MTeach: a simple production framework for context-based educational hypermedia.

In this article we present the main results obtained through our experience in designing and implementing MTeach, our authoring framework for didactic hypermedia production. MTeach is based on a methodology supported by a language and a compiler, and its goal is to help the author of a didactic production to work at a higher level than conventional authoring tools. Starting from a formal description of the lessons in MTLDL language, the running code of the multimedia application is generated in standard formats, providing a self-consistent and easy-to-navigate hypermedia structure.

**********

Automatic link generation enhances reusability of the multimedia material after it is organized in a lesson structure. This approach allows authors to create different context-based presentations starting from the expected skill of the target users. Advanced indexing schemes will be further available for dynamic and interactive hypermedia generation. The MTeach framework is standard compliant in all phases of the multimedia production process: the output of each activity in MTeach can be easily integrated with open e-learning platforms in a complementary fashion.

The MTeach project is also a workspace for nearly real-time multimedia production. Interfacing the teacher's console with a multimedia-processing unit, lessons and knowledge models can be directly edited and assembled in a final hypermedia product. Therefore, this article presents also the cost and time effectiveness of MTeach as a fast prototyping environment for experiencing courseware production and reuse in advanced distance-learning methodologies.

KEY FACTORS IN DIDACTIC HYPERMEDIA PRODUCTION

Didactic represents a natural application field for multimedia systems and authoring techniques, since traditional live lessons are "multimedia": the teacher talks, while explaining slides and pictures, showing objects and so on. With some improvements and adaptations, live lessons are published on an interactive CD-ROM or a web site, and adopted as courseware for media-based training and distance learning courses. Advanced models too require the leading communication be still represented by a virtual teacher's figure explaining "how," "where," or "why;" nevertheless, the accustomed, book-fashioned description of the argument will be gradually replaced by dynamic hypermedia guiding systems. The teacher can decide in advance the path the student has to follow, so that the lesson's outline is the same for all students. This is suitable for virtual classrooms having a well-known and mostly equal knowledge level, but it might have different effects on the learning performance of each student. Such a teacher-driven approach can be enhanced providing distinct learning paths for different difficulty levels, or incremental profiles: in many models, basic profiles can view a simpler version of the arguments, while details and/or advanced explanations are accessible only to certified users or gradually obtained after an intermediate evaluation test.

Context-based adaptive systems are more flexible and let the learning path to be created dynamically, on behalf of the single student's knowledge level. Starting from a common set of elementary multimedia information related to a particular argument, lessons are built up dynamically and paths are targeted on the student's profile. While users play their learning session, each interaction is recorded and analyzed by an expert system, trying to identify the student's knowledge level and to match his expectations; in this case, the learning approach is student-driven, so that the student will work by him/herself to discover the answers.

End users are commonly accustomed to the learning method imposed by the traditional didactic approach, that is, live lessons that are normally accompanied by papers, books, and personal lecture notes. Therefore, many multimedia courses strictly respect the logical organization that characterizes traditional courses and lessons. However, it is likely that in a few years this constraint will no longer be present. The increasing diffusion of personal and mobile computing devices, such as e-books, PDAs, and third generation mobile systems and services, will give a larger application horizon to computer-based learning systems. As a consequence, this will radically transform the way of teaching and thus the concept of learning. From the authors' point of view, even if the final product will be based on innovative didactic models, the didactic elements should be collected and organized at first within a traditional schema, guarantee the completeness and to help the identification of every semantic relationship.

The advantages of both traditional and innovative didactic models join themselves in a synergic fashion with the growing experience in computer-based video and multimedia production. Unfortunately, a well-known effect is that even few hours of teaching can generate a huge amount of information (text files, audio, video, pictures, etc.). Managing and handling large files and maintaining their discernibleness in ramified data structures become, in practice, a difficult task for both end users and producers.

To guarantee the lifetime and the maximum scalability of the multimedia learning environment, the teaching material must be regularly structured and reusable. Such requirements imply a set of technical issues regarding both data sources and hypermedia description formats. In order to obtain the full modularity at each layer, starting from data up to the project schema, the original multimedia sources should preserve their own identity in the output data structure. Open data formats should also be preferred in hypermedia production for both sources and generated output. This assumption achieves the largest cross-platform compatibility and allows collecting, modifying and reusing the training material at any time and with the lowest conversion cost. It also gets through the disadvantages of most proprietary file formats, which require their own run-time libraries and are often platform-dependent.

Any structure of the hypermedia navigation deriving from an automated generation system should support a well-defined set of cognitive and didactic methodologies. On the other side, the organization of the didactic material depends strictly from the explicit or implicit pedagogic approaches presented by the authoring tool. In a modern approach, alternative strategies in didactic methodology require a flexible production platform, where the same information elements preserve their own distinctness and can be easily combined using different lesson description models.

This article is organized as follows: the next section addresses the relationships with previous works, then the main scopes and technical concepts of the MTeach project are presented, and the next section explains the reasons of a language-based approach in our tool. There are three sections that describe the workflow in the MTeach framework, including details of the language for describing the lessons and the compiler, while in one of those sections a step-by-step example explains how MTeach leads to a real-time multimedia production (virtually no post-production is needed). Further sections illustrate the cost and time effectiveness of MTeach in a case study of didactic hypermedia production. Finally, the last two sections address the integration of MTeach with e-learning platforms, and introduces the rising affinity between MTeach and innovative, adaptive, didactic models.

