PSLC Social Communicative Thrust
www.learnlab.org/
Summary
During PSLC’s first four years, its Interactive Communication Cluster has studied interactions between a student and a tutor (either human or computer) or, less frequently, two students interacting with each other. Most of the experimental manipulations and subsequent analyses have focused on the cognitive content of interaction through learning space analyses, in other words, the what and when of instruction. Study results investigating the effect of interaction, although somewhat mixed, have largely supported the hypothesis that focused interactionpromotes cognitive aspects of learning such as attention to the most important knowledge components in a domain, deeper cognitive processing, and increased engagement with thecontent. VanLehn and colleagues (2007) present a thorough review of this literature as well asresults from recent investigations. These results encouraged early PSLC effortsto “unpack” the nature of communicative interaction in instruction and learning. Rummel and colleagues (Diziol,Rummel, Kahrimanis, et al., 2008a, 2008b), for example have recently evaluated interactions with a rating scheme analysis that quantifies the quality of an interaction on a number ofdimensions. This work represents an important step towards the type of up close inspection of communication that many scholars believe is necessary if we are to understand, and be able tomanipulate for instructional purposes, the way that communication works to produce robust learning.
In our re-named Social-Communicative Factors thrust, we propose now to expand our investigations of communication as a core enabler of robust learning to include detailed study ofpatterns of interaction, the role of social variables in initiating and sustaining learning, and theeffects on motivation, self-attribution and commitment to a learning group that are associated with learning through social-communicative interaction. Specifically, we propose to investigatehow human linguistic interaction works in instruction and learning, and how participants in learning exchanges (both teachers and students) can best be taught productive forms ofinteraction. We draw from our extensive prior work related separately to classroom discourse(Chapin & O’Connor, 2004; Bill et al., 1992; Resnick et al., 1992) and collaborative learning(Gweon et al., 2007; Joshi & Rosé, 2007; Rummel & Diziol, 2008). We note that, although the classroom discourse and collaborative learning communities have proceeded mainly independently from one another, the conversational processes identified as valuable within these two communities are strongly overlapping.
Investigations of valuable conversational contributions have been conducted both within communities exploring the cognitive foundations of group learning as well as the sociocultural community. Regardless of the theoretical framework, the same ideas have surfaced under anumber of different names including accountable talk (Michaels, O’Connor & Resnick, 2007), transactivity (Berkowitz & Gibbs, 1984; Teasley, 1997; Weinberger & Fishcer, 2006; King,1999), productive agency (Schwartz, 1999), and uptake (Suthers, 2006), and have been demonstrated to predict learning both in collaborative learning contexts (Azimita & Montgomery, 1993; Joshi & Rosé, 2007) and classroom contexts (O’Connor et al., 2007). For example, one cognitive justification for the value of transactive conversational behavior is its connection with cognitive conflict (Piaget, 1985), where transactive conversational moves highlight differences between the mental models of collaborating students. One can argue that a major cognitive benefit of collaborative learning is that when students bring differing perspectives to a problem-solving situation, the interaction causes the participants to considerquestions that might not have occurred to them otherwise. This stimulus could cause them to identify gaps in their understanding, which they would then be in a position to address. This type of cognitive conflict has the potential to lead to productive shifts in student understanding.It has the potential to elicit elaborate explanations from students that are associated with learning(Webb, Nemer, & Zuniga 2002). From the sociocultural perspective, based on Vygotsky’sseminal work (Vygotsky 1978), we can similarly argue that when students who have different strengths and weaknesses work together, they can provide support for each other that allows them to solve problems that would be just beyond their reach if they were working alone.
As an example of the connection between these separate characterizations of similar conversational behavior, the Accountable Talk principle of Challenges and Counter Examples, which instructs teachers to insert prompts such as “Is that always true?," is designed to elicit transactive conversational contributions such as Counter Consideration moves, which point out a circumstance under which a principle put forward by one student might not hold. Another Accountable Talk principle, Revoicing, teaches participants to listen carefully to the content ofeach other’s contributions by paraphrasing them and seeking confirmation that the paraphrasecaptures the original speaker’s intended meaning. Paraphrasing is another type of transactiveconversational move.
