Facilitating peer knowledge modeling: Effects of a knowledge awareness tool on collaborative...

9
Facilitating peer knowledge modeling: Effects of a knowledge awareness tool on collaborative learning outcomes and processes Mirweis Sangin * , Gaëlle Molinari, Marc-Antoine Nüssli, Pierre Dillenbourg Ecole Polytechnique Fédérale de Lausanne (EPFL) – School of Computer and Communication Sciences, 1015 Lausanne, CH, Switzerland article info Article history: Available online 6 July 2010 Keywords: Computer supported collaborative learning Group awareness Knowledge awareness Peer knowledge modeling Spatially distributed groups abstract We report an empirical study where we investigated the effects, on the collaborative outcomes and pro- cesses, of a cognition-related awareness tool providing learners with cues about their peer’s level of prior knowledge. Sixty-four university students participated in a remote computer-mediated dyadic learning scenario. Co-learners were provided (or not) with a visual representation of their peer’s level of prior knowledge through what we refer to as a knowledge awareness tool (KAT). The results show that, pro- viding co-learners with objective cues about the level of their peer’s prior knowledge positively impacts learning outcomes. In addition, this effect seems to be mediated by the fact that co-learners provided with these objective cues become more accurate in estimating their partner’s knowledge – accuracy that predicts higher outcomes. Analyses on the process level of the verbal interactions indicate that the KAT seems to sensitize co-learners to the fragile nature of their partner’s as well as their own prior knowledge. The beneficial effect of the KAT seems to mainly rely on this induction of epistemic uncertainty that implicitly triggers compensation socio-cognitive strategies; strategies that appear to be beneficial to the learning process. Ó 2010 Elsevier Ltd. All rights reserved. 1. Introduction A large body of research has shown that collaborative learning is more effective than individual learning, but not systematically. The effectiveness of collaborative learning mainly relies on the quality of social interaction during collaboration (Dillenbourg, 1999; Dillenbourg, Baker, Blaye, & O’Malley, 1996; Suthers, 2006). Collab- orative learning is an added-value when co-learners actively con- struct shared knowledge through socio-cognitive processes such as knowledge externalization and elaboration (Webb, 1989, 1991), knowledge elicitation (King, 1991, 1999), conflict resolution (Doise & Mugny, 1984), and knowledge negotiation (Fischer & Mandl, 2005; Roschelle & Teasley, 1995; Teasley, 1997). These so- cio-cognitive processes allow elicitation of knowledge, open up multiple perspectives and foster new opportunities to challenge one’s understanding by taking into account peers’ perspectives. However, it is not sufficient to place students in a group rather than alone and expect them to engage in these effective collaborative learning behaviors (Dillenbourg, 1999; Slavin, 1983; Sollers, Lesgold, Linton, & Goodman, 1999). Past research reports that co-learners do not spontaneously engage in efficient social interac- tions and cognitive processes (Cohen, 1994; Fischer & Mandl, 2005; Kreijns, Kirschner, & Jochems, 2003; Weinberger, Ertl, Fischer, & Mandl, 2005). On the social level, knowledge co-construction may be impeded, because learners often opt for quick consensus instead of building on each others’ contributions and establishing shared conceptions of a problem (cf. Chinn & Brewer, 1993; Clark, Weinberger, Jucks, Spitulnik, & Wallace, 2003; Nastasi & Clements, 1992). On the cognitive level, learners sometimes disregard strate- gies, theories or specific aspects of collaborative learning tasks (Crook, 1994). For instance, learners rarely seem to follow an ade- quate sequence of problem-solving steps and often adopt the first solution that fits their needs (Chinn, O’Donnell, & Jinks, 2000): learners often digress, oversimplify and orient themselves towards minimal requirements of collaborative learning tasks. Thus, the de- sired effects of collaborative learning often fail to emerge. Learners construct inaccurate knowledge, which they cannot apply and are not enabled to take multiple perspectives on a subject matter (e.g., Tudge, 1989). The main reason for these failures is that the construction of a shared understanding is a demanding task that relies on complex cognitive and social processes. These include cognitive processes such as the active use of prior knowledge, the deep processing of new information, and elaborative social activities (Elshout-Mohr, Hout-Wolters, & van, 1995). The active use of prior knowledge al- lows learners to process new information and effectively integrate it with their conceptual framework (Gijlers, 2006; Joshua & Dupin, 0747-5632/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.chb.2010.05.032 * Corresponding author. Tel.: +41 21 624 33 37; fax: +41 21 693 60 70. E-mail addresses: [email protected] (M. Sangin), [email protected] (G. Molinari), marc-antoine.nuessli@epfl.ch (M.-A. Nüssli), pierre.dillenbourg@ epfl.ch (P. Dillenbourg). URL: http://craft.epfl.ch (M. Sangin). Computers in Human Behavior 27 (2011) 1059–1067 Contents lists available at ScienceDirect Computers in Human Behavior journal homepage: www.elsevier.com/locate/comphumbeh

Transcript of Facilitating peer knowledge modeling: Effects of a knowledge awareness tool on collaborative...

Computers in Human Behavior 27 (2011) 1059–1067

Contents lists available at ScienceDirect

Computers in Human Behavior

journal homepage: www.elsevier .com/locate /comphumbeh

Facilitating peer knowledge modeling: Effects of a knowledge awareness toolon collaborative learning outcomes and processes

Mirweis Sangin *, Gaëlle Molinari, Marc-Antoine Nüssli, Pierre DillenbourgEcole Polytechnique Fédérale de Lausanne (EPFL) – School of Computer and Communication Sciences, 1015 Lausanne, CH, Switzerland

a r t i c l e i n f o

Article history:Available online 6 July 2010

Keywords:Computer supported collaborative learningGroup awarenessKnowledge awarenessPeer knowledge modelingSpatially distributed groups

0747-5632/$ - see front matter � 2010 Elsevier Ltd. Adoi:10.1016/j.chb.2010.05.032

* Corresponding author. Tel.: +41 21 624 33 37; faxE-mail addresses: [email protected] (M. Sangin

(G. Molinari), [email protected] (M.-A.epfl.ch (P. Dillenbourg).

URL: http://craft.epfl.ch (M. Sangin).

a b s t r a c t

We report an empirical study where we investigated the effects, on the collaborative outcomes and pro-cesses, of a cognition-related awareness tool providing learners with cues about their peer’s level of priorknowledge. Sixty-four university students participated in a remote computer-mediated dyadic learningscenario. Co-learners were provided (or not) with a visual representation of their peer’s level of priorknowledge through what we refer to as a knowledge awareness tool (KAT). The results show that, pro-viding co-learners with objective cues about the level of their peer’s prior knowledge positively impactslearning outcomes. In addition, this effect seems to be mediated by the fact that co-learners providedwith these objective cues become more accurate in estimating their partner’s knowledge – accuracy thatpredicts higher outcomes. Analyses on the process level of the verbal interactions indicate that the KATseems to sensitize co-learners to the fragile nature of their partner’s as well as their own prior knowledge.The beneficial effect of the KAT seems to mainly rely on this induction of epistemic uncertainty thatimplicitly triggers compensation socio-cognitive strategies; strategies that appear to be beneficial tothe learning process.

� 2010 Elsevier Ltd. All rights reserved.

1. Introduction

A large body of research has shown that collaborative learning ismore effective than individual learning, but not systematically. Theeffectiveness of collaborative learning mainly relies on the qualityof social interaction during collaboration (Dillenbourg, 1999;Dillenbourg, Baker, Blaye, & O’Malley, 1996; Suthers, 2006). Collab-orative learning is an added-value when co-learners actively con-struct shared knowledge through socio-cognitive processes suchas knowledge externalization and elaboration (Webb, 1989,1991), knowledge elicitation (King, 1991, 1999), conflict resolution(Doise & Mugny, 1984), and knowledge negotiation (Fischer &Mandl, 2005; Roschelle & Teasley, 1995; Teasley, 1997). These so-cio-cognitive processes allow elicitation of knowledge, open upmultiple perspectives and foster new opportunities to challengeone’s understanding by taking into account peers’ perspectives.However, it is not sufficient to place students in a group rather thanalone and expect them to engage in these effective collaborativelearning behaviors (Dillenbourg, 1999; Slavin, 1983; Sollers,Lesgold, Linton, & Goodman, 1999). Past research reports thatco-learners do not spontaneously engage in efficient social interac-

ll rights reserved.

