Keywords

1 Introduction

The interest in studying a foreign language is increasing in the last few years. The globalization process and the general access to Internet can be pointed out as the main factors that explain this tendency.

The number of students that are enrolled in languages academies and schools is increasing each year. Moreover, the number of bilingual studies is also increasing at all levels: Primary, Secondary school and universities.

At the same time, the number of different methodologies to learn a foreign language is increasing in the same proportion.

Technology has taken part in this process as it does in many other fields. So, there are many applications and tools offered to users that are related to this learning process. The rapid growing of smartphone usage has provoked that mobile apps be, together with the Web, one of the favourite environments to learn a foreign language.

The development of this kind of application is something complex due to the diversity of learning language methodologies, the variety of execution environments (Web, mobile and desktop) and the number of different technologies that can be used. This why the proposal presented in this article, make use of Model Driven Development and Model-Driven Architecture approaches, as a mean to develop complex solutions maximizing the savings in time and cost.

This article presents a first result in this direction. It consists in a metamodel that models the main features that are present in most of the learning language methodologies. The solution is illustrated by modelling two similar activities from two language learning methodologies, using almost the same model.

The paper is organized as follows: Sect. 2 describes the research context together with the related work. Section 3 presents the metamodel proposed, including a description of the main blocks with which is composed. Section 4 shows an example that illustrates the concepts. Section 5 includes an analysis of additional modelling capabilities derived from our proposal. Finally, the conclusions and future work are presented in Sect. 6.

2 Research Context

This section contains the research context, which includes two main elements. First, essential concepts and general features extracted from different learning methodologies are presented. Secondly, selected related work in the field of Model Driven Development is presented.

Learning a foreign language is a process that involves different methods, techniques and tools, each one appearing more effective than the others. In this paper, we have focused in those methodologies that offer some kind of technological support (Web site, mobile app or similar).

Among the methodologies object of study are Lexiway [1], Duolingo [2] Lexiway is a newer methodology that emphasizes the pronunciation, both in vocabulary and the sentence composition rather than in grammar. Lexiway offers a variety of tools: a virtual classroom and an iOS app. Duolingo is a method to learn English that takes advantage of the generalization of mobile apps. This method emphasizes the student motivation and it is based on gamification [6].

Although these methodologies offer different viewpoints, it is possible to identify some common elements. The Model Driven Development approach will be responsible to abstract all the elements that are present in the different methodologies. However, before we can perform this abstraction process, we first must study some of the most relevant methodologies to learn a foreign language.

The Model-Driven Architecture approach is proposed by the OMG (Object Management Group) in 2011. In the MDA Guide [5], MDA is defined as a solution to use models in the software development. This solution gives a main role to the models in the systems under development. As the name indicates, it is said model driven since it runs all the phases from inception, design, building, development, maintenance and modification.

In the MDA core, there are a number of important OMG standards: the Unified Modeling Language (UML), the Meta Object Facility (MOF), XML Metadata Interchange (XMI) and the Common Warehouse Metamodel (CWM). These standards define the central MDA infrastructure, which has been successfully used in the modelling and development of modern systems.

In the field of Human-Computer Interaction it can be found an approach that makes use of models to obtain the user interface (Model Based User Interface Development or MB-UID) [3]. In this way, since this work is focused in the development of interactive systems, MB-UID can provides us some interesting aspects.

The general opinion is that MDD is going to play an important role in the near future of software development, if only more power MDD tools be delivered. Maybe this is the reason there are so few related works that formalize by means a metamodel to support a foreign language learning methodology.

Nevertheless, there are a number of related works that use a MDD or MDA approach in the development of Web applications [8,9,10]. In [4], MDA is used to develop a music learning application.

A methodology to model e-learning applications can be found in [7], although this methodology is not focused in developing foreign language learning applications. Moreover, our approach presents a set of models at different abstraction levels (architecture), which show different application viewpoints depending on the abstraction level.

