Adaptive Course Creation for All
Alexandra Cristea
Information Systems
Department of Mathematics and Computer Science
Eindhoven University of Technology
Abstract
Personalization in education is supposed to be the answer to enhanced learning. Adaptive hypermedia has the tools to create this type of needed personalization. However, authoring of adaptive hypermedia is still a complicated problem. This paper presents a system that tries to tackle this problem by allowing a leveled authoring of the adaptive behavior of adaptive hypermedia, corresponding from beginner authors to advanced authors. Moreover, this type of approach allows authors to implement their own conceptions about instructional strategies, corresponding, for instance, to specific learning or cognitive styles.
Keywords: adaptive courseware, adaptive hypermedia, authoring, adaptation language, learning styles, cognitive styles, instructional strategies
1. Introduction
Personalization in every domain is a striving tendency of the modern society. Especially in education, personalization aims to mimic the excellent teaching practices and strategies used by teachers from the beginning of history. In this new society, however, there are some major problems with the historical approach to education. The new trend towards globalization has pushed forward the one-size-fits-all approach, which is more technology-driven than aiming at an educational goal.
A response to the need of personalization is given by adaptive hypermedia, which can respond to user models based, usually, on simple rule systems. The granularity of these user behavior descriptions is, however, very low. This means that rules have to be specified for each access of each piece of material, for each situation in particular. This turns out to be quite difficult for authors, especially in the educational domain.
Previously, we have been developing a theoretical framework, LAOS [6], for specifying the components of the conceptual structure of a course, as well as the meta-data needed. We have already implemented a part of this framework in a system called My Online Teacher (MOT) [7], tested both in theory and in real classroom setting [5].
This paper, however, concentrates on the dynamics of the LAOS model and its implementation in MOT, also called the adaptation model. Specifically, we look at how instructional strategies can be implemented with the new extension of the MOT model. These instructional strategies can reflect specific learning styles, creating adaptive strategies that can serve as a testing bed for accumulating evidence in support of or against specific learning styles theories. Similarly, instructional strategies can correspond to specific cognitive styles. Finally, they can just correspond to some arbitrary educational goals set by the authors themselves. We will exemplify the above with some examples. The learning styles approach is justified by cognitive psychology, and is appealing because it forces the author to adapt to the student and not the other way around. In other words, educators should not ask, ‘Is this student smart?’ but rather ‘How is this student smart?’.
However, although believing that personalization is the answer to many problems in today’s online education, and that adaptive hypermedia is the appropriate vehicle for it, our main purpose is to make it easier for authors to create their own ideas of instructional strategies, and not to enforce a specific approach.
Moreover, a major issue is that the created adaptive strategies corresponding to instructional strategies are reusable by other authors.
The remainder of this paper is structured as follows. Section 2 analyzes similar research and systems. Section 3 shows how various instructional strategies (corresponding or not to specific learning styles) can be implemented in MOT, as well as how updates can be implemented also during the interaction learner system. Then we draw some conclusions in section 4.
2. Background study
There are very few researches that have studied the implementation of learning styles in adaptive hypermedia. We shall shortly review them in the following, grouped by the learning style that they are trying to respond to.
For instance, a popular learning style treated recently is the field-dependent (FD) versus the field-independent one, as in AEC-ES [24]; this division shows similarities (in features and treatment) with the global-sequential dimension of the Felder-Silverman model, as in LSAS [1] and CS388 [4]. FD learners prefer experiences in a global fashion, adhere to structures, learn material with social content best, attend best to material relevant to own experience, require externally defined goals and reinforcements, need organization, be more affected by criticism, use observational approach for concept attainment (learn best by using examples). AH systems respond to these needs by providing navigational support tools (e.g., concept map, graphic path indicator, advanced organizer) understanding the structure of the knowledge domain. In terms of adaptive hypermedia, these techniques fall into the category of adaptive navigation support. FI learners perceive analytically, make specific concept distinctions, little overlap, prefer impersonal orientation, may need explicit training in social skills, are interested in new concepts for their own sake, have self-defined goals and reinforcement, can self-structure situations, are less affected by criticism, use hypothesis-testing approach to attain concepts. AH systems respond with learner control options, allowing arbitrary succession of course material. Some system allow explicit learner directed switch between the two treatments.
