7. Conceptual Approach
7.1. Relevant Conceptual Components and Models in Research
7.1.4. The Learning Environment
The IPA is employed in the virtual world in a way that pedagogical services are needed in favor of the learner. In research, the design of the virtual environment is done towards different purposes. The target is to focus on services that lead to better learning support. Elements relevant to the pedagogical
aspects of the environment with potential integration of a pedagogical agent are visited in this Section.
PEGASE in Virtual Reality
Buche and Querrec (2011) propose to extend the intelligent tutoring system in VLE to virtual reality with the PEdagocial Generic and Adaptable SystEm (PEGASE). The work focuses on the importance of representation of knowledge of the environment and its use with pedagogical agents to provide instructive assistance in the virtual environment. PEGASE consists of five models, namely:
domain model, learner model, pedagogical model, interface model, and error model. The domain model provides semantics needed by the artificial agent “to be able to construct a representation of the environment and to act together to reach their goals” (Buche and Querrec, 2011). The domain model expresses three types of knowledge: domain concepts, interfacing with the environment, and entities behavior. The learner model defines learner personal characteristics and gives the temporal condition of his/her knowledge. The pedagogical model provides knowledge for taking teaching decisions with the task to be performed in the environment with knowledge about the pedagogical situation. The interface model is concerned with the learner actions in the environment. The error model is used to detect and identify errors of the learner. With the error model, distinction is made between observational errors and their causes. Thus it gives rise to the importance of the error cause for the instructional purpose. The system compares learner actions to the expected ones from domain model and uses the error model to identify errors. The steps from learning interaction to providing pedagogical assistance involving the five models are depicted in Figure 26. The actions of those steps are conducted by different role agents as autonomous entities that infer and share from the models. For example, the InterfaceAgent observes learner actions while the ErrorsAgent detects and identify errors. The PedagogicalAgent offers pedagogical assistance and the TeacherAgent selects pedagogical assistance to be displayed by the InterfaceAgent.
Figure 26: Five models in an instructive process of PEGASE (Buche & Querrec, 2011).
HERA
The Helpful agent for safEty leaRning in virtuAl environment (HERA) is proposed by Amokrane, Lourdeaux, and Burkhardt (2008) to provide learning and training in an ITS based VLE by means of learner tracking. HERA is based on five conceptual model elements: domain model, learner model, errors model, risks model, and pedagogical model (see Figure 27). In this approach, the domain model provides details about the learning activities. The learner model keeps track of learner activities. The errors model contains a classification of error types that might occur during learning with causes and consequences of actions committed in the VLE. The risks model describes the risks produced by errors. The risks model also contains measures used to prevent or limit the effects of such risks. The pedagogical model provides the training rules of when, why to intervene, and to explain the learner errors. HERA modules are: interface module, recognition module, learner module, pedagogical module, and risks module. The interface module is an observatory entity that is concerned with observing learner actions and communicating them, in addition to retrieving learner location and the objects in the learner field of vision. The recognition module determines what the learner is doing with its ability to infer learner task plan by using the intelligent agent tactic of plan recognition function. A task plan is defined as “the set of actions carried out by a learner to reach a goal that represents the desired state of the world”. The learner module manages learner task plans. The risks module determines the risks produced as of
learner errors. It contains the measures used to eliminate, prevent, protect, or limit the effects of risks. An example risk in the 3D virtual environment is not wearing a face mask causing the possibility of inhaling toxic substances. In this case, the risk module is triggered by an error with approval from the pedagogical module, and sends a message through the interface module with the consequences of the error. The pedagogical module, in the context of this research, intervenes to help the learner and provide explanations to errors, or reminds the learner about tasks and in communication with the other modules.
Figure 27: HERA (Amokrane, Lourdeaux, & Burkhardt, 2008).
Model after Sklar and Richards
Sklar and Richards (2006) describe a model in relevance to learning with pedagogical agents. The model is composed of six components: domain knowledge, teaching component, user interface, student model, system adaptivity model, and control component. The teaching component is an instructional model that guides the student through the domain knowledge. The system adaptivity model is how the system adapts to student behavior. The control component manages the pieces together.