In order to work toward a common conceptual framework that unifies the classroom discourse, collaborative learning and instructional tutoring communities, we plan to develop aconcrete and precise formalization on a linguistic level of what counts as performing these valued conversational moves. This concrete formalization will provide a common language fordocumenting and investigating the specific ways in which social-communicative practices canpromote (or hinder) learning of complex mathematics and science content and reasoning skills.Our work will explicitly examine causal connections between these communicative processesand learning by running in vivo experiments in which specific social-communicative practices are introduced into well-defined mathematics and science units of study. These studies will make use of techniques from automatic collaborative learning process analysis (Rosé et al., inpress; Wang et al., 2007; Donmez et al., 2005) and script-based support for productivecollaboration (Dillenbourg & Jermann, 2007; Kollar, Fischer, & Hesse, 2006; Rummel & Spada,2007; Diziol, Rummel, Kahrimanis, Spada & Avaris, 2008; Diziol et al., 2008; Walker, Rummel,McLaren & Koedinger, 2007) to carefully manipulate these properties of conversation in highlycontrolled and context sensitive ways.
We build on the evidence of robust learning effects (including transfer, long-term retention and facilitation of subsequent learning) resulting from sustained participation in classrooms characterized by high levels of content-specific interactive communication. There is a considerable corpus of studies in the literature, where similar kinds of discourse-intensive instruction in difficult “traditional” high-demand subject matter have produced unexpected results in American classrooms. (c.f., among others, Lee, 2001, in literature; Ball & Lampert, 1998, Boaler, 2003, Boaler & Greeno, 2000, Chapin, et al., 2003, and Empson et al., 2006, in mathematics; Minstrell, 1989, in physics; Rosebery et al., 2005, Warren & Rosebery, 1996, in elementary science c.f., among others, Goldenberg, 1992/3; Beck et al., 1996; Chapin, O’Connor & Anderson, 2003; Lampert & Ball, 1998; Resnick et al., 1992. Scholars in several European countries have reported similar results (e.g., van Dooren, et al., 2005; Fischbein, Jehiam & Cohen, 1995; Merenluoto & Lehtinen, 2002; Tsamir, 2003; Van Vergnaud, 1989; Vosniadou, Baltas & Vamvakoussi, in press).
Questions and Hypotheses
Our research foci fall into five categories, one devoted to building tools for (semi)-automated analysis of social communicative interaction and a fourth devoted to specific questions abouthow social-communicative interaction works in instructional and learning.
1. Tools for Analysis of Extended and Complex Human Discourse
Progress in the study of transactive language interaction as a facilitator of learning has beenlimited, up to now, by the technical difficulties of collecting, transcribing, coding and analyzinglarge amounts of linguistic data. We propose to collaborate with PSLC’s DataShop in order tobuild a bank of interactive learning protocols that can be used by PSLC and other researchers tostudy interactive communication in learning and to create at least partially automated coding anddata mining processes for analyzing interactive data.
2. What Patterns of Discourse are Causally Associated with Robust Learning Outcomes?
Using our bank of transcripts and analytic tools we plan to examine which specific linguistic“moves” and patterns of interaction are causally related to improved learning for students ofdifferent degrees of preparation and different social backgrounds. Our research will focus on thestudy of human language interactions that support learning, although an eventual outcome maybe specifying interactive features that can be built into computer-based tutors.
We will begin with a set of hypotheses developed by Michaels et al. (2007) specifyingproductive Accountable Talk moves in classrooms. This work identifies specific language “moves” that characterize Accountable Talk in classrooms.
3. What Specific Cognitive and Metacognitive Processes are Enhanced by Particular
Strategies of Social-communicative Interaction? Cognitive Process Elaboration. Much research has demonstrated the potential effectiveness ofcollaborative settings for improving students’ problem solving and learning (e.g., Hooper, 1992;Slavin, 1996; Teasley, 1995). A couple of different mechanisms are held accountable for thebenefits of collaborative problem solving and learning (Hausmann, Chi, & Roy, 2004; Meier,2005; Rummel & Spada, 2005; Meier, Spada, & Rummel, 2007). These mechanisms can be seenas learning opportunities unique to a collaborative setting; however, only when students takeadvantage of them will their learning be improved.