Other systems implement strategies corresponding to the different sensory preferences, as in ARTHUR [12], iWeaver [24], MANIC [21], CS388 [4]. The adaptive presentation in AH can switch automatically between textual – (text, hypertext), auditory - (sounds, streaming audio), visual - (streaming video, slideshows, hypermedia) and kinaesthetic preferences (animations, simulations, puzzles).
Other learning styles, such as the sensing-intuitive dimension in the Felder-Silverman scale [10], are used in Tangow [16]. Sensing learners tend to like learning facts, intuitive learners often prefer discovering possibilities and relationships. Sensors often like solving problems by well-established methods and dislike complications and surprises; intuitors like innovation and dislike repetition. Sensors are more likely than intuitors to resent being tested on material that has not been explicitly covered in class. Sensors tend to be patient with details and good at memorizing facts and doing hands-on (laboratory) work; intuitors may be better at grasping new concepts and are often more comfortable than sensors with abstractions and mathematical formulations. Sensors tend to be more practical and careful than intuitors; intuitors tend to work faster and to be more innovative than sensors.
Sensors don't like courses that have no apparent connection to the real world; intuitors don't like "plug-and-chug" courses that involve a lot of memorization and routine calculations. To respond to these types of preferences, teaching strategies involve selection of appropriate meta-data (such as concept attributes of the form ‘example’, ‘activity’, ‘theory’, ‘exercise’) and corresponding AH adaptive presentation.
Finally, Kolb’s learning styles are applied, e.g., in INSPIRE [20]. Kolb learning styles model is based on two lines of axis (continuums): our approach to a task - (preferring to do or watch), and our emotional response (preferring to think or feel). The resulting learning styles are: activist (accommodator: doing and feeling preferences, or concrete-active); diverger (diverger: watching and doing, or concrete-reflective); theorist (assimilator: watching and thinking, or abstract-reflective); pragmatist (converger: thinking and doing, or abstract-active). The AH response to these styles is to show different sequences of information, by using adaptive navigational support techniques with link annotation.
Many of the systems above assess learning styles via psychometric questionnaires, and thus classify learners into stereotypical groups before the actual learning starts, and no updates are possible. AH actually offers the tools to bypass this problem as we shall see.
The main difference to our approach is that we offer an authoring tool that can build different adaptive strategies corresponding to different learning styles, but is not bound by it. There are no explicit limitations to what exactly the adaptive strategies may represent, or to what subjects and concept structures they can be applied, as shall be seen in the following section.
3. Adaptive strategies in MOT
In this section we will look at the different ways instructional strategies can be created via the adaptive strategies written in the MOT AH authoring system. MOT[1] [17] was first introduced in [7], so we will not describe the system again here. However, the functionality of the system can partially be seen via the examples given. Adaptive strategies in MOT can reflect ‘simple’ instructional purposes as well as more complex learning styles and their respective instructional strategies, as can be seen in the following.
3.1. Ingredients of different presentations
One of the basic, major ingredients of learning styles is to be able to provide different learners with different presentations of the learning material (such as explanations, theory, exercises, etc.). Figure 1 depicts how different possible presentations can be authored in MOT. The left frame represents the hierarchy of created concepts within the concept map entitled ‘Intelligent Tutoring Systems’; the right frame shows the different possible presentations of a specific concept, called ‘Anderson’s Geometry Tutor’. If we would be the creator (and not ‘Genius’) we would also see in the left frame a button called ‘add attribute’ which would allow us to add an unlimited number of other different attributes. These attributes can be in concordance with a given learning standard (such as SCORM [20], LOM [18], etc.).
