Technology Approach and Detailed Plan Our research will be built around an emerging architecture for distributed systems that we call the “Pragmatic Object Web”. This notes the ongoing convergence of web and distributed object technologies to form what is usually called the object web. This is currently approached from four major points of view: CORBA (from an Industry Consortium) the Object Management Group, Java from Sun Microsystems, COM from Microsoft and a set of technologies from the web consortium W3C including XML and a document object model (DOM). These approaches are somewhat complementary but often competitive and in constructing real systems, we pragmatically propose to use the best of each approach – this assumes that some complex unpredictable worldwide process will blend these four giants into a composite distributed system architecture and technology base. Our pragmatic approach appears more likely than any other to lead to systems that are both powerful today and likely to be quite consistent with future changes. These ideas are described in a book that we are writing “Building Distributed Systems on the Pragmatic Object Web” which gives examples and detailed discussion of the concepts. The object web revolution has been driven as much by the adoption of open standards such as HTML, JDBC and IIOP as by the more obvious remarkable software artifacts and technologies such as browsers and the Java language. These object web standards and technologies appear to offer significant potential for improvements in universal access. In particular the object web standards allow the development of a more structured uniform information space wherein reusable universal access interfaces can be developed and used in a wide variety of circumstances. Although this potentially possible, it is by no means guaranteed as an unguided haphazard development of object web applications could lead to a situation actually worse than that now with increased information served by more and not less data structures. As a relevant example, XML technology could allow the definition of the structure of glossary items used to support more or less all education and training applications. We could then optimize universal access for this structure. On the other hand, it is also possible for each web site to develop and use a different XML syntax for their glossary and force the costly and inefficient scenario with a separate universal access mechanism in each case. This proposal aims to help and accelerate the development of common information structures that can both express the application in a general fashion and support well cross disability interfaces. In this fashion, our project will help the development of both universal access and the ongoing activities defining key object web standards. The Trace center is already a participant in the key W3C object model discussions. Our results will have broad applicability to general information spaces but we will concentrate on “distributed educational objects” (DEO) as these illustrate general principles and allow focussed technology, outreach and testbed activities. We now discuss the technical approach in more detail. We adopt a conventional hybrid information object model and define a DEO with a tuple (Page_URL, Component_DOM). This views information as a collection of components (labeled by Component_DOM) arranged in pages labeled by Page_URL. Pages are accessed through web address, file location, CORBA or Java naming service or whatever hierarchical naming scheme evolves on the object web. A “Page” is for traditional education, the basic curriculum unit. It is a “screenfull” or “foil” which is discussed by the lecturer or studied by the student as a single unit with cross referencing between concepts not requiring tiresome browsing and reloading of the browser page. A hierarchical labeling of Page_URL seems quite natural for future web education and training with some name like university/college/department/program/course/lecture. However the information within a given page is much less structured and consists of some often-haphazard arrangement of multimedia information nuggets. In this proposal, we will focus on universal access issues connected to the control and display of the complex information within a page, which is described by Component_DOM. We will start with activities built around the existing proposed W3C DOM. However the latter clearly can be improved for universal access and in the last two years of this project, we intend to design and test more advanced document object models. The proposal activities can be divided into five broad areas: 1) UA Architectures and Systems Design 2) UA Object Web Technology Implementation 3) UA Hardware Interfaces 4) UA Curriculum Development 5) UA Testbed and evaluation Here we discuss the first two of these areas. Both the hardware and software infrastructure of the object web is changing with remarkable speed and so our plans are necessarily tentative especially in out years. However we believe that the activities discussed below illustrate clearly our approach and in some sense represent a lower bound to our goals for they do not require any major new object web base technology developments. Of course, we will take advantage of any significant new relevant technologies that become available during the performance period and modify our plans accordingly. We intend two major technology thrusts. First investigate UA in terms of the existing W3C DOM with educational objects of varying degrees of sophistication and different implementation strategies. Secondly we will investigate a novel approach to a document object models, which appears to allow a more powerful approach to UA. We assume in all activities, the basic architecture sketched below: TangoInteractive manages the sharing of educational objects and allows each client to optimize its view of the DEO based on user preferences and capabilities of the client machine and network connection. This capability is available in any system using a shared event collaboration model, which allows separation of display and shared object specification. As a simple example, a client with a low bandwidth network connection would request the low resolution version of an image and one serving a user with impaired vision, the audio augmentation of this image. As shown above, we encapsulate this optimized choice of DEO component display in terms of a knowledge agent. Collaborative systems like TangoInteractive, can be used to share DEO’s between different users or between different display devices for a given user. This replication of object between different display modalities can be implemented within a single machine or between multiple machines serving a single user. Note that although it may seem extravagant, using multiple machines for a given user is quite practical given the rapidly decreasing hardware prices. In fact, we regularly use this strategy when teaching, so that one puts the key functionalities of audio/video conferencing, chosen curriculum page and chat/white boards on different