Copyright ©2004 W3C ® ( MIT , ERCIM , Keio), All Rights Reserved. W3C liability, trademark, document use and software licensing rules apply.
The document provides a summary of several techniques and best practices that Web site authors and solution providers may employ when creating and delivering content to a diverse set of access mechanisms.
This section describes the status of this document at the time of its publication. Other documents may supersede this document. A list of current W3C publications and the latest revision of this technical report can be found in the W3C technical reports index at http://www.w3.org/TR/.
This document is a W3C Working Group Note. It represents the views of the W3C Device Independence Working Group at the time of publication. There are currently no plans to amend this document further. Publication as a Working Group Note does not imply endorsement by the W3C Membership. This is a draft document and may be updated, replaced or obsoleted by other documents at any time. It is inappropriate to cite this document as other than work in progress.
This document is published as part of the W3C Device Independence Activity by the Device Independence Working Group. It is a deliverable as defined in the Charter of that group.
Comments on this document may be sent to the public www-di@w3.org mailing list (archived at http://lists.w3.org/Archives/Public/www-di/).
Patent disclosures relevant to this document may be found on the WG patent disclosure page.
Drawing on the experience of technology providers and Web authors, this Note presents an overview of many existing techniques and best practices that may be used to deliver tailored content and applications to a wide variety of devices. It shows, in practical terms, how one might address the problems of content creation, content maintenance and content adaptation. It explores the issues associated with managing an application's interaction with a user where different devices and modalities are present.
This Note is restricted to content and applications that are intended for Web delivery, employing technologies that are associated with the Web, and in particular those technologies that have been recommended by the W3C or are being developed by the W3C. Furthermore, while the Web can be used to deliver a wide variety of media types, this Note shall focus on media that is represented by, or referenced through, markup languages. XHTML and its associated technologies shall play a key role in this Note, but the legacy of older technologies shall be recognized as an important domain of interest.
Implementation-specific issues shall be recognized, but not elaborated, in this Note. This includes issues such as performance and scaling, security, resource consumption etc. It is recognized that different implementations of the same authoring techniques, and their associated adaptation techniques, will have different properties and that these differences provide scope for business opportunities. These differences are not a concern of this Note.
This Note includes discussions on the features of authoring tools, site creation and maintenance tools, storage, delivery, adaptation, end-user devices and software on such devices.
This is a rapidly changing area and as a consequence some of the techniques described will be superseded or obsoleted by new developments as well as the emergence of new techniques made possible by new developments. Readers are encouraged to seek additional information through the references and other sources.
This Note is one of the deliverables of the Device Independence Working Group. According to the charter, section 3.2.2, the purpose of this document is to:
"propose some techniques for authors to achieve greater device independence. [...]
Firstly, using existing techniques, to:
Secondly, using techniques based on XForms, to:
The quoted charter items lead to stipulation of the following goals for this document:
DIAT:G-1: Authoring Practice
Identify the current and proposed techniques to support authoring for multiple devices, and present these as abstractions of the process from the original authoring step up to and including the final delivery of content to the access mechanism.
DIAT:G-2: Adaptation Techniques
Identify the range of content adaptation techniques and how they may be applied at the different stages of the delivery process (from server through intermediaries to the device).
DIAT:G-3: Requirements
Identify the requirements that should be satisfied by implementations of techniques identified in this document. In particular, this relates to the technical requirements for markup languages and other technologies offered by the W3C.
DIAT:G-4: XForms
Identify techniques, based on XForms, to support device-independent navigation and interaction. These techniques must support the tailoring of presentation, interaction modes and navigation within form-based Web applications.
This document shall identify known (published) practices and techniques that address the following issues:
The focus shall be on open and accessible standards and "best" practices, particularly those advocated by W3C Recommendations and by respected organizations.
This Note is intended as background material for people interested in the techniques associated with delivering content and applications from web sites to devices with very different capabilities. It assumes that the reader is familiar with the Authoring Challenges document [AC] that preceded this document.
In particular, the audience for this Note includes:
The objective of authoring is to create content for an audience. The objective of content delivery is to convey the authored content to the audience. In general, the objective of Device Independent Authoring Techniques is to facilitate content delivery to as wide an audience as possible while:
Each technique has technical and non-technical motivations, described below. Technical motivations refer to technical issues raised by increasing variety of devices and device features. Non-technical motivations refer to issues raised by people (authors and end users), the way they create and consume content, the business processes and other human activities that form the context of the Web.