3 The Proposed Metamodel

After the study and analysis of the different foreign language learning methodologies, we are going to get, through an abstraction or generalization process, the common elements to all of them. In this section, these common elements and its formalization by means an ECORE metamodel with OCL (Object Constraint Language) constraints, are presented.

First, we are going to describe the common elements of the language learning applications used by the different methodologies.

All the analyzed applications manage concepts (it can be words and sentences). These concepts are organized hierarchically (for example lesson, units and so on). Each concept is associated to a set of resources, for instance, the concept “house” can have associated one or more images, audios, videos.

Secondly, all the language learning applications have exercises. There a great variety of exercises like multiple choice, filling the gaps, sentence composition and so on.

The third pillar in all methodologies is the established order by which the concepts are studied. For instance, in a particular methodology, a student only can begin a determined level if only she/he has passed the previous level. This established order defines the workflow.

In summary, the common elements are:

  • Concept and Concepts hierarchy

  • Resources

  • Exercises

  • Workflow

The next step is to model, by an abstraction process, all the common elements considering these general MDA principles:

  • Abstraction

  • Reuse

As a summary of the abstraction process, we can say that a methodology model (instance of Methodology) defines four different concerns: workflow (Workflow package), content (Content package), media (Media package) and presentation (Activity and Presentation package).

Figure 1 shows different packages representing the common elements. The central package of this model architecture has been called Methodology and it is divided into 5 packages: Workflow, Content, Media, Activity and Presentation. A sixth package called commons contain utilities that will be explained later.

Fig. 1.
figure 1

Metamodel package model

Full size image

The next paragraphs describe the mapping among the common elements and the packages.

The first common elements are the “concepts” managed by the methodology, which are modeled by the different elements of the package Content (it contains the metaclasses Concept and ContentContainer). The hierarchy is also modeled by these same package (specifically by the metaclass ContentContainer).

The resources are modeled by the elements defined in the Media package.

The exercises are organized into two packages (Presentation and Activity). The Presentation package represents the particular kind of exercise as, for example, multiple choice or filling the gap. The Activity package represents the content of the exercise as, for instance, the different options of the multiple exercise or the statement.

The last common element, the workflow is modeled by the elements defined in the Workflow package.

A detailed view of the metamodel is shown in Fig. 2, which presents the defined metaclasses.

Fig. 2.
figure 2

Metamodel that models language learning methodologies (Color figure online)

Full size image

The Content package (see purple metaclasses in Fig. 2) is used to represent the organization of the conceptual content that will presented to the student. This content has a tree structure that allows designers to establish relationships among concepts depending on the strategy defined by the methodology. For example, in a given methodology, a level can be composed by units and, a unit can be composed by lessons. The metaclasses Content, ContentContainer and Concept are defined following a composite design pattern.

The Media package (see turquoise metaclasses in Fig. 2) is used to represent the resources that will be used in the presentation. It defines four kinds of media: audio, text, video and image (Audio, Text, Video and Image metaclasses), which can be combined to represent complex media (for instance, text and speech).

The concepts of the Content package are associated to elements of the Media package.

The Presentation package (see green metaclasses in Fig. 2) is used to represent the graphical interface of the different exercises offered to the students in the learning process. This package defines the concept of slide (Slide metaclass), which is associated to a particular activity (Activity metaclass, from the Activity package). This feature allows the reutilization of a slide’s view (Slide metaclass) in different activities (Activity metaclass) and vice versa.

The package Activity (see orange metaclasses in Fig. 2) provides the capability to parametrize the functionality of the activities that are carried out in the learning process. Every Activity metaclass provides information to the user (Ground metaclass), which is used to define an exercise or activity through elements of the MediaContent (the text of the statement). Additionally, an Activity metaclass defines information provided by the user through the Gap metaclass in order to complete the exercise. A Gap metaclass defines set of options and answers using the Option metaclass. Option metaclass is associated to a concept (Concept metaclass), which is associated to a MediaContent.

Let us to illustrate this with an example based on multiple choice exercise. In this case, the exercise consists in playing a speech and showing different images where one of them corresponds with the speech. This exercise is represented by the MultipleChoicePhotoText metaclass, which defines the mechanism (playing the speech and the possibility to select different options). Both the speech (Ground metaclass) and the options (Gap) is parametrized by an instance of Activity.