: +41 21 693 60 70.), [email protected]üssli), pierre.dillenbourg@

tions and cognitive processes (Cohen, 1994; Fischer & Mandl, 2005;Kreijns, Kirschner, & Jochems, 2003; Weinberger, Ertl, Fischer, &Mandl, 2005). On the social level, knowledge co-construction maybe impeded, because learners often opt for quick consensus insteadof building on each others’ contributions and establishing sharedconceptions of a problem (cf. Chinn & Brewer, 1993; Clark,Weinberger, Jucks, Spitulnik, & Wallace, 2003; Nastasi & Clements,1992). On the cognitive level, learners sometimes disregard strate-gies, theories or specific aspects of collaborative learning tasks(Crook, 1994). For instance, learners rarely seem to follow an ade-quate sequence of problem-solving steps and often adopt the firstsolution that fits their needs (Chinn, O’Donnell, & Jinks, 2000):learners often digress, oversimplify and orient themselves towardsminimal requirements of collaborative learning tasks. Thus, the de-sired effects of collaborative learning often fail to emerge. Learnersconstruct inaccurate knowledge, which they cannot apply and arenot enabled to take multiple perspectives on a subject matter(e.g., Tudge, 1989).

The main reason for these failures is that the construction of ashared understanding is a demanding task that relies on complexcognitive and social processes. These include cognitive processessuch as the active use of prior knowledge, the deep processing ofnew information, and elaborative social activities (Elshout-Mohr,Hout-Wolters, & van, 1995). The active use of prior knowledge al-lows learners to process new information and effectively integrateit with their conceptual framework (Gijlers, 2006; Joshua & Dupin,

1060 M. Sangin et al. / Computers in Human Behavior 27 (2011) 1059–1067

1987). In order to communicate efficiently and engage actively inthe construction of a shared understanding, co-learners have toeffectively imagine or project how much their partner knows andunderstands (Crook, 1994). In natural face-to-face collaboration,learners have a large set of cues at their disposal to estimate theirpeers’ level of understanding. However, in distant collaborationsettings, these contextual awareness cues are either missing orare seriously limited. Socio-technical solutions have been proposedto make up for these limitations and are referred to as knowledgeawareness tools.

The aim of this article is twofold. First, we report on an empir-ical study where we investigated the effects, on the collaborativelearning outcomes and processes of a knowledge awareness toolproviding co-learners with objective cues about their peer’s priorknowledge. Second, this paper aim to shed light on the social andcognitive processes underlying the effects of the knowledge aware-ness tool on the collaborative learning processes.

1.1. The importance of knowledge awareness for communication

In collaborative learning settings, learners are simultaneouslyconfronted with their own and their partners’ prior knowledge.Different combinations of prior knowledge can trigger differentinteractions and learning processes (Gijlers, 2006). If two peersprior knowledge substantially overlaps, less elaboration and nego-tiation may be needed. When one of the peer knows what her part-ner does not know, social interaction opens opportunities for themore knowledgeable to fill the peer’s knowledge gaps. If both peersshare the same knowledge gaps, a pooled ignorance phenomenonmight take place (Xin, 2002). Therefore, awareness of peers’ leveland quality of knowledge is an important element of efficient so-cial interaction.

This idea has been thoroughly investigated in the fields of psy-cholinguistics (Clark & C.R, 1981; Krauss & Fussell, 1991). For Kraussand Fussell (1991), messages are formulated in order to be under-stood by a specific audience, and the speaker needs to take into ac-count what that audience does and does not know. They suggestthat communicators draw on two distinct sources of informationto formulate their messages: their prior knowledge and the currentfeedback. These are dynamically related and feed each other. Fur-thermore, this process of ‘perspective taking’ is necessarily probabi-listic and that there is no simple mechanism to identify withcertainty what is mutually known (Krauss & Fussell, 1991). Studiesshowed that people are biased in the direction of their own knowl-edge when estimating their audience’s knowledge (Chi, Siler, &Jeong, 2004; Nickerson, 1999; Nickerson, Baddeley, & Freeman,1987; Ross, Greene, & House, 1977; Steedman & Johnson-Laird,1980). More knowledgeable people are more likely to overestimatethe audience’s knowledge while less-knowledgeable have propen-sity to underestimate it. Consequently, being aware of one’s ownand one’s partner’s prior knowledge about specific content is a cru-cial element of effective knowledge co-construction.

Research on expert-layperson communication showed that ex-perts tend to either overestimate or underestimate laypersons’knowledge (Bromme, Jucks, & Runde, 2005; Nückles & Bromme,2002; Wittwer, Nückles, & Renkl, 2008). Wittwer et al. (2008) in-duced experts to systematically over- or underestimate laypersons’knowledge and subsequently analyzed the influence on communi-cation outcomes. Results showed that the experts’ biased estimationof laypersons’ knowledge impaired the laypersons’ understanding.Chi et al. (2004) found that tutors attributed a higher level of knowl-edge to students than they actually had. These results are in-linewith previous findings on the ‘‘illusion of mutuality” (Krauss &Fussell, 1991; Nickerson, 1999). People are biased toward theirown knowledge and understanding when they assess the knowl-edge of others. This bias usually occurs when contextual cues are

missing or are seriously impaired. Therefore, compensation solu-tions are needed to make up for these limitations that may takethe form of socio-technical supports, making available informationabout the knowledge of the partners (Dehler, Bodemer, Buder, &Hesse, 2009; Engelmann, Dehler, Bodemer, & Buder, 2009; Molinari,Sangin, Dillenbourg, & Nüssli, 2009; Sangin, 2009; Sangin, Nova,Molinari, Dillenbourg, 2007).

1.2. Knowledge awareness in CSCL

Researchers in the CSCL field proposed socio-technical tools tosupport knowledge awareness in distributed groups. For instance,Ogata and Yano (2000) proposed the notion of knowledge aware-ness to increase opportunities for shared knowledge constructionin an open-ended collaborative learning environment. The knowl-edge awareness tool (KAT hereafter) designed and investigated byOgata and colleagues (Ogata, Matsuura, & Yano, 2000; Ogata &Yano, 2000), uses as an input the interaction of learners with spe-cific pieces of knowledge in the shared environment. Although thistype of information may trigger collaboration opportunities, themain limitation is that it does not provide an accurate representa-tion of others’ knowledge. It rather provides cues about mutualinterest with regards to a specific knowledge content. We shalltherefore classify this type of KAT as activity-based KAT.

Leinonen and Järvelä (2006) proposed and tested a KAT using avisualization of self-reported cues about group-members’ under-standing of the task. Another example of subjective KAT is designedby Dehler and colleagues (Dehler, Bodemer, & Buder, 2007; Dehler,Bodemer, Buder, & Hesse, 2011; Dehler et al., 2009). The authorstested the effects, on the performance of learners, of self-reportedcues about self and the partner’s level of text comprehension. Theresults showed that the KAT positively affects the message produc-ers’ high level learning outcomes. Learners provided with the KATreported more elaborated knowledge construction. The tool alsotriggered effective knowledge elicitation processes. Finally, learnersadapted their explanations to the specific knowledge of their part-ner rather than the general knowledge. As a last example, in a net-based expert-layperson communication study, Nückles and Stürz(2006) provided experts with information about the novice’sknowledge on a specific inquiry topic, previously reported by thenovices through a survey. The results showed that experts providedwith such a tool, planned and communicated their answers moreefficiently to the layperson’s inquiry. The advantage of subjectiveKATs is that the information displayed is easily gathered throughsurveys and self-reported questionnaires. The drawback is thatthe subjective awareness cues are subject to bias and misestima-tions, as they highly rely on the learners’ metacognitive capacities.

