knowledge advancement) had a higher proportion of high‐level problem‐centred uptake and theory‐building moves. In these instances, ideas were more coherently built upon each other to address the inquiry problem; there were also more high‐level explanations theorizing mechanisms and more emerging questions to sustain the inquiry. The results also showed that high‐level discourse threads did not differ from low‐level discourse threads in terms of cognitive conflict, low‐level explanations, or nonbuild‐on, indicating that merely expressing different ideas in online discussion (conflict) did not necessarily bring about knowledge advancement. Disagreement and conflict may provoke explanations and questions that need further tracking, while merely disagreeing and offering intuitive explanations were not sufficient and seldom led to knowledge construction and community knowledge advancement.

We also found that high‐ and low‐level discourse threads did not differ in initiate inquiry (low) or initiate inquiry (high), indicating that, whether the inquiry thread was initiated as a fact‐seeking question (low‐level initiate inquiry) or an explanation‐seeking one (high‐level initiate inquiry), it did not affect inquiry thread quality. In group inquiry, it is usually not the individual's intention that determines the direction and depth of the inquiry but the group's shared cognition. It is argued here that we should not classify a given discourse pattern as fact‐seeking or explanation‐seeking, based merely on the questions asked (Hakkarainen, 2003); rather, we must examine how the community interprets and processes the question, and how its members uptake ideas for theory building. For example, a student may start a conversation by asking, “Can salt water conduct electricity?”—which is usually interpreted as a fact‐seeking question. However, the question could contain tacit explanation‐seeking overtones, depending on how the community processes it. After determining that salt water can conduct electricity, they could then also wonder why, and if it were the water that conducts electricity, the salt, or the combination thereof. With each response provided, the inquiry could deepen. However, if the community does not have an epistemic goal of problematizing collective knowledge, students are unlikely to be aware of these tacit meanings, nor the necessity for further explanation and therefore may only offer yes/no answers or simple factual responses. This finding indicates that, although encouraging students to initiate inquiry through good questions is important, it is even more important to scaffold the community to process the ideas in the questions and to deepen the inquiry.

5.3 Prediction of discourse moves on epistemic and conceptual understanding

The analyses on the relationship between discourse moves and epistemic and conceptual understanding showed that students' engagement in high‐level problem‐centred uptake discourse move predicted their epistemic cognition over and above prior epistemic cognition. These findings support our hypothesis about the relationship between problematizing and coherence and epistemic cognition in science. Our results suggest that the more coherently students built on their peers' ideas to address the central problem in KF discourse, the more likely they were to understand science as a collective theory‐building process.

The results also showed that high‐level explanation/theorizing moves predicted conceptual understanding beyond prior science knowledge. This is consistent with the relationships between depth of explanation and conceptual understanding shown elsewhere (Lee et al., 2006). However, we did not find that post epistemic cognition further added to the explained variance, which is different from the previous studies (Mason, 2010; Stathopoulou & Vosniadou, 2007) that suggested the role of epistemic cognition in conceptual understanding. This is probably caused by the small sample size. The complex relationship among discourse moves, epistemic cognition, and conceptual understanding needs to be investigated further.

Some researchers have suggested possible influence of dialogic teaching in developing students' epistemic cognition (Bendixen, 2016). For example, Ryu and Sandoval's (2012) study shed light on the role of classroom argumentative discourse in developing students' understanding of the epistemic criteria for argumentation. Our study contributes and extends research on student dialogue by examining the role of online theory‐building discourse in promoting students' epistemic understanding of the socially constructed nature of science.

This study also indicates that designing technology with epistemic implications is important for promoting students' epistemic understanding. The technological design of KF—that different ideas can be connected and built on for further improvement—has the epistemic implication that ideas are socially constructed and tentative. It also provides affordances to support students' epistemic practice, which impacts their epistemic understanding. As noted by Caswell and Bielaczyc (2002), KF could alter the relationship between students and scientific knowledge; as students engage in knowledge‐building discourse, the epistemic implications of their knowledge‐building practice may gradually help them understand the tentative nature of scientific knowledge and the progressive nature of scientific inquiry.

We also noted this study is limited, in that the sample size of the discourse threads is small, which limits the generalizability of the study's conclusions. Future research involving a larger sample size is needed to examine if consistent patterns can be observed. Additionally, the small sample size also limited our statistic approaches to examining the relationship among conceptual understanding, KF discourse moves, and epistemic cognition; future studies could test their interrelationships using larger samples in path analysis.