Authoring Techniques are ways in which various technologies can be utilized to minimize the number and types of materials that need to be created to satisfy a range of device types. The techniques also provide mechanisms for incremental additions to the set of materials created to support a Web Page to enable it to be better adapted to a device (or class of devices). Most of these techniques involve the provision of alternatives and a mechanism to choose between these alternatives at run time. These techniques must:
Ultimately, the most important feature of an authoring technique is its ability to produce a good rendering on the client of the original content created by the author. This is a subjective assessment and is beyond the scope of this document.
In a strict sense, a Device Independent Authoring Technique is a method of creating content that can be delivered to any conceivable device. In practice this is impossible, so the real techniques seek to minimize the number of materials needed to satisfy a range of device types. There are many different types of material that are needed to satisfy a user request for a Web Page. These include various types of media that constitute the content of the page as well as supporting material such as styling and layout rules that determine how the content is rendered.
The creation of these materials is what we call Authoring and typically these materials must be adapted for delivery to specific devices. The construction of adaptation processes may also be regarded as a form of authoring, though more usefully regarded as a form of programming. Authoring of raw material, styles, layouts and adaptation processes may be performed by different individuals with different skills, which support specialization of the various contributors. The technical challenge is to facilitate this separation of concerns.
One of the identified Authoring Challenges is Variability, by which is meant the ability of an authoring technique to vary the content according to the delivery context. Technically, this requires access to parameters defining the delivery context and a means of content selection according to these parameters. Some authoring techniques seek to expose as many parameters as possible, accompanied by powerful selection methods. Other techniques seek to abstract these parameters while making the selection process as implicit as possible. The former approach gives the author more control but requires more effort, while the latter sacrifices control to reduce effort.
A technique that is suitable for a small amount of content should also be suitable for a large amount. Ideally, the amount of additional effort should scale with the amount of content.
The non-technical motivation for Device Independent Authoring Techniques is centered around the requirements and constraints imposed by the creators of content and the audience for this content. In simple terms: time is precious, effort is expensive, users are demanding and consistency is desirable.
Creating content is an expensive and time-consuming activity. It is expensive because it consumes the time of the authors. It is generally accepted that additional effort by authors is required for Device Independence, and the various authoring techniques attempt to minimize this additional effort. There are other costs, such as the investment in adaptation mechanisms, but these tend to be less significant.
Authors intend their content to be accessed by as large an audience as possible, regardless of the constraints of the delivery context. The majority of these constraints come from the limitations of the end user devices, though similar constraints can be imposed by the end users themselves. An authoring technique should overcome the constraints so that the greatest number of delivery contexts can be supported.
Branding is a business motivation that seeks to present a consistent image of the business across the spectrum of delivery contexts. Consistency also reduces end-user confusion, thereby giving the end-user more freedom to migrate from one device to another as circumstances may dictate. A Device Independent Authoring Technique should support the consistency of the end-user experience.
Section 2 describes a classification of Authoring Techniques that shall be referred to in subsequent sections. Section 3 describes the Delivery Context, whose variability is the prime motivation for Device Independent Authoring, as explained in Section 4. Section 5 introduces some general authoring techniques whose application spans numerous classification, and explains that most Device Independent Authoring Techniques aim to support variability in six aspects: Style, Layout, Content, Structure, Navigation and Interaction. Techniques specific to these aims are covered in Sections 5 through 10. Section 11 summarizes the Authoring Techniques Workshop, the event that marked the beginning of this document. Section 12 draws conclusions from the document.
The focus of this document is on Device Independent Authoring, but this is inevitably associated with Adaptation Techniques, details of which are presented in Appendix B. The adaptation of content may also necessitate adaptation of structure, so a brief summary of content-related structures is presented in Appendix C. The appendices may be read independently of the rest of the document.
The Device Independence Working Group has defined a number of terms related to Device Independence. This Note adopts these terms as presented in the Glossary of Terms [DIWG Glossary].
Three broad classifications of authoring techniques are identified in this section. In all classifications, the following features are assumed:
The Authoring Challenges [AC] document highlighted a number of (overlapping) challenges facing content authors. The document identified 55 implications [DIAI] of Device Independence within 10 categories (listed in AC Section 8.1), reproduced below.
