Clarissa: A Laboratory for the Modelling of Collaboration

There is a continuing debate about how to organise collaborative activities for them to be educationally valuable. This organisation can be analysed both in terms of how to set up the situation and how to arrange the interactions between participants. Here, we are interested in the different ways in which to organise the interaction, and in the consequences for effective collaboration. For example, what constraints might be usefully applied to 'unconstrained' collaboration? The work described here uses computational modelling to provide the tools for a systematic investigation. The approach taken is based on the assumption that students should learn through the adoption of different ways of using dialogue (dialogue roles). Very few, however, have sought to examine this assumption through the computational modelling of explicit dialogue role assignment within collaborative situations. The Clarissa (Collaborative Learning As Realised In Simple Simulated Agents) system allows the exploration of collaboration with respect to dialogue roles and different policies used for their allocation. A simple problem solving domain context is used, which exhibits many of the properties of more complex situations. The system is described, and selected results obtained from modelling a range of types of collaboration are presented. Findings from the analysis of a set of different collaborative arrangements indicate that there are more effective ways of organising collaborative situations than the free adoption of dialogue roles. In this paper, a pair of such policies are used to explore this issue using a baseline policy chosen as a representative of a commonly accepted form of collaboration. Clarissa itself provides a novel laboratory which has some implications for a new range of software agents capable of plausible collaborative behaviour.