RELATED WORK

Specific works on authoring environments that are aimed at real-time production for didactic purposes are not as much developed as other branches of hypermedia research. On the other hand, we observe that the emerging of recent, innovative infrastructures for this special kind of hypermedia authoring and generation could directly benefit from more extensive models, systems and methodologies that have been worked on during last years within the hypermedia researchers' community.

The stages in the process of generating a hypermedia presentation are often represented in an "information pipeline" model: Hardman and Bulterman (1995) offered a well-defined methodological approach and traced some interesting guidelines in identifying the workflow for the hypermedia production. They also addressed the main research issues, particularly semantic relationships and consistency in hypermedia, that have been further experienced in different fields by various research activities (Mangiaracina & Maioli, 1995). MTeach extends a linear workflow in a complementary fashion, offering a production framework that is based on a concurrent approach, and that doesn't depend on implementative constraints, as shown in "MTeach in Action: Nearly Real-Time Hypermedia Production"

Some other studies (Teixeira, Paris, Madeira, Brisson-Lopes, 1999) addressed encouraging issues on the courseware reuse processes and the estimation of their effectiveness, and thus give a good motivation for experiencing and evaluating the MTeach flexibility in this application field.

A useful reference for hypermedia description languages may be found in literature (Roisin, 1998; see also Hardman, Bulterman, & Van Rossum, 1994; Van Ossenbruggen, Hardman, & Rutledge, 2002). Such models can be appreciated for their well-structured approach and appear to be extremely powerful when dealing with the correctness of a hypermedia in its formal aspects. In such a field, MTeach proposes a meta-language for high-level integration and structure design, to better address the semantic content of the hypermedia. A meta-language allows management, in a simpler fashion, of the whole hypermedia during the phases of information selection, data structuring, and final generation, as shown in depth by Hardman and Bulterman (1995). It also helps the authors keep the relationships among the elements independent from the complexity of the links at a lower level; such links can be mapped out referring to any hypermedia language, model, or description that has been developed within the researchers' community.

The MTeach project addresses these aspects to provide a simple, structured, and open framework to build cost-effective large didactic multimedia productions (Montessoro, Pierattoni & Cortolezzis, 2001). The framework can also be used to implement innovative strategies to organize and present the didactic material, and to experiment dynamic modeling of the student driving adaptive learning paths (Montessoro, Pierattoni, & Toppano, 2002).

OVERVIEW OF THE MTEACH PROJECT

The MTeach framework is based on a methodology supported by a language and a compiler, and allows the author of a didactic multimedia production to work at a higher level than conventional authoring tools. The main effort in the MTeach project consists of defining a simple methodology that preserves most of the normal preparation and teaching activity of the author and that leads to a straight sequence of production tasks.

Approaching the design of a didactic-oriented authoring framework, software developers should identify the correct guidelines to the final success of the project. These are strictly dependent from the target of the authoring tool; both author's and end user's prerogatives have thus to be satisfied, starting from an investigation on their practical requirements.

A language that provides a structured, high-level design of the hypermedia project is a key contributor to achieve this target; nevertheless, it allows keeping the project development at a semantic level, not depending on toolkit's internal representation rules. Links in the multimedia product are semantic-driven and therefore they can be automatically shaped from such a description. If some changes are made in the source, the final product is automatically updated or rebuilt through the MTeach compiler. Such a feature minimizes the manual work and keeps the author's focus on the contents instead of on the presentation's look. Moreover, semantic links in a didactic product are critical for the correct understanding of the arguments, and must be driven by the teacher. No automatic tool can do the work as well, and even a practiced hypermedia programmer may fail, if he/she doesn't supervise the project at a semantic level too; a language provides a simple, fast and portable way to describe and respect the relationships among the concepts.

Low production cost and skill, easiness of usage and a short production time are functional benefits of the MTeach framework. This is particularly important in hi-tech didactic publications, due to the fast evolution of the presented technologies that makes contents obsolete in a very short time. The distribution of such products occurs after a long and complex editing process, involving many resources and subjects, but sometimes the contents are up-to-date in a few months after the final release. An authoring tool based on a simple and module-oriented language allows the reviewers to easily modify or substitute at any time the only elements in the structure, which are still under development or subjected to periodic enhancements. If new arguments are added or integrated within an existing project, the inner coherence of the structure of the whole hypermedia will be automatically maintained, because each project build redeploys the dependence relationships among the arguments.

A recognizable modularity is further essential to obtain a good reusability of the parts of a product in a fast prototyping environment: this feature makes MTeach a flexible test bed on which new didactic methodologies can be experienced and applied on advanced hypermedia solutions.

ADVANTAGES OF A LANGUAGE-BASED AUTHORING TOOL

Most authoring tools present a graphic environment as user interface, which helps authors develop their projects in a friendly workspace, and often simplifies the identification of a large variety of toolkit features. Predefined models, wizards and templates enhance the accessibility of the instruments; on-line preview areas also show the direct effects of certain modifications on the final product.