Some, but not all, of these features are provided by current technology. It will be obvious to readers of this document that there are gaps between the identified challenges of Device Independence and the available authoring techniques. Closing these gaps is a prime motivation of the DIWG.
The following subsections introduce the three main classifications of authoring techniques. General descriptions are followed by concrete examples (including markup where appropriate). All of these authoring techniques rely on various adaptation techniques, discussed in Appendix B.
In the Multiple Authoring classification, the author creates a different version of the content for each (class of) device. Devices that have not been specifically addressed by the author may result in lack of service for such devices, in contravention of DIP-3, which states "It should be possible to provide a functional presentation, in response to a request for a Web page identifier, in any given delivery context that has an adequate access mechanism." To avoid such a contravention, a delivery solution may select an available authored version that is functionally compatible with the context, but there is no guarantee of such availability.
There are certain cases where Multiple Authoring is preferred. The production of multimedia resources may require additional author input, such as the creation of corporate branding resources (e.g. logos) at various sizes and resolutions. This is a case where "pixel perfect" control is an essential feature for the author with sound business reasons for avoiding less precise methods.
The ability to select from a set of possible versions of content can be supported by techniques such as URL Redirection (by server-side adaptation) or Server Selection (by an intermediate adaptation such as a Proxy). Client-side selection-based adaptation is also supported by various technologies.
In the Single Authoring classification, most of the author's effort is focussed on creating a single version of the content. An adaptation solution translates the single authored content into a form appropriate to the device. The author may have the requirement/option to provide additional information to assist the adaptation solution. Typically the effort involved in creating the single version of content is greater than any individual version created by a Multiple Authoring technique. However, this effort is expended only once and therefore Single Authoring techniques involve less total effort.
In some Single Authoring techniques, the author may be required to create one or more resources for each delivery context. The resources may take the form of styles, scripts, configurations etc. However, these resources can be re-used and are therefore considered to be a once-off effort. These resources are also "Single Authored".
Some Single Authoring techniques are described below. Corresponding adaptation techniques are described in Appendix B.
Hinting is the means by which an author conveys meta-data related to parts of the authored content. For example, an author may express the relative importance of parts of the content through hints. Hints may be represented within the authored content as additional markup, additional attributes, specially formatted comments etc. These can be added to familiar Web markup (e.g. XHTML) via an authoring-specific markup module. It is also possible to represent hints outside of the document if parts of the authored content can be addressed individually (e.g. using XPath).
Where custom modules or extensions to a markup language are used in the authoring solution, it is expected that these extensions are not visible beyond the authoring/adaptation environment.
In CSS Media Queries, the author defines rules based on contextual information. These rules are typically captured as a class and XHTML tags in a document may be assigned a class via the class attribute. The rules then affect the presentation of content associated with the class. These rules may be executed by the client, by an intermediate or by the origin server.
CSS Media Queries require that the author is familiar with specific device features. Abstract characteristics are not generally supported. The expression language cannot (currently) be extended. Future versions of CSS Media Queries may address these shortcomings.
The following sample (from http://www.w3.org/TR/2002/CR-css3-mediaqueries-20020708/#color-index) illustrates the use of Media Queries to indicate that a particular stylesheet is appropriate to delivery contexts involving displays with at least 256 colors:
<?xml-stylesheet media="all and (min-color-index: 256)" href="http://www.example.com/..." ?> |
Adaptation techniques associated with the authoring techniques in this section include the following: (More details may be found in Appendix B.)
In the Flexible Authoring classification, the author has complete freedom to combine Single and Multiple Authoring techniques. Thus the author may create single versions of some resource(s) for subsequent adaptation, and multiple versions of other documents when fine control or specific features are required. This flexibility may also be applied within a document where part of the document uses Single Authoring and the rest uses Multiple Authoring.
In a layout-based authoring technique, the author provides separate pieces of content. Some or all of these pieces are subsequently aggregated into a complete presentation using a layout as determined by the context. Portal systems are examples of such an approach, where each portlet is an individually authored piece of content that is aggregated (and possibly adapted) into a context-appropriate layout. A detailed discussion of layout authoring techniques is presented in Section 8.
The alternative content authoring technique is a form of Flexible Authoring applied within a document. It permits the author to express a set of alternative content fragments/resources and a means of selection within the set. An ordering may be imposed on the set to influence the selection mechanism. Client-side methods have been available for some time (viz. <frames> and <noframes>) but incomplete support in clients makes this useful only in limited cases. Current technologies provide more sophisticated solutions, such as the switch mechanism in SMIL as in section 7.2.1.