A last type of KAT design – we refer to as objective KATs – usesmore objective assessments of peers’ knowledge. To our knowl-edge, only one attempt to design objective KATs has been under-taken so far. Engelmann and Tergan (2007) asked learners in atriadic concept-map building scenario to build a concept-map ofthe problem individually. During the subsequent collaborativeproblem-solving phase, participants of the experimental conditionwere provided with their own as well as their partners’ individualconcept-maps. The results showed that participants provided withthe tool achieved higher problem-solving performance. Engelmannet al.’s tool is the most courageous attempt to objectively representpartners’ knowledge. The main advantage of this type of tool is thatthey provide objective information about peers’ knowledge in theform of an external representation of their mental model. How-ever, the implementation of these forms of awareness tools is te-dious and costly in terms of design and learning activity, makingthem less ‘‘design-friendly”. Furthermore, the information pro-vided by these tools is difficult and effortful for learners tointerpret.

M. Sangin et al. / Computers in Human Behavior 27 (2011) 1059–1067 1061

This overview of CSCL research on knowledge awareness showsthat KATs can be an efficient and effective social and cognitive sup-port to help the grounding and coordination processes when natu-ral awareness cues are missing or are limited. Carefully designedKATs can tailor effective verbal interaction processes such as elab-orative talk, negotiation of knowledge and consensus building.However, seldom research focused on empirically understandingthe social and cognitive mechanics underlying these tools.Engelmann et al. (2009) propose a descriptive model of thefunctionalities of KATs. This model conceptualizes collaborativelearning primarily as an exchange of information between twolearners. KATs, by externally representing partners’ knowledge,provide context information to support this exchange of informa-tion ‘‘by complementing the natural process of establishing knowl-edge awareness with an externally referable basis” (p. 953).

Nückles, Wittwer, and Renkl (2005) investigated the underlyingcognitive processes used by experts provided with a KAT to planand communicate efficient solutions to laypersons. Two alternativecognitive explanations have been tested. On one hand, the KATmay sensitize the experts to the layperson status of the audience,prompting them to carefully adjust their explanations to the typi-cal knowledge characteristics of a particular community of layper-sons (see for instance Nickerson, 1999). On the other hand, bypresenting experts with individuating information about a layper-son’s knowledge level, the assessment tool may enable the expertsto more quickly and accurately adapt their mental model of thespecific layperson’s knowledge. Their results suggested a specificadaptation rather than a purely sensitizing effect. Accordingly, byenabling learners to adapt their contributions to the learning part-ners and serving as a contextual help.

The expert-layperson communication paradigm implies asym-metric and unidirectional learning processes. Peer-to-peer collabo-ration implies bidirectional knowledge acquisition processes thatare rather symmetric or slightly asymmetric (asymmetry thatlearners are usually not aware of). In these settings, we can expectsome degree of mutual influence among the peers. Therefore, inthe context of peer-to-peer collaboration characterized by slightasymmetry in knowledge but not in status, we argue that the illu-sion of mutuality affects both peers and lead to suboptimal inter-action processes that may critically hinder the processes andoutcomes of collaborative learning. To be able to avoid the illusionof mutuality, learners need to have a rather precise idea of theirpartner’s level of prior knowledge. In order to successfully esti-mate the knowledge gap between them and their partner, theyalso need to have a rather precise idea of their own knowledge.However, it has been shown that learners have a tendency to besubject to an illusion of comprehension (Brown, 1987; Maki & Berry,1984; Renkl, 1997) or an illusion of competence (Koriat & Bjork,2005; Weinberger, 2003). Therefore, we argue that the very pres-ence of an objective KAT may have a sensitizing effect, by inherentlysensitizing learners to the fragile nature of their partners as well astheir own knowledge. This sensitization may trigger strategies tocope with the epistemic uncertainty generated that are beneficialto the collaborative learning processes. Rephrased differently, theKAT may act as a social metacognition tool triggering strategiesto cope with epistemic uncertainty such as knowledge verificationand negotiation and elaborative talk (Sangin, 2009). Accordingly,we argue that providing co-learners in a remote dyadic synchro-nous collaboration with an objective KAT may have a sensitizingeffect making co-learners aware of their partners and their ownknowledge gap.

1.3. Research questions and hypotheses

The present empirical study investigates the effects on collabo-rative learning outcomes and processes of an objective KAT, as well

as understanding the underlying social and cognitive processes.Co-learners in the experimental condition were provided withthe KAT, graphically representing their partner’s level of priorknowledge, whereas co-learners of the control condition werenot. We examined the effects of KAT on the collaborative learningoutcomes and processes as well as on the peer knowledge model-ing. Based on the literature presented hereinbefore, the followingresearch questions and hypotheses are formulated:

R1. Does providing co-learners with the KAT affect their knowl-edge acquisition performance?

H1. We expect the KAT to enhance co-learners learningperformance.

R2. Does providing co-learners with the KAT improve their mod-el of the partner’s actual knowledge, serving as a mediating factor inthe effects of the KAT on the knowledge acquisition performance.

H2. We expect the KAT to enhance the learners’ accuracy esti-mation of their peers knowledge, which serves as a mediator forthe knowledge acquisition performance.

R3. Does the KAT enhance the production of elaborative talk?H3. We expect the KAT to positively affect the production of

elaborated epistemic utterances.R4. Does the KAT orient the focus of collaboration towards

knowledge verification and negotiation?H4. We expect the KAT to positively impact the production of

knowledge verification utterances as opposed to task completionutterances.

R5. Does the KAT induce epistemic uncertainty and triggerstrategies to cope with it?

H5. We expect the KAT to positively affect the production ofepistemic uncertainty markers.

2. Method

2.1. Participants and design

Sixty-four first semester university students (18 women and 46men, mean age = 21.2 years) were recruited and remunerated toparticipate to the study. Learners with a high degree of knowledgeabout the instructional material (i.e. the neural transmission) werefiltered through a prior knowledge test and were excluded fromthe sample. The 64 participants were randomly assigned to 32dyads. Sixteen dyads were assigned to each of the following twoexperimental conditions: (1) the KAT condition, where the partic-ipants were provided with awareness cues about the level of priorknowledge of their peer; (2) the control condition, where the par-ticipants were not provided with this information. Peers withinsame pairs did not know each other before the experiment. Onepair of the control condition was excluded from the analyses be-cause technical issues occurred during the experimentation ses-sion that biased the session.

2.2. Instructional material

The instructional material consisted of an explanatory textabout the neurophysiologic phenomenon of the ‘‘neural transmis-sion,” based on textbook materials and developed with the help oftwo experts of the domain. The text was equally distributed amongthree different chapters: ‘‘the resting potential,” ‘‘the initiation ofthe action potential” and ‘‘the propagation of the action potential”.

2.3. Technical setup

We used an automated computer-based setup running on twoidentical computers distributed in two rooms. All phases of theprocedure were automated. During the collaborative phase, we

1062 M. Sangin et al. / Computers in Human Behavior 27 (2011) 1059–1067

used an online concept-map building software, CmapTools(�IHMC). During this phase, participants were provided with amicrophone headset in order to communicate through an audioconferencing tool (TeamSpeak�).