A language-based approach could not offer at first sight the same helpful features available in a graphic toolkit. For less experienced authors, the difficulty of learning a structured language could also seem larger than discovering the instructions of a graphical tool. The value of a language-based description is clear when high-level editing activities have to be modeled with sequences of procedures and then automatically applied on large hypermedia structures. For example, shifting and splicing pools of arguments and related clips, inserting batches of synchronized images or adding parts of existing lessons with their own data formats during the editing phase become easier, if all actions can be performed on a source file that describes in a structured language the hypermedia and its internal relationships. Further, the regularity of the hypermedia data structure and the reusability of the teaching material are often issues that are difficult to preserve, if management tools are complex and variously mixed with artistic features in a high-level graphical authoring environment.

Operations are easily reproducible when dealing with a plain text source: many text editors have a set of macros and automation components that are more powerful than repeating sequences of visual commands on a GUI. This fact represents a common issue of most graphic-based applications: system administrators, for example, encounter some difficulties in automating procedures, when related to sequences of actions on graphical elements that are neither mapped on console commands, nor manageable at a lower level of abstraction.

It is important to argue that "language-based" does not necessarily mean "hand-written." As well as the scripts to manage an operating system can be generated or modified by GUI-based programs, the MTeach description can be handled by external programs, as discussed in sections "MTeach in Action: The Courseware Production Process" and "The MTeach Compiler." The advantage of the language-based approach comes from the wide availability of software tools to handle ASCII files and from the easiness of writing specific programs.

STRUCTURE OF THE MTEACH ENVIRONMENT

The MTeach authoring framework (Figure 1) is grounded overall on a language (MTLDL) and a compiler. The automated code generation descends from a structured, production-oriented language, which describes the content of each lesson in terms of primitive multimedia elements: audio, video graphics, text, and so on. These elements are not related to any particular authoring system and are thus reusable in different contexts. The basic framework of the language designs a multimedia lesson with a description very similar to the index of a book. Two benefits come from this first approach: formerly, "traditional" teachers are not disoriented; second, the multimedia project can be compared to reference textbooks for completeness and coherence. The lesson description also contains the logical connections between its various parts in terms of keywords (flat links) and hierarchical structure. In respect of commonly used authoring tools, this approach produces a more rigid, and therefore more regular, hypermedia structure. Navigation becomes easier this way for both author and end user.

From the technical point of view, the system is extensible to several different platforms and operating systems, only by adding new interface modules to the code generator of the compiler. In particular, if an authoring tool provides application interfaces to external hi-level programs, MTeach can control it to reduce or eliminate the need for manual work. When integrated with an e-learning content management platform, the MTeach source code could be directly linked to upper layer control features and managed by an external graphic console. Source code can be directly written and managed by the author in the MTeach language; notwithstanding an automated code generation could be easily integrated in a graphical development environment.

[FIGURE 1 OMITTED]

MTEACH IN ACTION: THE COURSEWARE PRODUCTION PROCESS

MTeach provides a powerful methodology to build very large didactic multimedia products, following as close as possible the normal activity of a teacher. The MTeach Lesson Description Language (MTLDL), allows the teacher to describe the hierarchical structure of the lesson in the same way the index of a book is written; the keyword list associated to each element represents the logical interconnections among the arguments that will become glossary entries and hyper textual links.

Keywords are the entry point to a complex system of logical relationship among the arguments. If the keyword selection is a well-structured process, with a set of scalable hierarchies that may be combined in multiple dimensions, MTeach can support the generation of multiple learning and reference paths. This feature would be appreciated in particular when dealing with dynamic hypermedia delivery systems, so that keyword-based links are dynamically loaded during each run-time visualization according to different, parallel navigation paths.

The kernel of MTLDL, with its peculiar syntax, lexical conventions and data types, is very simple and effective, and thus accomplishes both clarity and easiness of usage. It also demonstrates a natural flexibility that makes it attractive for further extensions. This feature allows us to predict that both MTeach application and research fields will grow larger, and new information elements will appear in its open language, according to the feedback coming from actual applications in academic production environments.

Traditional chapters become "lessons" and the sections within each chapter are called "arguments." The analogy with the book simplifies the description of the logical schema, which can be written with little effort directly from the teacher, not necessarily from a multimedia-authoring expert.

Each argument is a module that groups a collection of different types of documents together, as shown in Figure 2. "Clips," the most important type, are made of audio/video tracks and several synchronized "visual" elements, typically "slides." Synchronization is not performed during the audio/video editing sessions: instead, it is featured by the automatic code generation. The way synchronization is achieved depends on the operating environment at the output: for example, the first prototype used a Real Player plug-in with JavaScript and Real Media synchronization features, while current implementation supports SMIL 1.0 Specification and the MPEG format capabilities. Nevertheless, only the compiler's code generation modules have been affected by this change, but neither MTLDL nor lesson sources had to be rewritten.

MTLDL requires that timing information is associated to each visual element, and related to the starting point of the audio/video track, as described in Figure 3. Timing information can be inserted using absolute or relative notation, in accordance with SMPTE recommendations (SMPTE, 1999) and MPEG time addressing scheme (ISO/IEC 11172-1 and 11172-2 Standard, 1993). This could appear more complex and less effective than a graphical interface, but plain ASCII format still allows easy editing and manipulation when the description regards a large tree structure. MTLDL source can be edited by hand, while batch operations are manageable using macros or scripts from any text editor or ad-hoc program tool. Interactive graphic interfaces do not scale equally well. Moreover, even MTeach may benefit of an additional program, with its own interface, to compute and insert timestamps into the source file (MTeach in Action: Nearly Real-Time Hypermedia Production).