In addition to the adaptation techniques associated with single authoring, the following adaptation techniques are prevalent among flexible authoring solutions: (More details may be found in Appendix B.)
The delivery context [described in DCO] is the set of all (available) parameters pertaining to the characteristics of the delivery channel, which includes the server, the communication media, intermediate active components, the edge device and the rendering mechanism. It is expected that a subset of the delivery context will influence what is delivered to the client and how it is delivered. In solutions that provide device independence, it is typically the case that the delivery context influences the content adaptation processes. The content author will optionally provide additional information to affect this adaptation within particular contexts.
In practice, several technologies permit the author to insert rules into the content whose execution is influenced by certain core device characteristics, including the selection of layouts. This means that the author is made aware of at least some of the delivery context parameters, and consequently the success of the authoring process may depend on the author's correct understanding of these parameters.
The more parameters that are offered to the author, the more confusing the authoring process can become. For this reason, context information can be aggregated and/or classified so that the author may apply decisions to greater ranges of possible contexts. As an example, consider an adaptation solution that includes the orientation of the screen (portrait, square or landscape) as a parameter, rather than offering the exact aspect ratio. Alternatively, the author may be provided with a library of methods to simplify the task. For example, consider a method called "isPortrait" that uses the specific aspect ratio to decide if the display is portrait.
CSS Media Queries are a good illustration of author awareness of context. Only a small number of attributes are present, and the operations on these attributes are limited. Nevertheless, this is sufficient for the author to control styling for various classes of device. Fine control of styling is not possible because of the limited set of attributes and limited operations. For this the author must resort to more complex techniques with a more complete set of presentation attributes. To this end, the DIWG is working to create a core set of presentation characteristics [CPCReq].
Context is used in all adaptation techniques since the objective of adaptation is to produce a result that is appropriate to the context. In some techniques (e.g. Selection) the author is aware of the context and supplies information related to the context (e.g. decision expressions). In other techniques (e.g. Decomposition) the relevant contextual information only plays a role after authoring.
There are two general techniques for Device Independent Authoring. They are, in accordance with the classification introduced in Section 2, as follows:
These, however, are extremes in a spectrum of techniques. MA/CS is an extreme because it places a significant burden on the author, and SA/CA is an extreme because it places a significant burden on the adaptation. In simple and constrained circumstances, these techniques may be viable, but generally they are inflexible, will not scale and are onerous to maintain.
Compromises between the two extremes can produce useful techniques, which will be explored in detail in subsequent sections.
To facilitate compromise, the objectives of authoring can be divided into several aspects, and different authoring techniques can be applied to the different aspects. The following (overlapping) authoring aspects are generally recognized:
Each of these aspects can be addressed by different authoring techniques if the aspect can be sufficiently detached from the others. For example, Style can be separated through the use of Cascading Style Sheets (CSS), which affords the author the opportunity to use multiple styles for the same raw content.
Sections 5 through 10 will explore different authoring techniques that enable the author to vary different authoring aspects according to the delivery context. These techniques would normally be used in combination to achieve maximum benefit and should not be considered as solutions in isolation. Many of these techniques rely on specific adaptations, which are subsequently documented in Appendix B.
Changes in style will change the presentation of resources without actually changing the resource itself. Sometimes the changes are considerable, such as making the content invisible. Sometimes the changes are more subtle, such as indenting by a few pixels. Since these changes predominantly affect the appearance of content, the appropriate authoring tools support WYSIWYG editing and/or device emulation. It is then the author's task to determine which styles suit the various delivery contexts.
The delivery context can influence the selection of styles. In particular, this feature is provided in CSS Media Queries. The author creates (or acquires) different style sheets appropriate to the range of style-capable devices to be supported and then determines the conditions under which each style will be used. These conditions are then expressed using CSS Media Queries. The client first retrieves the resource and then retrieves the appropriate style in accordance with the conditions.
The styles associated with elements in delivered content may be accessible (via a DOM) to scripting on the client. So, for example, the author may include script in the content that responds to events and changes the styles accordingly. This technique is used to provide content that becomes highlighted when the user points to certain regions (e.g. mouseover events) and regions of the content that "collapse" by changing the style to invisible.
