Skip to content

Social Communication Research Thrust

The Social-Communicative (SC) thrust investigated communication as a core enabler of robust learning, including detailed study of patterns of verbal interaction.

Social Dialogue and Robust Learning.  A key culminating output of the SC thrust was the book, Socializing Intelligence Through Academic Talk and Dialogue (Resnick, Asterhan, and Clarke, 2015) and we expect it to have a major influence going forward (see top left of Figure 6).  The book consists of 34 chapters written by leaders in the field including Micki Chi, James Greeno, Ken Koedinger, Gerry Stahl, Carolyn Rose among others. Theories and case studies are presented that provide theoretical framing and strong evidence for robust learning effects (long-term transfer and acceleration of future learning) of student participation in well-structured classroom dialogue.   It is written to appeal to researchers in education, learning and cognitive sciences, educational psychology, instructional science, and linguistics, as well as to teachers, curriculum designers, and educational policy makers.

In the early years of LearnLab, work in this area explored the conditions by which tutorial dialogue produces robust learning.  This line of work has been highly influential (e.g., VanLehn et al., 2007, cited by 326) and was foundational in the development of the KLI framework. Subsequent research pursued the creation and evaluation of automated text classification tools that facilitate other researchers and course developers to apply language processing to support student learning.  This work has also been highly influential (e.g., Rose et al., 2008, cited by 242).  A key discovery of that work is that the effectiveness of automated text classification depends not only on effective machine learning algorithms, but also on developing linguistic pattern detectors that provide the feature representation input for machine learning.  This insight is shared with the CMDM thrust result that developing or learning better feature representations improves downstream learning (Li et al., 2015).

Advanced Technology for Language-Based Learning.  SC research produced, and was enabled by, key language technologies including LightSIDE  (Mayfield & Rosé, 2013), formerly TagHelper (Rosé et al., 2008), for language analysis, Bazaar (Adamson & Rosé, 2012), formerly Basilica (Kumar & Rosé, 2011; Kumar et al., 2007), for managing dynamic support for online group learning, and TuTalk  (Rosé et al., 2001; Rosé & VanLehn, 2005; Jordan et al., 2007) for conducting interactive directed lines of reasoning with individual students and groups.  The work is also influenced by and influencing the KLI framework  (Koedinger, Corbett, & Perfetti, 2012, cited by 147) through the effort to provide a fine-grain theoretical explanation for the growth of dialogue and argumentation skills in terms of identifiable instructional and learning events and in terms of specific knowledge components that produce consistent conversational behaviors (Koedinger & Wiese, 2015).

Scaling Social Communication to Improve Learning.  While other thrusts have elaborated on the implicit learning processes of memory and induction that are part of the KLI framework, the SC thrust has advanced understanding of the explicit learning processes of sensemaking.  It has demonstrated how sensemaking can be enhanced through social dialogue and argumentation. We tracked student and teacher growth in productive dialogue (i.e., Accountable Talk moves) over two years (Clarke et al., 2013) by using an automatic coding (via LightSIDE, Mayfield & Rose, 2013) of transcript data from 30 classroom discussions.  A Key discovery was that uptake of Accountable Talk is enhanced, above and beyond teachers professional development, by engaging students in small group online Accountable Talk Activities facilitated by intelligent conversational agents (Adamson & Rosé, 2012; Dyke, Adamson, Howley, & Rose, 2013).

A recent summary of computer-supported collaborative learning results from the SC thrust (Rose & Ferschke, 2016) describes how dynamic support for collaborative learning has been enabled by an integration of text mining and conversational agents to provide novel support for productive discussion processes. This research has paved the way for emerging technologies that support discussion-based learning at scale in Massive Open Online Courses (MOOCs).

Following this path, the SC thrust successfully implemented a MOOC in which technology support for discussion based learning and help exchange was integrated.  Early results suggest a reduction in attrition over time by a factor of two for students experiencing a collaborative chat as part of their MOOC participation (controlling for the effect of level of engagement). Analyses of the discussion data in Twitter, the traditional threaded discussions, and the collaborative chats suggest that a significantly higher concentration of reasoning, especially reasoning about domain content, occurred in the collaborative chats in contrast with Twitter and the threaded discussions (effect size .5 sd).  This success has led to a partnership with edX, a satellite open source effort, which now includes 23 members from outside CMU, including both industrial partners and university partners, to design and build out affordances for discussion based learning under the umbrella of edX’s open source community (