These attributes are the meta-data that can be used in various interpretations of the learning contents, as specified by different learning strategies, as we shall see.
3.2. Ingredients of different ordering
The other major ingredient of the implementation of different learning styles is providing different learners with different ordering of the material. Ordering does not happen in MOT at the level of the concept maps as in Figure 1. This is due to the fact that ordering is reflecting the goal of the presentation, the audience that it is aimed at. MOT therefore allows a different layer, for storing relations between concepts that are inherent to the presentation. This layer is called in MOT the ‘lesson’ layer. Figure 2 shows an instance of the lesson layer in MOT, for the lesson ‘Introduction Into Adaptive Hypermedia Systems’. The validity of the introduction of this extra layer has already been proven by previous real-life tests with students [4].
As can be seen in Figure 2, the ingredients of the lesson layer are (almost) the same as the ones in the concept layer. Actually, the lesson layer is a restricted, constrained version of the concept layer. The type of restriction applied has something to do with the type of presentation desired – so can come as an answer to the requirement of a specific learning style. It is easy to see that restrictions can imply selecting only attributes of a specific type, such as only ‘explanations’ or only ‘exercises’. Here we are missing, for instance, attributes such as ‘exercise_intern’, ‘text_intern’, etc.
3.3. Adaptive strategies based on ingredients
In [9] we have introduced the basis of an adaptation language, which tries to identify and represent the repetitive patterns that appear in adaptive hypermedia, not in terms of concept representation, but in terms of (adaptive) concept use. This language allows the usage of general concepts, as well as concept instances. More importantly, for the purpose of the current paper, it allows the creation of adaptive strategies written in this adaptation language. This language is implemented as one of the newer components of MOT.
In the following, we show how simple instructional strategies can be written in MOT. Figure 3 shows the usage of the special, newly introduced attribute ‘text_novice’ from the concept map in Figure 1. This simple adaptive strategy says that, if the user model specifies the learner to be a ‘novice’, and there is a ‘text_novice’ attribute present in the concept[2], then this attribute (representing the version of the concept which is considered appropriate for this user) can be shown to the learner.
The ‘enough()’ construct is introduced elsewhere [9]. Here is suffices to remark that it guarantees that the two conditions written inside its brackets are true.
It is important here to note that this simple strategy of a few lines[3] can be applied to any concept map written as in Figure 1, and not just to the one example shown there.
Obviously, similar adaptive strategies can be written for the other extra attributes added, such as ‘text_expert’ and ‘text_intern’. Instead of showing that, Figure 4 shows how to deal with specific exercises, such as the exercises only designed for advanced users. The writing of the rest of the strategy in Figure is symmetrical to the one in Figure 3.
3.4. Adaptive strategy for Field Dependent learning style
For exemplification of how learning styles, paired with some instructional strategy, can be transformed into an adaptive strategy, we show first the treatment for the ‘simple’ field-dependent learning style.
In MOT, instructional strategies corresponding to learning styles can be authored via the same frame authoring tool [3] used for simple adaptive strategies. The authoring steps are as follows. First, the description of the strategy can be specified, as in Figure 5.
It is customary to create for FD learners a layered view on the subject matter, starting with a very general, structural view and then gradually advancing through the details. By using the lesson structure and order, as shown in Figure 2, such an FD adaptive strategy can be easily constructed. Figure 6 shows how each level is declared readable, with the newly created ‘readleveldepth’ procedure.
To keep things simple we are not going into the details of the definition of this procedure, we just need to explain shortly how it behaves. For the first level, it will allow, for the example in Figure 2, the reading in order of the items ‘title(Introduction Into Adaptive Hypermedia)’ and ‘text(Hypermedia …)’. Please note that the empty fields were jumped over. Next, the level will be increased, and the first row of sub-lessons under ‘[Applications]’ will be available for access. The numbers in Figure 2 serve for the prescribed reading order. This goes on until the whole lesson map is read.