One mechanism that has shown to be particularly important for students’ learning in acollaborative learning setting is giving explanations to the learning partner (Hausmann, et al.,2004; Ploetzner, Dillenbourg, Preier, & Traum, 1999; Slavin, 1996; Spada et al., 2005; Teasley,1995; Webb, 1989; Webb & Mastergeorge, 2003; Webb, Troper, & Fall, 1995). Duringcollaboration, students are required to verbalize their knowledge. Furthermore, when theirpartner has difficulties to understand their explanations they have to reformulate and clarify theirstatements. As O’Donnell (1999) pointed out, this verbalization of knowledge demandselaborating on the learning content and yields deeper processing. The particular benefits ofverbalization in a collaboration were also found in a study conducted by Teasley (1995).
Second, receiving explanations from a partner after errors or help requests can be beneficialfor students’ learning (e.g. Webb 1989; Webb & Mastergeorge, 2003; Webb et al., 1995). Asking for help and receiving explanations, however, only improves the learning outcome under specific conditions: It is important that the explanations have a high level of elaboration (e.g., Spada et al., 2005; Webb, 1989), and, it is beneficial if the help-seeking stuydent makes explicit where he needs help and persists in asking for further explanations when the initial answer has not been sufficient (Webb & Mastergeorge, 2003).
Third, a collaborative learning setting enables students to engage in co-construction or jointproduction of knowledge (Barron, 2000; Berg, 1993; Hausmann et al., 2004; Li, 2003;Moschkovich, 1996; Spada et al., 2005). In a study on collaborative learning in mathematicsclassrooms, Berg (1993, 1994) observed students that mutually contributed to the problemsolving process by finishing the sentences of their learning partner or taking turns in vocalizingthe solution steps. Second, Moschkovich (1996) describes how students mutually negotiated themeaning of mathematical concepts and engaged in elaborative discussion on complex questions,thereby constructing a common understanding of the mathematical background (Moschkovich,1996). As research has shown, the co-construction of knowledge is a good predictor for studentachievement in the domain of mathematics (Berg, 1993, 1994) and can result in a deeperunderstanding of mathematical concepts in student dyads (Moschkovich, 1996).Metacognitive processes: monitoring the problem-solving process. King (1991) promptedcollaborating students to elaborate on their problem solving process. In her study, students’interaction was guided by three sets of metacognitive questions. One set of questions aimed atstudents identifying the problem and planning how to solve it; another set guided them tomonitor their collaboration while solving the problem. Finally, when they had solved theproblem, they were instructed to evaluate the problem solving process by the third question set.They had to answer questions as “What worked? What did not work? What would we dodifferently next time?” (King, 1991, p. 309). Students were trained to use these questions twice aweek over a 3-week period. Interestingly, the instructions still influenced students’ interactionand problem solving even when students no longer received instructional support. The group thatwas guided in questioning outperformed the two control groups both in giving elaboratedexplanations and problem solving.
In Kramarski’s classroom study (2004), groups of four 8th-graders worked on a linear graphunit. In the control condition, students were instructed to engage in mathematical discourse. Inthe experimental condition, the mathematical discourse was guided by the IMPROVE method,instructing students to apply a series of cognitive and metacognitive questions. The use of thequestioning technique yielded an increased amount of elaborated explanations and an improvedmathematical discourse. Furthermore, students of the experimental condition outperformed theunsupported control group both on graph interpretation and graph construction in a post test.
4. What Motivational Processes are Enhanced by Different Forms of Linguistic Discourse?
Research in the social psychology of learning indicates that sociocognitive engagement,including exploration of personal disagreements between students seeking to reach a jointsolution to a problem can stimulate thinking and learning. For example, Doise and Mugny (1984;see also Perret-Clermont, 1980) demonstrated that sociocognitive conflict, defined as disagreement between children seeking to reach a joint solution to a problem, can stimulateintellectual growth under certain conditions. Substantial research has also examined therelationships between emotion and cognition (Damasio, 1994; Forgas, 2001; Schwarz & Clore,2007). This work has demonstrated that emotion can influence memory retrieval (Bower, 1981),judgment (Schwarz & Clore, 1988), and decision-making (Seo & Barret, 2007). Although muchis known about the emotion-cognition interface, its relevance for learning has not been explored.