The Workflow package (see grease metaclasses in Fig. 2) allows us to represent the application presentation’s flow (Workflow metaclass). In order to represent this information flow, we adopt the classical graph notation, where the nodes represent the states (State metaclass) associated to a representation (Slide metaclass in the Presentation package), and the edges represent the different transitions (Transition metaclass) among states or slides.

Finally, the Methodology package (see light grease metaclasses in Fig. 2) defines the metaclass Methodology, which represents the whole methodology.

All the metaclasses in this metamodel descend from the Entity metaclass, which is defined in a package called Commons (see yellow metaclasses in Fig. 2). This package provides other metaclasses with the identification (Entity metaclass) and extension (Property metaclass) capabilities.

Additionally, an editor has been developed to manage (create, edit and verify) the models according to the metamodel. This editor has been developed as an Eclipse plugin, by using the Eclipse Modeling Framework (EMF). This plugin has been defined by using OclInEcore, which is a dialect of the OMG Essential Meta-Object Facility (EMOF). OclInEcore allows us to define OCL constraints that are used to define the model invariants and queries.

4 Case Study

The main goal of this case study is to illustrate the flexibility and adaptability of the proposed metamodel to model different learning language methodologies. Therefore, we will show an example of model that includes an activity based on a multiple-choice exercise. During the activity, the student hears a word and she/he must select one option among four pair image-text.

This case study will show the common elements of two similar activities taken from two different methodologies (Duolingo and Lexiway).

Figure 3 shows how to model the presentation of the activity in the case of Lexiway. In that figure, we can appreciate two parts. First, the exercise (GUI) as is presented to the student and, secondly, the model is shown (particularly, the workflow).

Fig. 3.
figure 3

Two slides presentation and workflow in Lexiway (Color figure online).

Full size image

The example is based in Lexiway Junior (a version of the method specially adapted to young people), and it has only two slides: the first one contains only information on the exercise; the second one includes the multiple choices exercise.

Let us review the metamodel concepts explained using this example.

As it has been mentioned in the previous section, a methodology model (instance of Methodology) defines four different concerns: workflow, content, media and presentation (see Figs. 3 and 4).

Fig. 4.
figure 4

Presentation in Lexiway and Duolingo of a multiple-choice activity using the same elements.

Full size image

With the Content package we can represent any conceptual organization. In the case of Duolingo (a method to learn English using Spanish as a original language), the organization is based on 59 units (ContentContainers). Each unit has a variable number of lessons (ContentContainers). Each lesson contains a set of new and related words (Concepts). Once the user has studied these words, she/he has the possibility to reinforce them in later lessons. In the case of the Lexiway Junior, the method is organized in 6 blocks (ContentContainers). Each block is organized in 4 units (ContentContainers), which contains 2 lessons each. A lesson contains 16 words grouped in 4 quarters (ContentContainers).

In this example, we define a lesson that contains four concepts.

From the media model viewpoint (MediaModel metaclass), each of the four concepts has associated a set of multimedia resources (ComposedContent metaclass), which are composed with different image (Image) and text (Text) resources. This model also defines the media elements that are related to the exercise statement: audio (Audio) and text (Text). In this exercise, the statement text is defined as “¿Qué palabra has escuchado?” (“What word have you heard?” in Spanish) and the speech is “Box”. This requires the definition of another composed multimedia content resource (ComposedComponent).

The Activity package is used to define the needed information to carry out the exercise. This information is contained in an instance of Activity (Activity MultipleChoiceActivity) that includes, both the statement (Ground MultipleChoiceEnunciado) and the user’s answer (Gap MultipleChoiceGap). In this case, the Gap has 4 possible options (Option) and only one established as the right answer.

