Simulating Collaborative Discourse

Collaborative problem-solving (CPS) has widely been recognised as an essential skill for success in the 21st century. Because of this, many researchers have focused on trying to better understand CPS in efforts to find out when it is effective, when it is not, and how to make it a teachable skill. 

To study CPS, researchers need to collect discourse data that represents interactions among individuals, such as conversations. Importantly, any conclusions drawn from these data are limited either to the specific sample collected or to samples that are sufficiently similar to those collected. Of course, researchers can and should collect more data to test the generalisability of their conclusions. However, in most cases, collecting data of this kind is difficult and expensive. To address this issue, we propose to develop methods for simulating collaborative discourse data. 

Preliminary work by Swiecki and colleagues (in preparation) has shown that it is possible to accurately simulate collaborative discourse data in terms of patterns of interaction and patterns of discourse codes (labels). Moreover, decades of research in the social sciences has shown that it is possible to simulate complex social interactions using multi-agent models. Despite these advances, simulation methods have not been adopted to study CPS within the fields of the Learning Sciences (LS), Learning Analytics (LA), and Computer-Supported Collaborative Learning (CSCL)

To make simulation based methods available and accessible to these communities, we will begin by extending Swiecki’s prior work to cover less constrained collaborative scenarios. Next, we will develop programming packages for popular data science languages such as R and Python specific to simulating collaborative discourse data. In addition to providing researchers with functions and algorithms to simulate collaborative discourse under a large variety of conditions, these packages will contain sample simulated datasets that can be used to immediately begin addressing relevant questions. 

In time, we will integrate the packages and data into open-source, web-based tools that will allow researchers to simulate and analyse collaborative discourse data without extensive programming knowledge. These tools will be based upon similar successful tools such as NetLogo and Hash, but will cater to the specific needs of researchers interested in understanding collaborative discourse.