|Fig. 1. The DynaLearn Learning Space 2 (in Teachable Agent mode) including a learner made diagrammatic expression (LHS), and two interacting virtual characters (RHS).|
DynaLearn develops an Intelligent Learning Environment that allows learners to acquire conceptual knowledge by constructing and simulating qualitative models of how systems behave. DynaLearn uses diagrammatic representations for learners to express their ideas. The environment is equipped with semantic technology components capable of generating knowledge-based feedback, and virtual characters enhancing the interaction with learners (Figure 1).
Background and motivation
The importance of learners constructing their conceptual interpretation of system behaviour while learning Physics, Biology or Environmental Science has been highlighted in many studies. From these studies it is apparent that there is a need for software that supports learners in actively working and interacting with the theoretical concepts involved. This can be achieved by having learners create models and run simulations from which they can make predictions and derive explanations.
|Fig. 2. The DynaLearn software consists of three main components. A workspace for creating knowledge, a semantic based feedback generator, and a set of virtual characters facilitating reflective interaction.|
DynaLearn is motivated by this need, particularly for secondary and higher education (undergraduate level). The project integrates well-established, yet independent technological developments, and utilizes the added value that emerges from this integration (Figure 2). Using the DynaLearn workbench, learners construct conceptual models by manipulating icons and using diagrammatic representations. Simulating these models stimulates reflective thought on behalf of the learner, because it confronts learners with the logical consequences of their ideas and insights as expressed in the models. Moreover, within this workbench embodied conversational agents are situated and available for learners to further analyse and improve their models. This interaction is steered using knowledge technology that generates feedback based on a growing repository of models.
Learning by modelling
There is ample research that points out the importance of learners constructing conceptual interpretations of systems’ behaviour. But what kind of tool is needed to accommodate the true nature of conceptual knowledge? Addressing this question, DynaLearn has developed a set of six distinct representations, which act as scaffolds to support learners in developing their conceptual knowledge [Ref]. The representations are referred to as Learning Spaces, and are based on Qualitative Reasoning (QR) technology [Ref]. A qualitative model provides formal means to externalize thought. It captures the explanation that the creator of the models believes to be true of how and why a system behaves. The approach is domain independent.
In DynaLearn we utilize the full expressiveness and potential of the QR formalism. This allows us to divide the qualitative system dynamics phenomena over a range of learning spaces with increasing complexity, implementing a progression where at ‘the next level’ learners are confronted with additional and alternative expressive power for representing and reasoning about the behaviour of systems, and hence facilitating the construction of a better understanding of the phenomena involved.
Generating feedback using semantic technology
|Fig. 3. Key tasks and interactions with the semantic based feedback component.|
An innovative feature of DynaLearn is that conceptual models created by learners can be compared to models created by other learners or teachers, and based on that automated feedback can be provided. The feedback based on the semantic technology has three parts (Figure 3). First, OWL Export: conceptual knowledge contained in models is extracted and expressed in OWL and available for processing using semantic technology. Second, Grounding: the terms from the conceptual models are linked to external vocabularies. These grounded models are stored in a semantic repository. Third, Semantic feedback: alignment and reasoning techniques are applied to discover similarities and dissimilarities among models. Based on that feedback is generated and communicated to the learner. The result of this process is a networked pool of online aligned conceptual models (expressed as ontologies) anchored in common vocabularies, and representing specific scientific concepts. This has the potential of being a valuable Web resource for scientific progress in general and for semantic guided learning in particular.
Aligning model and simulation expectations
Constructing models involves transforming initially vague and general ideas into clearer and more formally specified representations, until the model (and its simulation results) match the expectations of the modeller. This is a reciprocal process in which on the one hand learners stepwise improve the details expressed in their models, while on the other hand they may change their expectations due to the discoveries done when confronted with simulation results of their earlier thoughts.
The diagnostic component helps learners to ultimately arrive at a model that matches their expectations. This is achieved by having learners formulate their expected simulation results. Based on the discrepancies between these expectations and the logical consequences of knowledge already expressed, the software is be able to identify flaws in the latter, and suggest improvements on how to overcome these. The software thus stepwise supports learners in constructing and discovering a correct explanation of how a system works.
Communicative interaction using virtual characters
A distinctive feature of the DynaLearn approach concerns a community of unique virtual characters that accompanies learners in an entertaining and motivating way while handling complex conceptual knowledge (Figure 4). The characters offer basic help as well as sophisticated feedback on the contents modelled by learners.
|Fig. 4. The DynaLearn virtual characters (from left to right): two students, the critic, the teacher, the quizmaster, and the mechanic).
Click here to see them in action.
A cartoonish design combines expressivity with entertainment to stimulate the learner’s work. The characters move freely on the computer desktop and have the ability to speak with individual voices. Being hamsters, the characters reference real-life adoptable pets and resonate DynaLearn’s focus on environmental science. The virtual characters resemble a classroom setting and take roles such as ‘a teacher who delivers knowledge and training’ or ‘students that need to acquire knowledge’. The quizmaster provides a special option. It can be used to assess the learner’s knowledge on a topic.
Environmental science – Application and evaluation
Based on the DynaLearn approach, partners have created didactic materials for teaching and learning environmental science. Particularly, over 200 models have been created covering 61 topics across the seven DynaLearn themes in environmental science (Figure 5 and 6). Using this material, 49 evaluation studies have been conducted in real learning situations and settings (such as, school, college, and university classes), involving 736 participants, varying in evaluation goals, sample size, research design, and duration. Overall the studies showed the great potential of the DynaLearn approach for supporting: the growth of causal system thinking; the acquisition of scientific reasoning skills; the ability to learn about complex ecosystems; the gradual construction of content knowledge; the incremental development of the conceptual modelling approach and skills.
The DynaLearn project has developed an intelligent learning environment that supports active learning, and engages learners in acquiring conceptual knowledge of how systems behave. The learning environment encompasses six learning spaces, which facilitate a gradual progress towards mastering the subject matter. While developing their knowledge learners may obtain formative feedback by grounding their model ingredients in common external vocabulary, and by asking for a comparative analysis with repository models. When the simulation results turn out different from what learners expect, the diagnostic component can be called to suggest improvements to overcome these. The virtual characters are available to engage learners, and induce motivation using specific settings including the Teachable agent and Quiz modes. Evaluation studies confirm the potential of the DynaLearn approach.