machines assigned to the teacher. Students in this example typically view the curriculum on their own lab machine while a single machine handles audio and video for collocated students. We note that our model for instruction includes both asynchronous and synchronous modes supported in a common fashion for UA. We assume that in each case, students and teachers access curriculum material stored as DEO’s on web servers, object brokers or equivalent. Asynchronous or self-paced learning occurs when each participant accesses this material in his or her own time. Synchronous learning occurs when this same material is replicated among a class and discussed interactively. This model allows a single approach to universal access, which is independent of learning model. In our first thrust, we use the existing W3C DOM as supported by Netscape and Microsoft version 4 and higher browsers. We will also make extensive use of XML, which is here viewed as the most powerful way available to express the structure and internal linkage of document components. XML will be used both on the client and server side. We will look at Universal Access for the following types of educational pages which show increasing sophistication in terms of authoring tools and hence internal W3C DOM structure. 1) Plain HTML Pages 2) PowerPoint exported to the web using Microsoft’s Internet Assistant and modest restructuring (with an existing filter) to better define object components (Component_DOM). 3) PowerPoint accessed via COM components, which allows to properly define base object model. Existing NPAC technologies allows one to export this to the web using XML templates. This is an example of a modestly sophisticated object structure defined in a clean way via XML rather than the heuristic choices that need to made in interpreting the HTML tags in the first two cases. 4) We can embroider the object structure in page types 1) through 3) in various ways, such as through the addition of glossaries, notes and quizzes in fashions popularized by tools like WebCT. If the results of the first studies suggest it would be valuable, we will consider the effect of such additional components which can be added using dynamic HTML (DHTML) so that they appear indeed as direct extensions of the DEO Component_DOM rather than as additional windows. 5) In the last phase of this thrust, we will look at educational material produced with high end authoring tools. Macromedia Authorware is quite popular and typical of the systems we have in mind. The state of the art will probably change over the next two years but we expect there to several good systems capable of authoring the rich dynamic pages with essential multimedia flavor that are characteristic of Macromedia pages. One can also expect embedded interactive Java applets to be of growing importance. We intend to take state of the art pages of this type and explore the ramifications of this richer DOM for universal access. We emphasize that the basic linkage of TangoInteractive to pages of this type will be available through work funded at NPAC by other sources. Thus this proposal focuses on studying the UA issues for these different document object structures. TangoInteractive is particularly well suited for this study as it has a native JavaScript interface, which can access the full W3C DOM structure. For instance, we can identify the images contained in a document and so manage the rendering of these in the modality required by the user. TangoInteractive also captures all events in a page and so precisely shares all form input and output. Again this control of form fields and buttons, allows it manage alternative cross disability text and button input or output display devices on the different clients sharing the form. We can illustrate the difference between page types 2) and 3) above. In case 2), one must use a heuristic to distinguish the image which corresponds to the PowerPoint slide web export from those images which are buttons defining “home”, “next”, “previous” etc. In case 3), the XML structure defines exactly which image is which as XML defines the document structure. It also allows one to associate with image multi-resolution and sonified versions. So what’s wrong with this approach? Well many things no doubt but here we note a critical flaw in the current W3C DOM that it does indeed define reasonably the individual page components. However their relationship requires an understanding of the page layout and the dynamic structure of any DHTML which must be either done heuristically of by an approach such as 3) above which essentially captures the COM structure in XML. In general, good universal access requires both a definition of the individual components and their relationship (such as order of presentation) as well a clear definition of alternative forms needed in universal access. In our second technical thrust, we will investigate a new approach to document object models, which is designed to support both an easier definition of the overall structure of the document and the dynamic linkage of input-output devices to components. One approach that is attractive today is to use Sun Microsystems JavaSpaces and Jini technologies but these are of course only illustrative of appropriate technologies and better choices may become available. We suppose that all the components for a given lecture or self-paced study session form a JavaSpace together with all relevant devices. Note that we intend to use classic hierarchical naming and access schemes for coarse-grained views of educational objects such as those at the course and higher levels. However in any given session, it seems more appropriate to support a richer structure both inside each document and for the presentation material that one wishes to pick and choose between with an order that is determined in real-time. Jini offers its leasing concept to support dynamic component structure and applet download of interfaces to support dynamic device capabilities. In this model, the agent in the above figure, is a matching service between the curricula and device entries in the JavaSpace. As explained above, the agent will also effectively generate the dynamic “index” supporting navigation between components. Note that this architecture illustrates our Pragmatic Object Web with multiple object models coexisting. Java provides the content and display device registration and discovery. We expect to use XML to define the properties of the JavaSpace components and of course HTML to define final layout. The base educational objects can perhaps be served from a CORBA object broker and originate (as in PowerPoint) with a COM specification. We will build a prototype of such a rich object model linked to TangoInteractive. This will be in last half of the project after we have further experience from using the existing W3C DOM. We expect this JavaSpace based model to give considerable insight to future designs of object models with richer navigation models, definition of document components and their dynamic linkage as well as their interface to input-output devices.