2.4. Procedure

The two peers were separated in two different rooms and wereinstalled in front of two identical computer setups running theinstructional material. The experimental session lasted about90 min and consisted of six phases including two main learningphases: an individual explanatory reading phase and a remote col-laborative concept-map building phase. We briefly describe themain phases:

2.4.1. Prior knowledge verificationTo detect and remove from data potential experts of the do-

main, we asked the participants to write down all they knew aboutthe phenomenon of neural transmission through an open-endedquestion. Participants had four minutes to answer the prior knowl-edge verification question, after they were automatically directedto an instruction page about the individual learning phase.

2.4.2. Individual test comprehension phaseDuring the second phase, participants were asked to carefully

read the instructional texts in order to understand and rememberas much as they could about the neural transmission. During thisphase, they were able to freely access the three texts by clickingon the corresponding buttons at the bottom of the screen. Theywere also free to read the texts in any order they wanted. Theyhad 12 min to complete this phase and were provided with acountdown in the form of a reverse-bar.

2.4.3. PretestAfter the individual reading phase, the first learning test was

administered. It consisted of an automated multiple-choice ques-tionnaire of 30 items. The questionnaire comprised two types ofquestions: six multiple-choice questions and 24 inference verifica-tion questions.

2.4.4. Collaborative concept-mapping phaseParticipants were provided with instructions about the collabo-

rative concept-map building phase and a short video tutorial(2:34 min) on how to use CmapTools�. During the collaborativetask, participants were invited to draw a collaborative concept-map about what they learned during the individual task. Thisphase lasted 20 min. Peers were able to communicate orally usinga headset, thanks to a vocal conferencing tool (TeamSpeak�). Dur-ing this phase, participants in the experimental condition wereprovided with the KAT on the bottom part of the screen (seeFig. 1). Participants of the control condition were not provided withthe KAT. Participants of both conditions received exactly the sameinstructions; the sole difference between the conditions was thepresence or absence of the KAT during the collaborative concept-mapping phase.

Fig. 1. knowledge awareness tool (KAT). Each bar represents the level of kn

2.4.5. PosttestAfter the collaborative task, the second learning test was

administered. It consisted of the same items as the pretest, pre-sented in a different order.

2.4.6. Peer’s knowledge estimation questionnaireAfter the posttest, participants were provided with the knowl-

edge estimation questionnaire. For each of the three chapters ofthe instructional material, we asked the participants to estimatetheir partner’s knowledge with a 7-point Likert scale ranging fromvery low to very high.

2.5. Experimental conditions and dependant variables

2.5.1. Experimental conditionsWe manipulated our main factor (KAT-factor) during the collab-

orative phase. Participants in the experimental condition were pro-vided with cues about their partner’s knowledge in the form of aknowledge awareness tool (KAT). The KAT was provided only dur-ing the collaborative concept-map building phase. It consisted of agraph representing only their partner’s scores (in percentage) onthe pretest with regards to the three chapters (see Fig. 1). The con-trol condition’s participants were not provided with the KAT. How-ever the space allocated to the concept-map construction was thesame across the two conditions.

2.5.2. Learning outcome measuresWe used the learning gain between the pretest and the posttest

to measure the learning gain. The pretest and posttest consisted ofthe same 30 items: six multiple-choice questions and 24 inferenceverification items. All items were designed to be comprehensionquestions, instead of memorization questions. Two multiple-choice questions and eight inference verifications were associatedto each of the three chapters. The multiple-choice questions in-cluded four answer propositions including one or more possiblecorrect answers. The minimum score for these items was 0 andthe maximum was 4. As for the inference verification items, partic-ipants had to decide if an assertion was true or false. The score was0 for incorrect answers and 1 for correct answers. The pretest scorefor every chapter ranged from 0 to 16 and the overall range for thetests was from 0 to 48 points. All items were validated by expertsof the domain and their internal validity and variability weretested in a pilot study.

We calculated the relative learning gain as the learning outcomesmeasure with respect to the three chapters. To take into account thevariability on the pretest and the relative margin of progression (orregression, given that some of the learners did actually report nega-tive learning gain scores) adapted formulas were used to computethe learning gain. The formulas used are as follows:

for i 2 f1;3g;RLGi ¼ðposttesti�pretestiÞðMax�pretestiÞ

h i�100; if ðposttesti�pretestiÞP 0

ðposttesti�pretestiÞðpretestiÞ

h i�100; if ðposttesti�pretestiÞ< 0

8><>:

RLGi = Learner’s relative learning gain score with regards to chapteri.pretesti = Learner’s pretest score with regards to chapter i.post-

owledge of the peer on the associated chapter of the instructional text.

M. Sangin et al. / Computers in Human Behavior 27 (2011) 1059–1067 1063

testi = Learner’s posttest score with regards to the chapter i.Max =the maximum score obtainable for chapter i. (i.e. 16)

The total relative learning gain is obtained by calculating themean relative learning gain for the three chapters. In the analysesreported hereafter, we use the total relative learning gain as alearning outcome measure and refer to it as RLG.

2.5.3. Knowledge modeling accuracyTo measure the quality of the model of the peer’s knowledge,

we calculated the accuracy of participants in their estimation oftheir peer’s knowledge about each of the three chapters, by com-paring their knowledge estimation to the peer’s actual posttest score.For each chapters, the knowledge modeling accuracy was computedby taking the absolute value of the difference between the stan-dard values of the participants’ knowledge estimation and theirpartner’s posttest score. The total knowledge modeling accuracy(KMA) score is calculated by performing the mean as follows.

for i ¼2 f1;3gX

KMAi ¼ ½1� jzðKEiÞ � zðpeers posttestiÞj�

KMAi = Learner’s knowledge modeling accuracy score with regardsto chapter i.KEi = Learner’s knowledge estimation with regards tochapter i.p_posttesti = the learner’s peer’s posttest score with re-gards to the chapter i.

2.5.4. Verbal process variablesDuring the collaborative concept-map building phase, we re-

corded the dyads’ verbal interactions. The verbal interactions weretranscribed verbatim. In order to test our hypotheses about the dif-ferences in terms of quality of interaction, we used quantitativecontent analysis methods (Strijbos, Martens, Prins, & Jochems,2006). We followed Strijbos et al.’s recommendations in our proce-dure. The segmentation and coding were separated in order to en-hance the reliability of the method. Utterances were used as an unitof analysis. An utterance was defined as a ‘‘unit of meaning” of con-ceptual understanding (Newman, Webb, & Cochrane, 1995; vanBoxtel, van der Linden, & Kanselaar, 2000). The proportion of agree-ment between two independent trained ‘segmenters’ on 13% of thecorpus (4 randomly selected pairs) allowed us to compute the reli-ability of segmentation. The level of agreement reached 86%.

The segmented corpus was first and foremost divided into fourmain categories regarding the informational content of the utter-ances: domain knowledge, knowledge quality, activity regulationand miscellaneous. The current paper focuses mainly on domainknowledge and knowledge quality utterances.

2.5.5. Domain knowledge utterancesDomain knowledge units are utterances related to epistemic

activity (i.e. the subject matter). We used a 2 � 2 two-dimensionalcoding scheme to further analyze the domain knowledge utter-ances with regards to the informational content and the qualityof utterances.

We made the assumption that KAT condition’s co-learnersshould be more focused on knowledge verification and negotiation

Table 1Examples of utterances for the different domain knowledge categories.

Dim. 1 Dim. 2 Example of

Knowledge verification and negotiation Elaborated I think thatThe ‘Na–K

Non-elaborated I rememberWhat is the

Task completion Elaborated I’ll add a neWhat do w

Non-elaborated I’ll add ‘myWhat do yo

and less focused on task completion. To test this assumption, wedistinguished the knowledge domain utterances expressed in theform of knowledge verification and negotiation from utterances ex-pressed in the form of task completion. As the learning activitycan be approximated as collaboratively translating the knowledgeacquired in the text into a concept-mapping formalism, learnershave two different ways to introduce and discuss content: onone hand, they can express the information in a form close to thetexts they read and their own mental model (e.g. ‘‘the neuronhas an axon”). On the other hand, they can formulate the informa-tion in terms of its concept-map translation (e.g. ‘‘we should addtwo boxes called ‘neuron’ and ‘axon’ and connect it with a ‘to have’link”). Given that the learners received as an instruction to build aconcept-map of their understanding of the instructional material,we argue that learners subject to the illusion of mutuality are fo-cused merely on completing the task rather than negotiating theknowledge. Therefore, they should produce more utterancesphrased in the concept-map form. Accordingly, we refer to utter-ances expressed in the ‘‘concept-map” form as task completionand utterances in the ‘‘non-concept-map” form as knowledge veri-fication and negotiation utterances (see Table 1).