Other documents may enrich the lesson, such as texts, animations, and images, even if not linked to any audio/video track. The language and the compiler are under extension, in order to include further data types and formats, such as PDF documents, static and dynamic HTML pages, and browseable web links.

As disclosed before, a key feature of MTeach is the automated generation of the several links that logically interconnect the topics. A list of keywords (such as "Ulnar," "Exposure," etc.) is associated with each element in the description: this additional information will be parsed to build a multilinked glossary during the code generation. Each keyword (it can be a word or a phrase) in the glossary will show the lists of related topics in the hierarchical index, that is, the whole set of arguments in which the keyword is recalled. We observe that keywords are not mandatory: if absent, no link to the related element will appear in the glossary. Another set of automatically generated links provides a sequential visit of clips of the multimedia lessons. An incoming extension of the compiler will build up glossary links in a multi-dimensional style: this feature will be appreciated in context-based hypermedia generation, supporting user-based personalization and expanding the accessibility of different navigation paths.

MTEACH IN ACTION: NEARLY REAL-TIME HYPERMEDIA PRODUCTION

The MTeach research project is currently a stable workspace for "nearly real-time" multimedia production. MTeach becomes this way the core system in the realization of a stable, real-time production environment.

The MTeach processing and production unit is directly connected with a digital audio and video capture system; live recording is performed through a digital video camera and software for MPEG audio/video encoding. Clips can be recorded in a dedicated room or in a real classroom during a regular teaching session. Special features such as audio/video remixing, needed by practical lab exercises and PC-based training, can adjust the recordings during a post-processing phase. Figure 4 shows the logical schema of the production environment; according to the MTeach production process (see also Figure 1), the sequence of activities may be addressed as follows:

1. Definition of lesson contents and organization

First, the structure of the lesson gets described in MTLDL, either manually or by an automated process on the production unit. This step provides the guideline for the following production phases, and guarantees that the audio and video recordings will match the logical flow defined by the lesson's schedule. However, later modification of the structure in a didactic multimedia product often requires that the teacher's speech is recorded and dubbed again to fit the order of the arguments.

2. Production of visual supports

The teacher prepares slides, pictures, and animations to show during the lesson. In particular, the MTeach platform needs just a preliminary conversion of the slides in a format supported by the visualization plug-in on the client side. Actual implementations of MTeach require an Apple QuickTime plug-in on the HTML browser, which is compatible with GIF, PNG, PICT, BMP, and TIFF images, and proprietary QuickTime and Photoshop image formats as well. Animated GIF images with their related timestamps can also incorporate sequences of animations on single slides. This solution has the advantage that the teacher can use his preferred graphic editor (Microsoft PowerPoint, LaTeX, and Xfig, etc.), but the conversion to a portable format is a standard feature on most design and presentation tools. Most of the didactic material and the lesson's schedule should be ready before the lesson starts; anyhow, teachers can hand-draw pictures during the class, if more exercises or detailed clarifications are needed.

[FIGURE 4 OMITTED]

3. Lecture recording

After that, the lesson is audio/video recorded, possibly live with students, to make the recording as natural as possible. If studio recording is preferred, it is possible to record arguments one by one, possibly following the same order they will appear in the final product. A GIF image browser runs on the teacher's console, which is linked to the processing unit and, eventually, with the video projector in the classroom. The interface between the console and the processing unit sends a synchronization signal in correspondence of a teacher's action (i.e., a slide forward command or a timestamp inside an animated GIF). The processing unit intercepts these signals and includes their information on the fly during the automatic generation of the MTeach source code.

4. Lesson review

When the recording is done, it is a good idea to review what has been shot and whether something, like a few last-minute slides, should be added in order to have a better, more complete final product. This step involves the refinement of the MTLDL schema to produce its final version, the one that is going to be compiled.

5. Video cut and editing (only if needed)

Audio/video recording is further edited, so that live recordings are split up into the elementary modules (clips) defined by the structure of the lesson. Actually, this postprocessing activity will be fully automated (see Figure 4) in the next release of the prototype, currently under development. The MTeach production unit will thus perform a real-time separation and synchronization of the video recordings during live sessions, driven by the teacher's actions and following the MTLDL lesson schema. If the teacher must show live actions on her computer (e.g., showing running videos or software programs) a video mixing console can be used, if available, during the recording. Otherwise, lecture recording video and animated screen capture can be manually postprocessed to mix them into the main video track. Slides, pictures, and animations are modified, if necessary, according to the actually recorded talk; for example, new slides can be prepared to reproduce additional drawings, which could have been hand-drawn on the fly during live lessons.

6. MTeach compiler execution

At this point, all the elements needed for the code generation are ready. The structure described in MTLDL contains all the information necessary to build the links of the hypermedia structure. Compiler's template files will be used as definition of the graphical aspect of the output.