5. Do Different Forms of Linguistic Discourse Differentially Affect Groups of Students?
A number of researchers have focused especially on the importance of linguistic interactionin content areas for minority students and other academically underprepared students—both as aresource and as an obstacle in academic achievement generally (Adger et al., 2002; August &Hakuta, 1998; Ballenger, 1999; Baugh, 1999; Lee, 2001; Cazden, 2001; Delpit & Dowdy, 2002;Heath, 1983; Abedi & Gandara, 2006; Moll et al., 1992;Walqui & Koelsch, 2006) and morespecifically in mathematics learning (Cocking & Mestre, 1988; Moschkovich, 2000; Moses &Cobb, 2001; Resnick, Bill etc.). Similarly, in the collaborative learning community, exchange ofhelping behavior has been demonstrated to decrease tensions in inter-racial classrooms, as wellas increase important qualities such as liking for other students, identification with a learningcommunity, motivation, and retention (Sharan, 1980; Slavin, 1980; etc.) as well as impartingfeelings of agency and empowerment among “low status” students (Elbers & Hann, 2004).
Planned Research
We will proceed with two interacting research strategies: one, Expanding capacities forrecording, coding and analyzing interactive communication that can be at least partiallyautomated; and two, conducting in vivo experiments on ways of teaching participants the most promising patterns of interactive communication and testing the effects of these patterns onmeasures of robust learning.
Expanding Capacities for Recording, Coding, and Analyzing Interactive Communication
We plan to start the process by putting hand-coded data (of existing transcripts) into DataShop tobegin work early on developing systematic descriptions of patterns of interactions. These willinclude transcripts of mathematics lessons led by Victoria Bill (Bill et al, 1991; Resnick et al,1992) that raised African American participants’ average standardized test scores fromapproximately the 30th percentile to above the 80th percentile. They will also include transcriptsfrom Chapin and O’Connor’s interventions in Chelsea, Massachusetts (Chapin & O’Connor,2004). This will supplement already existing collaborative learning data sets (Walker, Rummel,& Koedinger, 2007; Kumar et al., 2007a). We will collect further transcripts (from newclassrooms and in vivo experiments) and build automated coding systems of their interactivecharacteristics.
We will work on a linguistic formalization of the types of discourse moves that have been described on an abstract level in the collaborative learning literature and the classroom discourse literature. Our team brings to the table methodologies and technologies for automatic analysis as well as insights from syntax, formal semantics, formal pragmatics, sociolinguistics, and philosophy of language (Rosé et al., 1995; Rosé, 2000; Rosé & VanLehn, 2005; Arguello & Rosé, 2006; Joshi & Rosé, 2007; Wang et al., 2007; Wang et al., 2008).
Our team brings much infrastructure from our previous work for automatic conversationanalysis (Rosé et al., in press; Wang et al., 2007; Joshi & Rosé, 2007), report generation forgroup learning facilitators (Kang et al., to appear-a; Kang et al., to appear-b), and dynamically triggering interactive support for reflection and collaboration in the midst of ongoingconversations based on that automatic analysis (Kumar et al., 2006; Kumar et al., 2007-a; Kumaret al., 2007-b; Chaudhuri et al., to appear; Kumar et al., submitted-a; Kumar et al., submitted-b).Thus, we are well positioned to automate the analyses of valuable conversational processes aswell as to use those automatic analyses to trigger interventions that elicit these types of behavioras in our previous studies where automatically or manually triggered interventions increased the
quality of interaction (Gweon et al., 2006; Wang et al., 2007-b) and learning (Gweon et al., 2006;Kumar et al., 2007).