To test a potential effect of the KAT on the quality of verbalinteractions, we dissociated non-elaborated knowledge domainutterances, from elaborated utterances. Utterances were coded asnon-elaborated when they were providing simple and uniqueinformation. Utterances were coded as elaborated when the speak-er produced causal relations, conditions or disjunctions, or defini-tions and descriptions (see Table 1).

2.5.6. Uncertainty markersTo test our assumption of a potential sensitizing effect of the

KAT, we analyzed the knowledge domain utterances to see if theywere accompanied by uncertainty markers (see Brennan & Ohaeri,1999 for further details). We counted the number of uncertaintymarkers produced by each learner. If an utterance included morethan one uncertainty marker, we counted it only once. We didnot take into account utterances where the learners reported a fail-ure to remember a piece of information, without providing newelements (e.g. ‘‘There was something about this potential, but Idon’t remember”). We detected an overall of 156 uncertaintymarkers in the knowledge domain utterances. Among these uncer-tainty markers, we detected 64 uncertainty markers for the controlcondition and 92 for the KAT condition. These uncertainty markerswere formulated in various ways. Table 2 presents a taxonomy ofthe main uncertainty markers and their number of occurrences.

3. Results

3.1. Unit of analysis and treatment checks

In order to define the unit of analysis for the statistical testsconducted, we performed intraclass correlation (ICC) between sub-ject A and subject B’s data (A and B being peers of the same pair), in

utterance

having the myelin sheets on the axon allows the transmission to go faster.pump’ was kicking out the K+ and pulling the Na+ in, right?

that the resting potential was negative‘saltatory stuff’?

w ‘‘myelin,” here close to the ‘‘transmission efficacy;” and we may relate them.e add as value for the potential, 65mv or 0.65mv?elin sheet’ down hereu want me to add in ‘axon’?

Table 2Taxonomy of principal uncertainty markers (translated from French).

Tentative English translation Occurrence

. . . I don’t remember well but. . . 6

. . . I remember that . . . 4According to what I remember. . . 14. . . I didn’t understand well. . . 8. . . but I don’t know. . . 36. . . I don’t know anymore. . . 16. . . It seems to me that . . . 15. . . I believe that . . . (used as ‘‘I think that”) 29I think that. . . 2. . .If I’m not wrong. . . 4. . . stuff like that . . . 13Miscellaneous 14

1064 M. Sangin et al. / Computers in Human Behavior 27 (2011) 1059–1067

order to find out if a group level effect introduced variance in thedata (see Kenny, Kashy, & Cook, 2006 for further discussion). Wecomputed the intraclass correlations for the main dependent vari-ables (i.e. RLG and the KMA), as well as the main process variables.No evidence of a group effect was supported by the ICC analyses.Therefore, individual-level measures are used as unit of analysiswith standard inferential statistical methods (t-tests and regres-sion analysis). It’s worth mentioning that the all the conditionsfor the use of variance analyses were tested and verified (see San-gin, 2009 for further details).

3.2. The effect of KAT on learning outcomes

T-tests were performed to test the effect of the KAT on learningoutcomes. Due to technical problems (client crash during experi-ment), the data from two dyads of the KAT condition was not avail-able and had to be removed. We tested the effect of the KAT on theRLG. The t-test reported a significant difference between the KATcondition participants (M = 13.4) and the control condition partic-ipants [M = 3.6, t(1, 60) = 2.73, p < .01, Cohen’s d = 0.7]. Hence, pro-viding learners with cues about the prior knowledge of theirpartner appears to enhance their learning gain. This result confirmsour hypothesis H1. The presence of the KAT positively affects theco-learners learning performance.

3.3. The mediating effect of peer knowledge modeling

We examined the peer knowledge modeling accuracy (KMA) asvariable potentially mediating the effect of the KAT factor on theRLG. Regressions statistical methods where used to verify themediation effect (see Baron & Kenny, 1986) three steps wereundertaken to assess mediating effects using regression methods.Simple linear regression analyses were performed to verify thetwo first steps by taking the KMA as a potential mediator, theKAT factor as the independent variable and the RLG as the depen-dent variable. The linear regression confirmed that the KMA is sig-nificantly related to the KAT factor (b = .381, p < .01, r2 = .15). TheKAT factor was also significantly and positively related to the RLG(b = .332, p < .01, r2 = .11). In the last step we tested in a multipleregression the relation between KAT factor and the RLG whencontrolling for KMA. The multiple regressions showed that theKAT factor was no longer a significant predictor (b = .210; p > .1)whereas the KMA was still a significant predictor (b = .32, p < .01).Thus, it can be concluded that KMA did mediate the KAT factor’seffect on the RLG. The Sobel significance test for indirect effects(Sobel, 1982) was significant [z = 1.99, p < .05]. These results sup-port H2. The precision in the partner’s knowledge estimationseems to be a dominant mediator in the effect of the KAT on thelearning gain.

3.4. The effects of KAT on elaborative talk

We tested the effects of KAT on the production of elaborativetalk. To do so, we compared the ratio of elaborated utterances withrespect to the domain knowledge utterances. The results showed apositive and significant effect of the KAT on the ratio of elaboratedutterances produced. KAT condition learners produced significantlymore elaborated utterances (M = 0.27, SD = 0.07) than learners ofthe control condition (M = 0.20, SD = 0.08; t(60) = 3.4, p < .01; Co-hen’s d = .8). Accordingly, our hypothesis H3 is verified. The KATseems to positively affect the production of elaborative epistemictalk.

3.5. The effects of KAT on knowledge verification and negotiation

To test if the KAT oriented the collaboration process towardsknowledge verification and negotiation, we computed the ratio ofknowledge verification and negotiation utterances with respect tothe overall knowledge domain utterances. The results reported apositive and significant effect of the KAT on the ratio of knowledgenegotiation utterance. KAT condition learners produced a signifi-cantly higher ratio of knowledge verification and negotiation utter-ances (M = 0.31, SD = 0.08) than learners of the control condition(M = 0.25, SD = 0.09; t(60) = 2.8, p < .01; Cohen’s d = .8). This con-firms our hypothesis H4. The KAT seems to orient the epistemic ver-bal interactions towards knowledge verification and negotiation.

3.6. The sensitizing effect of KAT

We performed an independent sample t-test to test potential ef-fects of the KAT on the production of uncertainty markers. The testreported a positive and significant effect of the KAT condition onthe production of uncertainty markers. Learners in the KAT condi-tion produced significantly more uncertainty markers (M = 3.54,SD = 3) than learners in the control condition [M = 2.00, SD = 1.83,t(39.1) = 2.26, p < .05; Cohen’s d = .64).

We also analyzed the relation between the production of uncer-tainty markers and the production of specific types of epistemicutterances. First, the results reported a positive and significant rela-tion between the production of uncertainty markers and the pro-duction of elaborated epistemic utterances (r(60) = .46; p < .001).Furthermore, the results also reported a significant correlation be-tween the production of uncertainty markers and the productionof knowledge verification and negotiation utterances (r(60) = .57;p < .001). This verifies our hypothesis H5. The KAT seems to sensi-tize co-learners to the fragile nature of their prior knowledge andtrigger epistemic uncertainty.