The final code generation is now targeted for HTML output files with SMIL-based content presentations (W3C-Synchronized Multimedia Integration Language 2.0 Specification, 2001) and MPEG audio/video encoding and synchronization (ISO/IEC 13818-1 Standard, 2000). The first prototype required to export all animations in the animated GIF format, including timing information: now it is possible to use a larger set of slide formats, according to their compatibility with the browser's plug-in.

THE MTEACH COMPILER

A brief recap of the process. The teacher prepares the slides and a draft of the MTLDL source. During the recording of the lessons (that can take place in a studio or, better, live with students) actions performed by the teacher, such as passing to the next slide, are intercepted by the processing unit. The timing information they provide is inserted into the description file. This way the production unit can synchronize the elements on the fly. It is also able, thanks to the data already written in the source, to automatically edit the clips. At this point, the only thing left to do is compile the MTLDL file, an operation that can be automated as well.

Automatic link generation makes MTeach different from the conventional authoring tools and provides a good reusability of the multimedia material after it is organized in a lesson structure. Moving a part of a lesson description (clips, slides, timing, keywords, etc.) to another project does not lose the structural information. All the links are rebuilt when the new code is generated: this is a nice feature, since even very short lessons may contain hundredths of links. The MTeach compiler has been developed using lex (Lesk & Schmidt, n.d.) and yacc (Johnson, n.d.). Or, more precisely, their latest incarnations: respectively, flex and bison.

These are tools from the Unix world born to help in developing lexical and grammatical analyzers. A lexical analyzer looks into a stream of characters and recognizes whether those sequences of bits are or are not valid terms of the lexicon of the language it is built around. This is called "tokenizing," because the characters are grouped in tokens. These are then sent to the grammatical analyzer, whose job is to determine if the way the tokens are arranged satisfies the grammar of the language. If this is the case, some code is executed.

As for MTeach, to achieve the largest compatibility across platforms, ANSI C instructions constitute that code. Their goal is to pull out the information contained in the MTLDL file and store it into a data structure that internally describes the hypermedia structure. To respect the hierarchy established by the formal description, the data is stored in a tree structure. With the C language, there is an easy way to create such a structure, and it is by means of linked lists. These are lists whose instances have the capability of pointing to other instances of the same list. In this case, every instance is formed by six elements: three store actual data, three are pointers. The number of pointers is tied to the number of levels an instance--or, better, a tree node--may see: the same level (a sibling node), one up (the parent node), one down (a child node). Most accurately, a node may point to the very next sibling, to the only parent it can have, and to the first of its cute little children. But, as the say has it, one picture is better than a thousand words, so Figure 5 shows a pretty incomplete but, nevertheless, clarifying example.

Figure 5 also shows keywords, even if they are not part of the tree. They are stored in a parallel list that has links allowing them to point to the tree elements they are associated with in the MTLDL source file. Once the tree structure has been built, the compiler uses it to generate the final code. In this phase, the compiler reads a set of template files and inserts into them the actual information described by the internal data structure. The templates contain the instructions related to the visualization, the look, the way the data is shown on the screen, while they obviously lack the actual data. This approach allows the use of conventional editing and authoring tools to build and test the templates. Of course, this is possible only if the internal structure of the file is known, or at least some Application Programming Interfaces (APIs) are accessible to control the editing feature of the authoring tool.

At the current implementation, the generated code is HTML, JavaScript, and SMIL, but in the next MTeach version this is going to change towards an ad hoc browser for a more stable distribution policy. Current directions in the MTeach development progress aim at the integration with MPEG-based streaming services, and at a larger exploitation of the features of both XML and the recent SMIL 2.0 standard. The main purpose of this approach is to improve the flexibility of MTeach and to ensure the conformance of the output process to largely used standard formats, offering it as core engine for web-based distance learning instruments and integrated systems. Open formats take advantage of the features provided by the most common web browsers, instead of using "ad hoc" client solutions, and can radically simplify the distribution and fruition phase of a hypermedia-based distance learning program.

EXPERIENCING MTEACH IN COURSEWARE REUSE

During our activity in multimedia production for didactic purposes, we have dealt at different levels with many commercial authoring tools and e-learning platforms, even before starting the development of the MTeach framework. Our experience in this field allows us to individuate a set of criteria for evaluating the generation process of a didactic hypermedia. These criteria are focused on the following main objectives:

[FIGURE 5 OMITTED]

* global production time,

* easiness of usage, in terms of number of people involved in a production process and their related skill, and

* production cost versus reusability of the product.

The global trend that we observed using common authoring tools revealed that the production lead-time should be inversely proportional to the number of persons involved in the production team. Nevertheless, both time-to-market and product effectiveness depend on a synergic fashion on the coordination among the people in the team. Guidelines should be preferably addressed by the author, who becomes the project manager of the hypermedia, or by an editorial staff, which represents an additional structural cost in the production system. The quality of a didactic production relies thus on the ability of an author to preserve with her management activity the global consistency and the regularity of style, which are key factors to gain the end user's comprehension. From the teacher's point of view, timings and methods of such management activity are often quite different from the schemes of creating a course or a lesson. A practical consequence in this case is that multimedia production costs and management skills are often a big barrier for universities to arrange high level didactic sessions, such as those needed for degree- or master-level courses, that cannot count on the profits deriving from massive sales. MTeach is thus shaped on a methodology that preserves the teacher's rule and, at the same time, helps the organization of a production team.