Experiments on Teaching: Effects of Promising Patterns of Interactive Communication We will begin by replicating and extending a series of in vivo experiments on the effects of Accountable Talk in low-income urban classrooms with high proportions of English languagelearners in Chelsea, Massachusetts (O’Connor et al 2007; NHSF REC 0231893, PI: O’Connor).In a tightly controlled series of three-day studies in 5th and 6th grade classrooms, O’Connor’sgroup sought to determine whether it was possible to get evidence supporting a hypothesizedcausal relationship between selected discourse-intensive instructional practices and studentmathematics learning. In previous non-experimental studies in Chelsea, students had shownlarge gains on standardized tests after a year or more of discourse-intensive instruction, but itwas not possible to test the specific features of the intervention that produced these effects. Thus it was possible that cognitive and metacognitive abilities might improve over months of practicein clarifying, justifying and describing mathematical ideas, whether or not explicit transactivecommunication strategies were employed. Similarly, student motivation might have improveddue to long-term participation in an intensive mathematics program, without a specific impact ofparticular forms of linguistic participation.The short-term, highly controlled in vivo experiments were designed to test more specifichypotheses concerning specific Accountable Talk moves. Two versions of 3-day lessons on newmaterial were designed, differing only in the use of the core Accountable Talk moves. Students in the two groups were otherwise exposed to the same examples, problems, assignments andactivities. Unexpected mathematical content emerging in discussions in the experimental classeswas quickly incorporated into the direct instruction in the control classes. An ANCOVA showed a significant (p<.008) main effect of instruction type and a substantial effect size (Cohen’s d:0.833) at the end of the 3-day lesson. [c37]Post-test item analysis showed that the two conditionsdid not differ in performance on computation-only items, but differed significantly on itemsrequiring the ability to explain, compare, justify solutions and the like.We will begin by replicating these in vivo studies in middle and high school mathematicsclasses and in college level science and math courses. The secondary school studies will becarried out in Pittsburgh Public School classrooms, initially in the new University PartnershipSchool, which represents acollaboration between the public schools, the University of PittsburghSchool of Education, and LRDC’s Institute for Learning. College level experiments will becarried out in teacher training courses at the University of Pittsburgh’s School of Education. To support the expansion of our in vivo experiments, and to explore ways of spreadingsuccessful practices to multiple classrooms, we plan to extend the collaborative scripts approachof Rummel and colleagues and other researchers in the area of script-based collaboration (for a recent overview of relevant work see Fischer, Kollar, Mandl, & Haake, 2007). Collaborative scripts appear to be a powerful way to teach participants (both students and teachers) ininteractive learning groups various subsets of the language moves and study the effects onseveral measures of robust learning. Particularly when implemented in computer-supportedlearning setting they provide a convenient way to manipulate specific types of conversational moves. Initial studies will be highly controlled interventions, starting with tests of the Accountable Talk moves described above, and expanding as new patterns of interactive collaboration are identified in DataShop transcripts and analyses. Using automatic collaborative learning process analysis technology, later investigations may evaluate context-sensitive versions of these scripts where the script may dynamically adjust based on specific properties of the interactionbetween students. We will collect measures of cognitive enhancement (e.g., learning taughtconcepts, domain-specific and general reasoning skills, academic language), motivational enhancement (e.g., students’ estimate of the importance of learning the domain, of their ability tosucceed in learning the domain) and metacognitive enhancement (e.g., students’ estimates oftheir own knowledge, students’ ability to manage their reasoning and learning) in order toexplore there connection with these types of conversational moves.
Subsequent studies will test a larger intervention that includes training in the most effective conversational moves and collaborative scripts with implementation in a number of classrooms.The studies will focus on math and science learning topics. Initial studies will be carried out in the Pittsburgh Public Schools (beginning in the University Partnership School that will beopened in 2008-09 in collaboration with the University of Pittsburgh and extending to additionalsites); in classrooms in the Boston and Worcester, Massachusetts area with easy access from BUand Clark); in Germany; and also in university science courses. The scale-up interventionstudies will be conducted in school districts (in Texas, California, Michigan, Maryland and otherstates) that are part of the Institute for Learning partnership. We will also explore possibilitiesfor building scripts for collaborative interaction into tutors.
People