In some cases, aspects of the delivery context can be determined by client-side scripts, which enables them to make adjustments to the styles applied to the delivered content. For example, the font size could be reduced when the content is rendered in a small window.
Client-side scripting tends to be the most device-sensitive technology and consequently non-portable. Therefore the author must resort to creating different scripts for the different devices, and sometimes no scripted features at all when the behavior is not supported by the device. Scripted style cannot be considered a viable device independent style authoring technique.
In cases where the client does not support styles, it is still possible for the author to use styles in the original content. An adaptation process in the origin server (or intermediate) would replace the styling with explicit presentation features, such as <font> tags. The WYSIWYG aspect of authoring in these circumstances needs to incorporate the adaptation process in order to determine the effects of the adaptation.
The context-sensitive capability of technologies such as Media Queries can be supported in server-side styling with the aid of context carrying technologies (e.g. CC/PP and UAProf). Scripted style selection is generally not possible, since this is a client-side function.
XForms provides an "appearance" attribute that can be specified on all Form Controls. This provides an author with the ability to provide a hint to the component responsible for rendering XForms Form Controls. There are three pre-specified values: full; compact; minimal; and an ability for an author to define their own value but with no pre-specified meaning.
The attribute does not specify what exactly is to be rendered but is merely a hint. The XForms <select1> control is a good example to consider. The following are an explanation of how the hints should be interpreted for this control:
To enable support for the wide variation in the output capability of different delivery contexts, it is necessary to be able to provide a variety of different layouts. Variations in the size and aspect ratio of displays, for example, can mean that a different physical layout is required to support a harmonized user experience.
The DIWG Authoring Challenges specify the following requirement for layout.
DIAC-4.3: Layout: Authoring techniques that support DI should provide mechanisms that allow authors to express the layout of material that varies between different devices with different delivery contexts. In particular, they should support different spatial and temporal layout of material.
Although it is possible to arrange for explicit representations of layout using commonly available technologies, such as XHTML and CSS, it is unusual to find such an approach being used in current web sites. This section covers the basic principles associated with explicit provision of layout, and examines a number of current techniques.
The basic principles of using explicit layouts are the same as those for any properties that may vary between delivery contexts. First, there must be an explicit representation of the properties, in this case the layouts. Second there must be a method by which the layout can be referenced in a manner that is independent of the delivery context.
The kind of information needed for an explicit layout is information that will allow the basic differences between delivery contexts to be taken into account. For example, one of the major differences between delivery contexts for display devices is the number of pixels that can be used. Another is the aspect ratio.
Layout representations typically divide the output into a number of areas in which content can be placed. An individual area might contain a single piece of content, such as an image, or a large amount of content, such as several paragraphs of text. These areas are not limited to representing spatial display. Some systems allow them to represent temporal information allowing control over spoken output, for example.
To make use of an explicit layout representation there must be a way to assign content to particular areas of the layout. Current techniques make use of direct references from the markup that defines the content to the layout representation. These references use names. Areas in the layout have some unique name or ID. The content to be assigned to a particular area references it using that name or ID. Where multiple pieces of content reference the same area, simple rules define the ordering of output within the area. Usually this is based on the order in which the references are encountered within the content.
CSS is able to provide the basic functions referred to in DI Principles. For example, it is possible to associate size and position information with, for example, <div> elements, with particular IDs, in a style sheet. By using <div> elements with those IDs when authoring content, material can be targetted to a particular place in the layout.
The ability to associate subsets of a style sheet with particular properties associated with a device makes it possible to have several versions of the <div> styles in a single sheet and to select the one most appropriate.
Current capabilities for selecting different styles from within a single CSS are rather limited for general use in device independence. However, systems with specialist adaptation mechanisms, which can select between different versions of entire style sheets, exist and could be employed to give greater control. Some commercial systems are capable of this kind of adaptation of style sheets.
SMIL 2.0 includes an explicit layout notion based on its <layout>, <root-layout> and <region> elements. SMIL contains a number of layout modules for specific types of rendering, including multi-window visual display and audio display. As such, it is a good example of the ability of layouts to represent both spatial and temporal information.