Our studies and experiments rove over the whole set of activities required in developing and publishing distance learning courses. In this section we present the results of the MTeach project that come from its recent application as start-up platform for e-learning university degree courses.

1. Courseware conversion

The absolute time spent for a multimedia production depends in general on many contingent or subjective factors: the most relevant among them are the type of media, the hardware requirements, the personnel's skill, and the demanded quality of the output. This is also the amount of time associated with courseware creation and production only, and does not include the time spent by the teacher for his/her management tasks. For keeping our evaluations as independent as possible from the variability of such factors, we need a paradigm for comparison. Thus, starting from the case of standard video-recorded lessons (case A in Figure 6), which are still the most common way of implementing distance learning through satellite broadcasting, we'll keep the average production time needed for one study hour as reference for the other cases. In particular, our paradigm regards a studio-recorded lesson, with a fairly small postprocessing for editing and synchronization of slides and video clips, such as lab sessions, practical examples or exercises on the whiteboard, and whose output was a TV-quality master videotape. Beside the teacher, the production team involved in this example at least two video technicians. We observe that lesson preparation activity takes 2/3 of the total time required for publishing a video lesson, while the remaining is necessary for recording and postediting the lesson material.

The same lesson was further reproduced offline for computer-based distance learning (case B in Figure 6), starting from the original lesson material (raw audio/video recordings, PowerPoint slides and the original teacher's outline) and using a graphical authoring tool. In this case we proved that the average additional production time for converting just one study hour was more than 1.5 the time spent in case A. The data conversion and postprocessing phase required the original video material to be acquired, split up into smaller, elementary clips, and then mixed and exported in the authoring tool's format, taking around 1/3 of the additional time. Video clips got further synchronized with slides and captions and imported by the hypermedia authoring tool on a graphical user interface. This was the most time-expensive process: the main activities regarded in this case the inclusion of each visual information, the synchronization and the link establishment between new and existing elements in the developing hypermedia. Even if such operations are quite simple in any graphical authoring environment, it took a long time to repeat them when dealing with a very large number of information elements, that is, about 2/3 of the additional time for courseware reuse. The output was hypermedia in proprietary format on a master CD-R. The production team involved a web developer with proven experience in multimedia authoring, dealing with the teacher for assessing the semantic consistency in the development phase.

[FIGURE 6 OMITTED]

Choosing the MTeach framework for reprocessing the same lesson (case C in Figure 6) allowed us to reduce the lead-time for courseware reuse in a very effective way. Similar to case B, MTeach still requires a postprocessing phase, for the original material to be acquired and split up into smaller, elementary clips: in this phase, to the conversion and re-encoding of video clips in MPEG, we spent approximately the same time as in case B, but it was more than 2/3 of the whole additional work. After converting and editing the elementary clips according to the lesson's outline, the time spent for manual work in hypermedia assemblage was shorter, thanks to the fast prototyping feature of the MTLDL description and the compiler-based approach. The production team involved this time a technician with a fair knowledge of digital video and image processing, while the teacher just prepared the original lesson's outline. The output was hypermedia in standard formats, ready for a multi-channel distribution, that is, for publishing on different kind of media (CD-ROM, Web). With MTeach, the additional production time for reconverting one study hour averages out at 90% of the time spent in case A, with savings of about 44% if compared to case B, that is, using a graphical authoring tool.

2. Courseware adaptation and reassessment 

The previous example is still a kind of courseware reuse. But more important is the possibility to further reduce the production time by re-using the digital material that was previously developed with an authoring tool. The reusability of parts of the final product is mostly limited in common authoring tools, and descends from the strict embedding of the original data sources (audio, video, and image files) inside the code generated by the authoring tool. Some platforms provide a certain concept of modularity, but often it is optimized for data storage and content mining, with a scarce semantic coherence. Cross-portability between different production environments is nearly impossible, since each tool has an internal coding system for the project description, accepting sources and producing archives with different formats. This increases the ratio between production cost and reusability of the product.

MTeach gives the teacher a fast way to reuse compact units of knowledge, which are focused on specific subjects, and a structured methodology for managing the course redeployment process. We tried to enlighten a set of 30 one-hour lessons that were previously published with a graphical authoring tool, giving as output a course with 20 one-hour lessons. All the material had to be re-edited manually by an authoring expert inside the original authoring tool, but all tasks took more time than expected. This happened mainly because it was difficult to individuate the operative guidelines for the course redeployment, and the relations among materials and arguments had to be reconstructed almost from scratch. It also required the teacher to exert a very strict coordination with the operator(s). Repeating the same activity on the MTeach version of the course, it took on average only 30% of the time we spent in the former case.

It is obvious that the results may vary depending on the quality and the complexity of courseware and the level of re-use that it is planned. Our experience demonstrates that a well-structured courseware production, offering a layered and recognizable modularity supported by a language-based approach, is powerful to adapt courseware to specific groups.

EXPERIENCING MTEACH IN NEARLY REAL-TIME PRODUCTION

Dealing with strict time constraints on the publication of the hypermedia, we also experienced the innovative features of current MTeach framework as nearly real-time production environment. As described before, the MTeach production workspace combines recording with postprocessing activity at almost the same time: the manual work for editing the sources is limited to marginal adjustments before the final compilation.