4. Discussion

In the present study, we investigated the effects of a cognition-related knowledge awareness tool on the remote collaborativelearning outcomes and processes. The results showed that provid-ing co-learners with objective cues about their partner’s priorknowledge level enhances learning gain. In addition, the resultsshowed that learners of the KAT condition performed better in esti-mating their partner’s knowledge after the collaboration phase.Furthermore, the accuracy in estimating the partner’s knowledgewas positively related to learning outcomes. Finally, the resultsprovide evidence of this mediation effect. Providing learners withcues about their partner’s prior knowledge allows them to be moreaccurate in estimating their partner’s actual knowledge and byextension, enhances their learning gain. One explanation is thatthe KAT helps learners to build a better representation of theirpeer’s knowledge. Establishing a better understanding of what

M. Sangin et al. / Computers in Human Behavior 27 (2011) 1059–1067 1065

their partner knows and needs to know allows them to adapt moreefficiently to his/her knowledge and provide explanations at anappropriate elaboration level; processes that have been proven toguarantee efficient learning outcomes (King, 1999; Teasley, 1995;van Boxtel et al., 2000; Webb, 1989). These findings are in-linewith previous works (i.e. Krauss & Fussell, 1991; Lockridge & Bren-nan, 2002), which highlight the importance of processes such asaudience design and perspective taking for collaborative knowl-edge construction and coordination.

Further analyses of the verbal interactions showed that the KATenhances the production of elaborated utterances. The KAT seemsto catalyze the production of elaborative talk. This result supportsthe assumptions of audience design and perspective taking. Learnersprovided with cues about their partner’s knowledge seem to makean extra effort to provide elaborated explanations and/or elicittheir knowledge in an elaborated fashion.

Another assumption was that the presence of the KAT may ori-ent the collaboration processes towards knowledge elicitation andnegotiation, instead of mere task completion. The assumption wasverified by the results. Learners of the control condition seems tobe more focused on task completion and produce what Mercercalls cumulative talk (Mercer, 1996). Interactions in this conditionare mainly characterized by the accumulation of already knownconcepts, fast consensus building, and a collaboration modeoriented towards quick task completion (i.e. building the con-cept-map). This tendency can be accounted for by two underlyingexplanations. First, learners in the control condition may be lessfocused on knowledge verification and negotiation utterances be-cause they do not feel the need to verify the veracity of their pro-duction that they overconfidently take for granted. Second, they donot feel the need to negotiate a piece of information beforehandbecause they overconfidently take it as shared by the peer.

In contrast, KAT condition’s learners are oriented towards knowl-edge verification and negotiation. The interaction style seems to bebetter described by what Mercer (1996) calls ‘‘exploratory talk”.More effort is allocated to first negotiate knowledge and then collab-oratively translate it into its concept-map form. This effort is charac-terized by rich interactions where learners engage critically,constructively and ‘‘transactively” in each other’s contributionsand collaborate in the construction of a shared understanding (seeSangin, 2009 for further analyses). These results seem to be in-linewith the sensitizing effect hypothesis. Accordingly, the presence ofthe KAT may have triggered an elaborative mode of communicationas a strategy to cope with knowledge uncertainty. The KAT may havemade co-learners aware of the fragile nature of their understanding.On the dyad level, this sensitization effect may have triggered a modeof collaboration relying on strategies to diminish uncertainty suchas elaborative knowledge verification and negotiation.

We investigated the assumption that the sensitizing effect of theKAT may trigger an extensive use of uncertainty markers andhedges to attenuate learners’ propositions and explanations. Theresults showed that learners of the KAT condition indeed producedsignificantly more uncertainty markers than learners of the controlcondition. Furthermore, the results showed that the production ofuncertainty markers is positively related to the production ofknowledge verification and negotiation utterances. This seems tosuggest that there is a relation between the production of elabora-tive talk and the production of uncertainty markers. One must staycareful not to interpret such relations as causal. However, qualita-tive investigations support the idea that uncertainty markers aremainly produced when the speaker is providing explanations andmaking statements (see Sangin, 2009). These investigations showedthat as an implicit effect, the KAT allows learners to become awareof their own as well as their peer’s knowledge gaps. This implicit ef-fect is mainly explained by the fact all users exchanged their scoresat the beginning of the session. The awareness of these knowledge

gaps seems to have triggered epistemic uncertainty and conse-quently, implicit strategies to cope with it. Learners seem to havebecome aware that their prior knowledge is questionable, whichimplicitly implies the production of more hedges and uncertaintymarkers. Speakers tend to take a provisional stance when providinginformation and provide pragmatic cues of their self-assessed trust-fulness about the quality of their prior knowledge and understand-ing (see Brennan and Ohaeri, 1999). Besides, this may serve thepeers in their process of building and maintaining a more accuratemodel of the partners’ knowledge.

Another explanation is that the production of uncertainty mark-ers may also be motivated by a face-saving strategy. The KAT mayput the co-learners in a social comparison situation (see for in-stance Festinger, 1954). Co-learners, more specifically those thatappear to be more knowledgeable, may produce more uncertaintymarkers to save their own face and/or their peer’s face. Severalcues in the dialogue analysis suggest that co-learners tend to avoidentering an overt social status asymmetry situation.

In summary, the production of uncertainty markers seems to bebeneficial to collaborative learning. Uncertainty markers implicitlyinvite the listener to critically engage in the speaker’s epistemic pro-ductions. Hedges are also important cues in apprehending the ‘‘col-laborativeness” of verbal interactions. As claimed by Brennan andOhaeri, 1999, one role of hedging is to grant the addressee the oppor-tunity to reject or modify an utterance through processes such ascasting doubt or adding cautious support. In accordance, we believethese hedges and doubt expressions to be ‘‘markers of provisionali-ty” that allow speakers to take a stance that invites peers to questionand/or acknowledge statements and contributions, processes thathave proven to be beneficial for collaborative learning (Dillenbourg,1999; Stahl, 2002, among others). It has been shown that learnershave a tendency to be subject to an illusion of comprehension (Brown,1987; Maki & Berry, 1984; Renkl, 1997) or an illusion of competence(Koriat & Bjork, 2005). The important challenge for educationresearchers and practitioners is to find solutions to compensate forthese inherent suboptimal processes. Effective learning occurs whena student can target a valuable reason for learning and establishing‘‘ownership of knowledge” (Goodnow, 1988). Therefore, even col-laborative learning is not an effective method if co-learners aremerely motivated to ‘‘satisfice” a shallow knowledge, characterizedby potential false beliefs. We argue that sensitizing learners to thepotential gaps in their partner’s prior knowledge may make up forwhat has been referred to as the illusion of evidence (Bromme et al.,2005) or the curse of knowledge (see Birch & Bloom, 2007). Becomingaware of potential discrepancies between one’s own and the peer’sknowledge might encourage learners to more carefully plan theircontributions as well as to assess the quality of the partner’s contri-butions. This may have accounted both for the augmentation of elab-orative talk (as a desire to effectively help the peer) and theproduction of knowledge verification and negotiation utterances(as a desire to reassure the quality of one’s knowledge and sharedknowledge). Rephrased differently, providing co-learners with feed-back about the quality of the partner’s knowledge (and indirectlyabout their own knowledge), appears to be an effective indirectway of orienting co-learners towards efficient knowledge negotia-tion processes. We conclude that a certain level of epistemic uncer-tainty leads to uncertainty reduction strategies that are beneficial tocollaborative learning. Of course, above a certain threshold, uncer-tainty would become counterproductive (see for instance Mäkitalo,Weinberger, Häkkinen, Järvelä, & F., 2005).