The following example illustrates how two pieces of text can be positioned using SMIL:
<smil xmlns="http://www.w3.org/2001/SMIL20/"> <head> <layout> <root-layout width="320" height="480" /> <region id="a" top="5" bottom="100" /> <region id="b" top="200" bottom="280" /> </layout> </head> <body> <text region="a" src="text.html"/> <text region="b" src="additional_text.html"/> </body> </smil> |
The layout defines two regions within a root-layout. The size of the root-layout determines the physical size of the displayed material, in this case 320 pixels in width and 480 pixels in height. Within this root-layout two regions are defined. Both are the full width of the root-layout. The first starts 5 pixels below the start of the root-layout. The second starts 200 pixels below the start of the root-layout.
The <body> element within the example defines two <text> elements. Each of these assigns text to one of the regions in the layout.
There is a clear logical separation here between the information in the layout and the way content is mapped to it. If the layout is modified, the final presentation can be varied without the content being changed. Similarly the content can be changed without affecting the layout. Only the id of the <region> element links the content with the layout.
Currently, it is possible for an implementation of SMIL to have its own mechanism for allowing different versions of a layout to be associated with some context attributes, and a mechanism to support custom attributes, thereby providing the means to use delivery context data.
In conjunction with a server side adaptation that, for example, constructs the SMIL markup from definitions of content and an appropriate layout, this markup shows nearly all of the characteristics needed for support of multiple delivery contexts.
Implementation details of several layout techniques are outlined in an accompanying Submission document [ATSub].
An author will wish to have control over the order in which parts of the content are delivered, typically to reflect dependencies within the content. Structures with a natural order can define a traversal of the content. Typically these structures will have a well-defined origin, which is important to identify the entry point. Of less importance is an exit point in the structure, since most usage patterns do not require a specific exit procedure.
The use of links to the next, and previous, content relative to the most recently delivered content can facilitate navigation of linear and hierarchical content structures. Several Web site authoring tools will automate the creation of these navigation aids. Many browsers maintain a navigation history that is accessible to client-side script, with which the author may provide a "Previous/Back" linking feature.
In more complex situations (e.g. e-commerce Web sites) the determination of the "Next" page is done by an application using session information. The author would employ a workflow approach to enable the application determine the order of delivery. Such techniques are beyond the scope of this document.
Selection techniques solve the problem of choosing from the available fragments of content to create a presentation for the user. Selections are typically based on matching certain characteristics of the delivery context to features of the content fragments. Below are examples of established selection techniques and the technologies that employ them.
The Synchronized Multimedia Integration Language (version 2.0) [SMIL 2.0] allows authors to create interactive multimedia presentations. Authors can describe the temporal behavior of a multimedia presentation, associate hyperlinks with media objects and describe the layout of the presentation on a screen.
In SMIL, the <switch> element enables an author to specify a list of alternative elements, selected according to Boolean tests. The first acceptable element in the list is chosen. A default selection can be defined as the last element in the set by having no constraints, as illustrated below.
<smil:switch> <html:img src="bigColorImages/logo.gif" title="Company logo" alt="logo" smil:systemScreenSize="768X1024" /> <html:img src="smallMonoImages/logo.gif" title="Company logo" alt="logo" smil:systemScreenSize="160X160" /> <html:img src="defaultImages/logo.gif" title="Company logo" alt="logo" /> </smil:switch> |
SMIL test attributes are permitted outside <switch> elements, though this usage does not permit selection from a set of alternatives, nor does it support the concept of defaults.
CSS Media Queries [Media Queries] is a W3C Working Draft to enhance the @media rules of CSS and the "media" attribute in HTML. Using a Media Queries expression, it is possible to select a stylesheet based on context properties. The recognized set of properties covers a small range that includes size, resolution, type of device and color capability. The following example illustrates the selection of a style sheet based on the device type:
<link rel="stylesheet" type="text/css" media="handheld" href="portabledevice.css" > |
Here is a another expression as it would be represented in CSS:
@media screen and (max-device-width: 300px) { @import url(narrowwindow.css) } |
Media Queries avoids the escaped versions of characters "<", ">" and "&" in expressions by using "min-" and "max-" prefixes. It also uses "and" for conjunctions, and "," for disjunctions. This makes the expression syntax easier to read, which is important if the expressions are to be hand-crafted and later read by humans (possibly for maintenance purposes). With the increasing use of authoring tools, this "legibility" feature of the expression language may become less important.
It is generally assumed that the Media Queries expressions will be processed by the client. This does not preclude the processing of Media Queries (and CSS in general) at some other point in the delivery path, such as in an intermediate system or in the origin server itself.