Standard video production is kept as paradigm for the comparison: as proposed in "Experiencing MTeach in Courseware Reuse," it refers to the average production time for a one-hour video-recorded lesson for satellite broadcasting. The actual MTeach implementation gives savings of around 50%; in a stable workspace, it is possible to target at most three hours of additional work for each one-hour lesson.

[FIGURE 7 OMITTED]

This is the time passing between preparing the MTLDL draft of the lesson's outline and mastering the generated hypermedia on a CD-ROM. Its average value is related to a live lesson, and includes also the time needed for preparing the mobile appliances of the MTeach workspace in the classroom. The production team involves a technician with knowledge of digital video and image processing.

MTeach is currently adopted to produce monographic multimedia courses at University of Udine, Electrical, Management and Mechanical Engineering Department. Fields of application span from microneurosurgery (the early prototype and further enhancements were produced with the collaboration of Dr. Alberto Alexandre, appearing in Figure 7) to computer networks. In the initial phase, some lessons have been produced starting from videotapes of the live lecture and PowerPoint files; the actual phase is based instead on the recent extension of MTeach as nearly real-time production environment.

MTEACH AND DISTRIBUTED E-LEARNING PLATFORMS

Previous trends in time and cost requirements have been confirmed also at a higher level, that is, on many e-learning platforms that are specifically designed to support the organizational structure, operations, and workflow of large educational enterprises. In addition to traditional instructor-led training, most of these powerful teaching and learning suites are able to manage curricula and learning paths that include self-paced online content. E-learning platforms are essentially high-level content delivery and management systems, and thus they could be seen as "containers of contents." Although most suites offer a built-in multimedia-authoring tool, the best ones also support standard formats, promoting interoperability with existing investments. This allows interface with third-party content production software, and to integrate them with external synchronous delivery systems, such as web collaboration, virtual classroom, and videoconference meeting systems.

MTeach does not have those features that an enterprise-class architecture for dynamic e-learning may offer for course management and workflow organization. MTeach is a production-oriented framework that is based on modularity and optimized for standard compliance in all phases of the multimedia generation process. Thanks to this fact, the most interesting result here is that the output of any activity in MTeach can be easily integrated with open e-learning platforms in a complementary fashion.

MTEACH AND DIDACTIC MODELS

The modularity of the MTeach language and the scalability of its framework can be appreciated in various teaching experiences and didactic methods. Starting from a well-defined methodology, MTeach allows collecting various elements of a structured lesson and assembling them in parallel learning paths; this way, a single argument or, by extension, an entire hypermedia product could be targeted to the different skill level of the audience. As result of our current research targets, the MTeach platform is going to be a modular system, where each argument inside a lesson corresponds to a particular function, which is linked to different schemes of exposition. In other words, a certain argument could be explained in different ways, using the same set of multimedia material that refers to that particular task. In a bidimensional space, single arguments can be expanded in various description models (e.g., basic-intermediate-advanced, or theory-exercises-practice); the result is a context-based hypermedia where the content is dynamically linked to a certain description model on behalf of the student's skill or interest.

On the other hand, a common set of information elements can also be assembled in different versions at a certain level of difficulty, according to the different project and teaching methodologies explored by the teacher. This approach represents a realistic substrate for dynamic and adaptive learning sources, and for the integration between multimedia production and advanced systems for hypermedia navigation. Additional indexing schemes based on user models are going to be added in parallel to the standard indexing method: this will offer a direct interface for adaptive and user-driven hypermedia navigation to expert systems, which are embedded in many web publishing platforms. This result will be achieved by the future release of MTeach, featuring a dynamic generation of the hypermedia structure on the basis of user-driven adaptation models; this enhancement requires an advanced indexing method, to guarantee the inner coherence and unity of the dynamically generated hypermedia.

CONCLUSIONS AND FUTURE WORK

An innovative environment for multimedia didactic production has been presented. The key feature is the MTLDL, the language for high-level description of the contents and organization of the lessons. A compiler generates the multi-platform running code and the navigation paths, according to the semantic relations between the arguments. This is the basis for dynamic and adaptive learning paths driven by the cognitive model chosen by the author.

The "nearly real-time" production environment will be further applied in seminars and lectures recording to be used in distance learning projects. This material will then be used to investigate three different research directions. First, the multi-channel delivery will be addressed to publish the same production on different media (e.g., CD-ROM, World Wide Web, etc.). Second, several cognitive didactic models will be applied to the recorded lessons in order to evaluate MTeach flexibility. Third, the MTeach production environment becomes yet the production engine for testing of a distributed e-learning platform at the University of Udine. MTeach radically simplifies the production cycle and, by extension, reduces the time-to-market of a hypermedia-based didactic project, improving the manageability of large production processes. Nevertheless, its language-based approach allows keeping the project development at a semantic level, not depending from toolkit's internal representation rules.