The Presentation package is used to model the activities views (Slides) of the methodology. The views in this example are the following: an activity with the introduction to the lesson and a multiple choice activity. The modeling of the introduction to the lesson, a presentation view is instantiated (FrontPageSlide portada-lección-01) that defines a null activity (Activity NullActivity), which contains the information of the introduction. To define the view of the multiple choice activity, a multiple choice view is instantiated (MultipleChoicePhotoText QueHasEscuchadoText), linked to the MultipleChoiceActivity.

The package Workflow is used to define the workflow model, where two states (States) and one transition (Transition) with the precondition established to true, are defined. Each of the states is associated to a view (see the blue lines in Fig. 3). So, the transitions among them define a change of views. Graphically, this transition is showed as a button (always enabled due to the precondition) on the user interface (see green line in Fig. 3).

The Fig. 4 shows the correspondence among the different elements of the model of the case study, the multiple-choice activity from both methodologies Lexiway and Duolingo. Figure 4 illustrates the expressiveness power of this technology, since it shows how the same model is used to represent two similar activities from two different learning methodologies.

5 Analysis of the Proposal

In this section the different modeling alternatives are analyzed. The different options will depend on the concerns that are defined in the learning processing methodology.

Considering the modeling of the learning methodology content (ContentModel), it allows define different kinds of structures. Let us see an example. In the case of Lexiway, a structure is modeled in 5 levels (level, block, unit, lesson and item); and in the case of Duolingo, the structure is modeled in 3 levels (units, lessons and words).

The resources can be organized using a tree structure through the MediaModel definition, allowing the reutilization of resources from different methodologies. The independence between the content and resource models (ContenModel and MediaModel) allows that a resource can be associated to different concepts (Concept) and vice-versa. Additionally, the independence between the resources (MediaModel) and activity (ActivityModel) models allows the reutilization or interchange of resources among different components (ActivityComponent) of an activity (Activity).

The modeling of the exercises is carried out by defining two models: the presentation model (instance of Slide metaclass) and the activity model (instance of Activity metaclass). This separation allows us to reuse different activities taken as invariant the same presentation behavior (see Fig. 4). So, it is possible to define different activities such as the following:

  • Filling the gaps (either drag and dropping or by writing)

  • Sentence composition

  • Sentence translation

  • Associate elements from two sets

  • Sentence repetition

  • Write the lacking word

  • Speech to text

  • Multiple choices (with different options)

Likewise, it is possible to represent different kinds of exercise. For instance, the introduction of a unit or lesson, different concept layouts distribution, and so on.

In the case of modeling the learning methodology workflow, it is possible to define any workflow by using a conditional graph as, for example, to establish an order for the lessons or to decide what lessons are enabled to be studied. In the case of Lexiway, for instance, the student must pass all the lessons in a unit before she/he can start the next lesson. Inside a lesson, the “presentation” is mandatory before accessing to the “practice”, “variations” or “spelling” lessons. Moving to Duolingo, there are lessons or tests that, once the student did pass them, she/he can avoid some lessons.

Finally, it is important to note that by using the Commons package, we can model class instances, with the capability to identify (Entity) and to extend (Property) them.

6 Conclusions and Future Work

In this article, a metamodel that allows us to collect and represent the main features that can be considered common to most of the learning language methodologies, is presented.

This metamodel is the first and fundamental piece in the Model-Driven Architecture, which will allow us to develop learning language applications.

Our MDA proposal consists in defining four models: Content, Media, Workflow, and Presentation. Each one represents a different view that allows us to get the projection of the modeled methodology.

To illustrate the flexibility of the proposed metamodel, an example has been presented. This example is used to compare the modeling a similar activity taken from two different methodologies. Therefore, it has been proved that the resultant models do not vary significantly and they have a high degree of reuse.

All the defined models follow the OMG standards, which guarantee the interoperability among them and other external models that follow the OMG. This interoperability is implemented using reflexive editors based on the Eclipse Modeling Framework, which allow us to create Eclipse plugins.

Future work includes to develop a GMF based tool to get and verify models obtained with our metamodel. After that, the convenient transformations will be developed to generate code from the models.

Finally, a customized tracking system for the users is being developed, which it will allow us to analyze their profiles and experience when using the language learning applications.