5. Conclusions

In the present paper, we explored the effects of a cognition-re-lated awareness tool providing co-learners with objective cues

1066 M. Sangin et al. / Computers in Human Behavior 27 (2011) 1059–1067

about their partner’s level of prior knowledge on the outcomes andprocesses of remote synchronous collaborative learning. As statedby Engelmann et al. (2009), research on knowledge awareness andassociated tools is still at its early stage. First results nonetheless,highlight the importance of this concept and these tools for theefficiency and effectiveness of CSCL processes. Our results consoli-date previous findings of the effects of knowledge awareness toolson collaboration outcomes. Providing co-learners with this knowl-edge awareness support positively impacts learning outcomes.However, seldom research dug deeper into the meditational effectsand the socio-cognitive processes underlying this positive effect.The present study is a first attempt to understand the effects ofknowledge awareness on the social and cognitive level processesof collaboration.

Our study opens some promising new perspective for the studyof group awareness, and more specifically knowledge awareness.However, as a first endeavor to investigate the knowledge aware-ness and peer knowledge modeling processes in their full effects,we essentially took the ‘‘learning by explaining” stance by investi-gating how implicitly encouraging co-learners to be more activeand proactive in terms of knowledge negotiation, affects individualknowledge acquisition. One limitation worth pointing out is thatthe ‘‘learning by being explained” aspect has been neglected anddeserves a proper focus as well. Another characteristic of the pres-ent study is that the KAT presents information about the level ofpeer’s prior knowledge instead of the quality. Future researchshould indeed focus on the effect of qualitative information aboutthe peers’ knowledge. However, it is important to point out that atradeoff must be found between providing precise information andthe risk of overwhelming the learners’ cognitive treatment pro-cesses at the expense of learning. Finally, we studied the effect ofthe KAT in a specific context where the peers did not know eachother beforehand. It is reasonable to expect that the effect of theKAT may drastically change in a situation where co-learners sharea common history. Dispositional information about the peer is ex-pected to interact with the objective cues provided by the KAT andmay lead to different results. Therefore, further work should focuson the systematic investigation of knowledge awareness support indifferent social and physical contexts as well as with different col-laborative tasks.

References

Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction insocial psychological research: Conceptual, strategic, and statistical consider-ations. Journal of Personality and Social Psychology, 51, 1173–1182.

Birch, S. A., & Bloom, P. (2007). The curse of knowledge in reasoning about falsebeliefs. Psychological Science, 18(5), 382–386.

Brennan, S. & Ohaeri, J. (1999). Why do electronic conversations seem less polite?The costs and benefits of hedging. In Proceedings, international joint conferenceon work activities, coordination, and collaboration (WACC ’99) (pp. 227–235). SanFrancisco, CA.

Bromme, R., Jucks, R., & Runde, A. (2005). Barriers and biases in computer-mediatedexpert-layperson communication. In R. Bromme, F. W. Hesse, & H. Spada (Eds.),Barriers and biases in computer mediated knowledge communication – and howthey may be overcome (pp. 89–118). New York: Springer.

Brown, A. (1987). Metacognition, executive control, self-regulation, and other moremysterious mechanisms. In F. E. Weinert & R. H. Kluwe (Eds.), Metacognition,motivation and understanding. New Jersey: Lawrence Erlbaum Associates.

Chi, M. T. H., Siler, S., & Jeong, H. (2004). Can tutors monitor students’ understandingaccurately? Cognition and Instruction, 22, 363–387.

Chinn, C. A., & Brewer, W. F. (1993). The role of anomalous data in knowledgeacquisition: A theoretical framework and implications for science instruction.Review of Educational Research, 63(1), 1–49.

Chinn, C. A., O’Donnell, A. M., & Jinks, T. S. (2000). The structure of discourse incollaborative learning. The Journal of Experimental Education, 69(1), 77–97.

Clark, H. H., & Marshall, C. R. (1981). Definite reference and mutual knowledge. In A.K. Joshi, B. L. Webber, & I. A. Sag (Eds.), Elements of discourse understanding.Cambridge University Press.

Clark, D., Weinberger, A., Jucks, I., Spitulnik, M., & Wallace, R. (2003). Designingeffective science inquiry in text-based computer supported collaborativelearning environments. International Journal of Educational Policy, Research &Practice, 4(1), 55–82.

Cohen, E. G. (1994). Restructuring the classroom: Conditions for productive smallgroups. Review of Educational Research, 64(1), 1–35.

Crook, C. (1994). Computers and the collaborative experience of learning.Routledge, London, UK.

Dehler, J., Bodemer, D., & Buder, J. (2007). Fostering audience design of computer-mediated knowledge communication by knowledge mirroring. In C. Chinn, G.Erkens, & S. Puntambekar (Eds.), Proceedings of the seventh computer supportedcollaborative learning conference. New Brunswick: International Society of theLearning Sciences, Inc.

Dehler, J., Bodemer, D., Buder, J., & Hesse, F. W. (2009). Providing group knowledgeawareness in computer-supported collaborative learning: Insights into learningmechanisms. Research and Practice in Technology Enhanced Learning, 4(2),111–132.

Dehler, J., Bodemer, D., Buder, J., & Hesse, F. W. (2011). Guiding knowledgecommunication in CSCL via group knowledge awareness. Computers in HumanBehavior [Special Issue on Group Awareness in CSCL environments], 27(3),1068–1078.

Dillenbourg, P. (1999). What do you mean by collaborative learning? In P.Dillenbourg (Ed.), Collaborative learning: Cognitive and computationalapproaches (pp. 1–19). Oxford: Elsevier.

Dillenbourg, P., Baker, M., Blaye, A., & O’Malley, C. (1996). The evolution of researchon collaborative learning. In E. Spada & P. Reiman (Eds.), Learning in humans andmachine: Towards an interdisciplinary learning science (pp. 189–211). Oxford:Elsevier.

Doise, W., & Mugny, G. (1984). The social development of the intellect. Oxford:Pergamon Press.

Elshout-Mohr, M., & van Hout-Wolters, B. (1995). Actief leren en studeren: Achtscenario’s. Pedagogische Studiën, 72, 273–300.

Engelmann, T., Dehler, J., Bodemer, D., & Buder, J. (2009). Knowledge awarenessin CSCL: A psychological perspective. Computers in Human Behavior, 25,949–960.

Engelmann, T., & Tergan, S.-O. (2007). ‘‘Knowledge and information awareness” forenhancing computer-supported collaborative problem solving by spatiallydistributed group members. In S. Vosniadou, D. Kayser, & A. Protopapas(Eds.), Proceedings of EuroCogSci 07 – The European cognitive science conference(pp. 71–76). Hove, East Sussex, England: Lawrence Erlbaum Associates.

Festinger, L. (1954). A theory of social comparison processes. Human Relations, 7(2),117–140.

Fischer, F., & Mandl, H. (2005). Knowledge convergence in computer-supportedcollaborative learning: The role of external representation tools. Journal of theLearning Sciences, 14(3), 405–441.

Gijlers, A. H. (2006). Confrontation and co-construction. Unpublished doctoral thesis.Twente University: The Netherlands.

Goodnow, J. J. (1988). Parents’ ideas, actions, and feelings: Models and methodsfrom developmental and social psychology. Child Development, 59, 286–320.

Joshua, S., & Dupin, J. J. (1987). Taking into account student conceptions ininstructional strategy: An example in physics. Cognition and Instruction, 4,117–135.

Kenny, D. A., Kashy, D. A., & Cook, W. L. (2006). Dyadic data analysis. New York:Guilford.

King, A. (1991). Faciliating elaborative learning through guided student-generatedquestioning. Educational Psychologist, 27, 111–126.

King, A. (1999). Discourse patterns for mediating peer learning. In A. M. O’Donnell &A. King (Eds.), Cognitive Perspectives on Peer Learning (pp. 87–115). Mahwah, NJ:Erlbaum.

Koriat, A., & Bjork, R. A. (2005). Illusions of competence in monitoring one’sknowledge during study. Journal of Experimental Psychology: Learning, Memoryand Cognition, 31(2), 187–194.