keyword_list ::= <identifier>
          [, <keyword_list>])

lesson ::= LESSON <identifier>
           [KEYWORDS = ([<keyword_list>])]
           {arguments}

argument ::= ARGUMENT <identifier>
         [KEYWORDS = ([<keyword_list>])]
         {contents}

arguments ::= argument [arguments]

contents ::= clip | visual [contents]

clip ::= CLIP <identifier>
           MOVIE = <filename>
           [KEYWORDS = ([<keyword_list>])]
           {visual}

visual ::= slide | text | URL [; visual]

slide ::= SLIDE <identifier>
           PICTURE = <filename>
           TIMING = <time>
           [KEYWORDS = ([<keyword_list>])]

Figure 2. Part of the kernel of MTLDL, the MTeach Lesson Description
Language

LESSON "Ulnar nerve exposure at the elbow"
KEYWORDS = ("Ulnar", "Exposure")
{
      ARGUMENT "Sites of entrapment"
      KEYWORDS = ("Entrapment", "Elbow")
      {CLIP "Sites of entrapment"
          MOVIE = "sites.mpg"
          KEYWORDS = ("Entrapment")
          {SLIDE "View of the elbow"
           PICTURE = "elbow1.gif"
           TIMING = 00:00:01
           KEYWORDS = ("Anatomy", "Elbow")

           SLIDE "Inside the elbow"
           PICTURE = "elbow2.gif"
           TIMING = 00:00:10
           KEYWORDS = ("Anatomy", "Elbow")
     }
  }

      ARGUMENT "Surgical steps"
      KEYWORDS = ("Operating techniques")
      {  CLIP "Skin incision"
         MOVIE = "incision.mpg"
         KEYWORDS = ("Skin incision")
         {SLIDE "Starting the incision",
          PICTURE = "skin_incision.gif"
          TIMING = 00:00:01
          KEYWORDS = ("Skin incision", "Elbow")
          ...
Figure 3. A simple example of a lesson described in MTLDL

Acknowledgements

Authors wish to thank Dr. Stefano Caschi for his active contribution during the development of the first MTeach software tool, and Dr. Elio Toppano for his valuable help in addressing dynamic hypermedia navigation driven by adaptive didactic models.

References

Apple Quicktime Developer Documentation (n.d.). [Online]. Available: http://developer.apple.com/techpubs/quicktime/quicktime.html

Hardman, L., & Bulterman, D.C.A. (1995, July). Towards the generation of hypermedia structure, Intelligence and multimodality in multimedia interfaces. [Online]. Available: ftp://ftp.cwi.nl/mmpapers/immil.ps.gz

Hardman, L., Bulterman, D.C.A., & Van Rossum, G. (1994, February). The Amsterdam hypermedia model: Adding time and context to the Dexter model. Communications of the ACM, 37(2), 50-62.

ISO/IEC 11172-1 (1993). Standard, Information technology. Coding of moving pictures and associated audio for digital storage media at up to about 1.5 Mbit/s--Part 1: Systems.

ISO/IEC 11172-2 (1993). Standard, Information technology. Coding of moving pictures and associated audio for digital storage media at up to about 1.5 Mbit/s--Part 2: Video.

ISO/IEC 13818-1 (2000). Standard, Information technology. Generic coding of moving pictures and associated audio information: Systems.

Johnson, S.C. (n.d.). Yacc: Yet another compiler--compiler. Retrieved November 12, 2002 from: http://dinosaur.compilertools.net/yacc/index.html

Lesk M.E., & Schmidt, E. (n.d.). Lex -- A lexical analyzer generator. Retrieved November 12, 2002 from: http://dinosaur.compilertools.net/lex/index.html

Mangiaracina, S., & Maioli, M. (1995, February). Building hypermedia for learning: A framework based on the design of user interface. Technical Report UBLCS-95-8. [Online]. Available: ftp://ftp.cs.unibo.it/pub/TR/UBLCS/95-08.ps.gz

Montessoro, P.L., Pierattoni, D., & Cortolezzis, D. (2001, October). MTEACH: A language-based framework for didactic multimedia production. Advances in Multimedia Information Processing--PCM 2001, 1002-1009.

Montessoro, P.L., Pierattoni, D., & Toppano, E. (2002, January). MTEACH: A simple framework for didactic and context-based hypermedia. Proceedings of SSGRR 2002W, International Conference on Advances in Infrastructure for Electronic Business, Science and Education on the Internet.

Roisin, C. (1998, November). Authoring structured multimedia documents. SOFSEM'98--25th Conference on Current Trends in Theory and Practice of Informatics, 222-239.

SMPTE 12M (1999). Standard, Television, Audio and Film--Time and Control Code. [Online]. Available: SMPTE web site: http://www.smpte.org/

Synchronized Multimedia Integration Language (SMIL) 1.0 Specification, (1998). [Online]. Available: http://www.w3.org/AudioVideo/

Synchronized Multimedia Integration Language (SMIL 2.0) (2001, August). [Online]. Available: http://www.w3.org/AudioVideo/

Teixeira, J.C., Paris, C., Madeira, J., & Brisson Lopes, J. (1999, July). Courseware reuse or courseware to reuse? Proceedings of GVE '99 -- Computer Graphics and Visualization Education '99, 199-202.

Van Ossenbruggen, J., Hardman, L., & Rutledge, L. (2002, May). Hypermedia and semantic Web: A research agenda. Journal of Digital Information, 3(1).

PIER LUCA MONTESSORO, DAVIDE PIERATTONI, AND RICCARDO CICUTTINI

University of Udine

Italy

[e-mail]