This model is depicted in Figure 28. This work refers to pedagogical agent to interact with the several components of the model and provide a typology of artificial tutors as pedagogical agents, peer learning agents, or demonstrating agents. Sklar and Richards (2006) stress the importance of the multi-agent model as a control component and discusses its relevant research.
Figure 28: Interactive learning system (Sklar & Richards, 2006).
MASVERP
In the context of safety training in virtual reality for risky situations, Edward, Lourdeaux, Barthès, Lenne, and Burkhardt (2008) propose the Multi-Agent System for Virtual Environment for Risk Prevention (MASVERP) focusing on human decision processes and human-behavior based errors. Three models are used with MASVERP agents: world model, risk model, and activity model (see Figure 29). The world model represents the environment objects, their state, and position while the COLOMBO module is in charge of managing changes in the environment state or an object state. The risk model is mainly composed of decision rules. The activity model helps the agent in the task of planning. The MASVERP agent is based on the BDI model (see Section 5.3.2). MASVERP is equipped with a planner, using an agent planning algorithm for selecting actions based on the agent goals, the environment, and to the individual characteristics of the agents.
Figure 29: Overall architecture and agent of MASVERP (Edward, Lourdeaux, Barthès, Lenne, & Burkhardt, 2008).
MASCARET
MASCARET stands for Multi Agent System for Collaborative, Adaptive and Realistic Environments for Training (Buche, Querrec, Loor, & Chevaillier, 2003).
MASCRET builds on the previously discussed PEGASE component as a multi-agent tutoring system in Figure 26. Chevaillier et al. (2011) further build on MASCARET by utilizing a semantic modeling approach to provide basic knowledge about the environment and the system. It covers the different aspects: 1) structure and behavior of the world; 2) interactions and tasks that users and agents perform in the environment; 3) knowledge items for the use by agents. The conceptual overview of the main components of the semantic MASCARET is depicted in Figure 30. In describing the framework, MATS is a multi-agent based tutoring system for tutoring that refers back to PEGASE described above.
VEHA is a meta-model of virtual environment entities. HAVE is a meta-model that describes interactions and activities of users and artificial agents. BEHAVE is a description of the activities that agents can support. Further investigation of this work yields details of the influence of objects or artifacts as well as the environment on the agent behavior. However, no enough details about the pedagogical model are provided in relation to the semantic enhancement.
.
Figure 30: Conceptual overview of the MASCARET framework (Chevaillier et al., 2011).
Framework for Virtual Embodied Collaboration
The framework proposed by Schmeil and Eppler (2009) can be considered a blueprint for collaborative learning in virtual worlds (see Figure 31). While this model does not consider the inclusion of artificial agents, it is helpful to identify learning collaboration elements that the pedagogical agent should deal with in an immersive virtual learning environment. The model has evolved from the authors’ effort to formalize the elements in the visual collaboration of a virtual world as well as identifying and incorporating collaboration patterns in the environment. This work identifies three dimensions, namely: syntactic dimension, semantic dimension, and pragmatic dimension. The syntactic dimension represents visible elements of collaboration. The semantic dimension is the alignment with desired objectives. The pragmatic dimension represents intentions supported by other layers.
The model provides a typology of each of the level elements such as typology of objects such as static, automated, or interactive as well as actions such as communicative, navigation, and object-related actions. The pragmatic dimension identifies three categories: collaborate, learn, and play. The learn category is depicted according to goals of Bloom’s Taxonomy. While this model provides a blueprint towards formalizing collaboration in virtual worlds, it does not provide indications of the added value of the used collaboration pattern as reported by the authors (Schmeil & Eppler, 2009).
Figure 31: Framework for virtual embodied collaboration by Schmeil and Eppler (2009).
IMA for Complex Learning Resources
Providing assessment is central to the learning process. And therefore, adopting an e-assessment method in the virtual learning environments is also central. A relevant work in e-assessment is found in Wesiak, Al-Smadi, and Guetl (2012) who provide the Integrated Model for e-Assessment (IMA), (see Figure 32) in both the domains of knowledge and skills assessment. The model addresses enriched learning experiences with based on Complex Learning Resources1 (CLR) and integrated assessment methods. The model includes a core methodology of four components: 1) Learning objectives, 2) Complex learning resources, 3) New forms of assessment, and 4) Evaluation and validation. Educational, psychological, technical aspects as well as standards and specifications are included in this model.