Krauss, R. M., & Fussell, S. R. (1991). Perspective-taking in communication:Representations of others’ knowledge in reference. Social Cognition, 9, 2–24.

Kreijns, K., Kirschner, P. A., & Jochems, W. (2003). Identifying the pitfalls for socialinteraction in computer-supported collaborative learning environments: areview of the research. Computers in Human Behavior, 19, 335–353.

Leinonen, P., & Järvelä, S. (2006). Facilitating interpersonal evaluation of knowledgein a context of distributed team collaboration. British Journal of EducationalTechnology, 37(6), 897–916.

Lockridge, C. B., & Brennan, S. E. (2002). Addressees’ needs influence speakers’ earlysyntactic choices. Psychonomic Bulletin & Review, 9(3), 550–557.

Maki, R. H., & Berry, S. L. (1984). Metacomprehension of text material. Journal ofExperimental Psychology: Learning, Memory, & Cognition, 4, 663–679.

Mäkitalo, K., Weinberger, A., Häkkinen, P., Järvelä, S., & Fischer, F. (2005). Epistemiccooperation scripts in online learning environments: Fostering learning byreducing uncertainty in discourse? Computers in Human Behavior, 21(4), 603–622[SSCI: 1116].

Mercer, N. (1996). The quality of talk in children’s collaborative activity in theclassroom. Learning and Instruction, 6, 359–377.

Molinari, G., Sangin, M., Dillenbourg, P., & Nüssli, M.-A. (2009). Knowledgeinterdependence with the partner, accuracy of mutual knowledge model andcomputer-supported collaborative learning. European Journal of Psychology ofEducation, Special Issue: ‘‘Social Dynamics in Judgment and Performance inAcademic Settings”.

Nastasi, B. K., & Clements, D. H. (1992). Social-cognitive behaviors and higher-orderthinking in educational computer environments. Learning and Instruction, 2,215–238.

Newman, D. R., Webb, B., & Cochrane, C. (1995). A content analysis method to measurecritical thinking in face-to-face and computer supported group learning.

M. Sangin et al. / Computers in Human Behavior 27 (2011) 1059–1067 1067

Available from: http://www.emoderators.com/ipct-j/1995/n2/newman.html[retrieved 5.4.2007].

Nickerson, R. S. (1999). How we know – and sometimes misjudge – what othersknow: Imputing one’s own knowledge to others. Psychological Bulletin, 125,737–759.

Nickerson, R. S., Baddeley, A., & Freeman, B. (1987). Are people’s estimates of whatother people know influenced by what they themselves know? ActaPsychologica, 64, 245–259.

Nückles, M., & Bromme, R. (2002). Internet experts’ planning of explanations forlaypersons: A Web experimental approach in the Internet domain. ExperimentalPsychology, 49(4), 292–304.

Nückles, M., & Stürz, A. (2006). The assessment tool. A method to supportasynchronous communication between computer experts and laypersons.Computers in Human Behavior, 22(2006), 917–940.

Nückles, M., Wittwer, J., & Renkl, A. (2005). Information about a layperson’sknowledge supports experts in giving effective and efficient online advice tolaypersons. Journal of Experimental Psychology: Applied, 11, 219–236.

Ogata, H., Matsuura, K., & Yano, Y. (2000). Active knowledge awareness map:Visualizing learners activities in a web based CSCL environment. Paper presented atthe international workshop on new technologies in collaborative learning,Tokushima, Japan.

Ogata, H., & Yano, Y. (2000). Combining knowledge awareness and informationfiltering in an open-ended collaborative learning environment. InternationalJournal of Artificial Intelligence in Education, 11, 33–46.

Renkl, A. (1997). Learning from worked-out examples: A study on individualdifferences. Cognitive Science, 21, 1–29.

Roschelle, J., & Teasley, S. D. (1995). The construction of shared knowledge incollaborative problem solving. In C. E. O’Malley (Ed.), Computer-supportedcollaborative learning (pp. 69–197). Berlin: Springer-Verlag.

Ross, L., Greene, D., & House, P. (1977). The ‘false consensus’ effect: An egocentricbias in social perception and attribution processes. Journal of Experimental SocialPsychology, 13, 279–301.

Sangin, M. (2009). Peer knowledge modeling in computer supported collaborativelearning. Unpublished doctoral thesis. Switzerland: Ecole PolitechniqueFédérale de Lausanne.

Sangin, M., Nova, N., Molinari, G. Dillenbourg, P. (2007). Partner modeling is mutual.In Proceedings of the seventh international conference of computer supportedcollaborative learning (CSCL), July 16–July 21, New Brunswick, NJ, USA.

Slavin, R. E. (1983). Cooperative Learning. New York: Longman.Sobel, M. E. (1982). Asymptotic confidence intervals for indirect effects in structural

equations models. In S. Leinhart (Ed.), Sociological methodology (pp. 290–312).San Francisco: Jossey-Bass.

Sollers, A., Lesgold, A., Linton, F., & Goodman, B. (1999). What makes peerinteraction effective? Modeling effective communication in an intelligent

CSCL. In Proceedings of the 1999 AAAI fall symposium: Psychological models ofcommunication in collaborative systems (pp. 116–123). Cape Cod, MA.

Stahl, G. (2002). Contributions to a theoretical framework for CSCL. In Proceedings ofthe international conference of computer supported collaborative learning (2002)(pp. 62–71).

Steedman, M. J., & Johnson-Laird, P. N. (1980). The production of sentences,utterances and speech acts: Have computers anything to say? In B. Butterworth(Ed.), Language production: Speech and talk. London: Academic Press.

Strijbos, J. W., Martens, R. L., Prins, F. J., & Jochems, W. M. G. (2006). Contentanalysis: What are they talking about? Computers & Education, 46, 29–48.

Suthers, D. D. (2006). Technology affordances for intersubjective meaning making:A research agenda for CSCL. International Journal of Computer-SupportedCollaborative Learning, 1(3), 315–337.

Teasley, S. (1995). The role of talk in children’s peer collaboration. DevelopmentalPsychology, 3(2), 207–220.

Teasley, S. (1997). Talking about reasoning: How important is the peer in peercollaboration? In L. B. Resnick, R. Säljö, C. Pontecorvo, & B. Burge (Eds.),Discourse, tools and reasoning: Essays on situated cognition (pp. 361–384). Berlin:Springer.

Tudge, J. (1989). When collaboration leads to regression: Some negativeconsequences of socio-cognitive conflict. European Journal of Social Psychology,19, 123–138.

van Boxtel, C., van der Linden, J., & Kanselaar, G. (2000). Collaborative learning tasksand the elaboration of conceptual knowledge. Learning & Instruction, 10,311–330.

Webb, N. M. (1989). Peer interaction and learning in small groups. InternationalJournal of Educational Research, 13, 21–39.

Webb, N. M. (1991). Task-related verbal interaction and mathematics learning insmall groups. Journal of Research in Mathematics Education, 22, 366–389.

Weinberger, A. (2003). Scripts for computer-supported collaborative learning. Effects ofsocial and epistemic cooperation scripts on collaborative knowledge construction.Doctoral dissertation, Ludwig-Maximilians-University, Munich, Germany. Availablefrom: http://edoc.ub.uni-muenchen.de/archive/00001120/01/Weinberger_Armin.pdf.

Weinberger, A., Ertl, B., Fischer, F., & Mandl, H. (2005). Epistemic and socialscripts in computer-supported collaborative learning. Instructional Science,33(1), 1–30.

Wittwer, J., Nückles, M., & Renkl, A. (2008). Is underestimation less detrimentalthan overestimation? The impact of experts’ beliefs about a layperson’sknowledge on learning and question asking. Instructional Science, 36(1),27–52.

Xin, C. (2002). Validity centered design for the domain of engaged collaborativediscourse in computer conferencing. Unpublished doctoral dissertation, BrighamYoung University: Provo, Utah.