Adaptivity components include a didactic model, knowledge model, and learner model.
The IMA model also includes a quality assurance component that aims at ensuring high quality standard in learning activities including best practices and
1 Collaborative and social learning, storytelling, simulation, and serious games (Wesiak, Al-Smadi, & Guetl, 2012).
pragmatic semantic syntactic
standards in delivery and managing ethical aspects such as data protection and plagiarism prevention.
Figure 32: Integrated Model for e-Assessment (IMA) by Wesiak, Al-Smadi, and Guetl (2012).
3D Adaptive Presentation and Navigation
Adapting the visual presentation in an immersive VLE brings positive learning experiences and should not be neglected. Generally changes in the scenes of a virtual word interface occur as a result of change in location. Few studies are performed on visual adaptation in virtual worlds and its consequences on learning and engagement experiences. Visual adaptation for learner personalization has been depicted by a model of Chittaro and Ranon (2007). This research also reports the lack of investigation on the effects of adaptation.
Chittaro and Ranon (2007) suggest careful adaptation as it can also yield negative experience, if not done properly. The model is based on the Adaptive Hyper Media Architecture (AHA) by De Bra, Aerts, Berden, de Lange, and Rousseau (2003). With the model by Chittaro and Ranon (2007), the system provides navigation support to guide the learner to suitable VE objects and navigated places while updates to the user model also takes place based on the learner interaction style and with the 3D object and movements in the environment. The system architecture is depicted in Figure 33. The authors describe the system as follows: “The usage data sensing component sends relevant usage data to a usage data analysis component, which turns the data into user model update and content requests for AHA!, which sends the personalized content through a content transformer component before it’s passed to the user’s browser. Dotted boxes represent optional transformers that can be added in specific domain or application scenarios.” (Chittaro & Ranon, 2007).
Usage data can be user position, orientation, or actions. AHA interacts with three components, namely: domain/adaptation model, user model, and application content (see Figure 33). The domain/adaptation model represents educational application adaptation rules and conceptual structure. While adaptation in AHA can provide adaptive presentation and adaptive navigation (see Figure 34) through rules, this model provides a generalization through the content transformer component. The model does not consider pedagogical agents and their possible roles in adaptation as well as interaction with other components.
Figure 33: The system architecture proposed by Chittaro and Ranon (2007) for adaptive hypermedia in a virtual learning environment.
Figure 34: Adaptive techniques in AHA (De Bra, Aerts, Berden, de Lange, & Rousseau, 2003).
Summary
In the visited work, it is found that the environment impacts and provides supporting conceptual models to learning with intelligent agents. Several works visited are in the way to formalize the design of the environments to be both pedagogical aware and supporting the agent approaches. Conceptual frameworks
such as PEGASE, HERA, MASVERP, MASCARET, and AHA provide supporting conceptual frameworks to learning in VLE. Several of which extend on intelligent tutoring systems concepts in 2D towards 3D virtual worlds with IPA support. The visited frameworks help identifying the following conceptual models: pedagogical model, domain model, error model, and risk model. The importance of prior identified models, including: learner model, domain model, task and activity models are also stressed in this study. The two identified models of error and the risk have special importance to manage learner errors and extraneous learner behavior in the environment that both have pedagogical values.
With the importance of collaborative learning activities, and the potential offering of virtual worlds, formalizing it by a collaboration model is highly desirable. The model by Schmeil and Eppler (2009) can act as a reference model for virtual embodied collaboration in virtual worlds, and therefore its components are relevant in formalizing an IPA model in a virtual world that considers collaborative learning. Another relevant and important aspect to learning in recent virtual environments is the need for a formalized an assessment model. And therefore, the IMA model (Wesiak, Al-Smadi, & Guetl, 2012) provides a reference to this need from an integrated and up to date perspective. The pedagogical agent aspects are not necessarily relevant to cognitive and decision abilities. The following Section sheds light into further aspects by visiting further frameworks that has particular support for intelligent pedagogical agents and their integration in an environment.