Unfortunately, Media Queries is limited and inextensible in its current form and is therefore suitable only for simple multi-device solutions.
A layout is a structure containing references to content fragments. Selection of fragments based on layouts usually involves two steps. The first step is to select the appropriate layout, based on context information. The second step is to populate the layout with fragments of content. This can be achieved by (indirect) references to fragments in the layout itself, or by a separate process that maps content fragments to portions of the layout. The process is similar to the well-established mail-merge operations of popular office automation solutions.
In incremental refinement, the set of possible subsequent "pages" of content is reduced each time the user follows a step within the navigation structure. After a number of navigation steps, the user has focussed on a specific subset of the content and no further refinement of the selection is necessary or possible. Informally, this may be viewed as a Zoom in/out feature.
Aggregation is the process of bringing together Authored Units from one or more sources to form a single fragment. Aggregation may be executed at the time of authoring, during adaption, or by the client. An authoring solution that supports aggregation must provide the author with the means of expressing which content fragments will be aggregated, and/or of expressing the process by which the content fragments will be selected for aggregation.
Since individual content fragments contain structure of their own, the aggregation of such fragments will inevitably produce a new structure. Any factors that influence aggregation may therefore be used to vary the structure of content, especially if these factors are determined by the delivery context.
Implementations of the aggregation technique include:
Decomposition is the act of dividing up one or more authored units to create a set of perceivable units appropriate for a particular delivery context. An authoring solution that supports decomposition must provide the author with the means of expressing how the authored units will be divided, and may necessitate a means of referencing the individual perceivable units thereby produced.
Decomposed content has a significantly different structure to the original content from which it was derived. This has implications for the manner in which the content is navigated, and possibly for linking to or within the content.
One form of decomposition commonly known as "pagination" occurs when the original content is linear in structure and is divided into an ordered sequence according to a simple quantitative rule. The ordered sequence permits the end user to select any page at random, though the preferred selection is indicated by the order of the sequence. Examples of pagination include:
Each "page" of a pagination could be referred to by its sequence number (e.g. Page 5 of 12) but the author cannot know these sequence numbers in advance. Thus it would be inappropriate for the author to say "see Page 5" because in different pagination contexts the same content may be rendered on a page other than the 5th.
A solution is to permit the author to make references such as "as mentioned X", (X being a named anchor) which may subsequently be rendered as "as mentioned in Page 5" in pagination contexts, or perhaps "as mentioned in Section 4.2" in a section-based decomposition context or "as mentioned here" where the "here" is rendered as a hyperlink in non-decomposition contexts. Ensuring the link is compatible with the surrounding text can be achieved through simple heuristics (e.g. deciding when to prefix the reference with the word "in"). The use of a phrase such as "click here" would avoid this issue but would not be in keeping with accepted Web accessibility guidelines.
Several pagination solutions are available to authors, including the many technologies from the Desktop Publishing and Word Processing domains, and markup technologies such as XSL Formatting Objects (XSL-FO) [XSL chapter 6]. In the case of XSL-FO, a "page-master" element describes the shape and orientation of the page(s), and a "page-sequence" element contains the content that flows into the pages defined by the page-masters. A "page-sequence-master" element determines the order in which page-masters are chosen. A "simple-page-master" element can be used on its own to produce a sequence of pages from a flow of content such that each page has similar dimensions. This is the most simple approach to pagination, and the discussion below will assume this simple approach of using equally sized pages.
Decomposition is an unavoidable necessity when the client is incapable of rendering the entire content. Knowing that decomposition is likely, authoring techniques exist that give the author the ability to influence or control the decomposition. Among these techniques are:
This technique is common in word-processing solutions. The author inserts a mark, typically after a paragraph of text, to indicate that this is an appropriate place to divide the text. The insertion of such a mark, and its position, can be made conditional on contextual information (e.g. page length). Fixed and optional page breaks can be supported by letting an adaptation system interpret existing separator elements such as <hr>. Alternatively, page breaks (such as the conditional "before start of new chapter" break) may be indicated explicitly by the author (e.g. XSL-FO). |
This is another word-processing technique in which a style is applied to the text such that paragraphs are divided so as to avoid leaving a single line on a page, separate from the rest of the paragraph. A "widow" is a division of N lines into 1 line on one page, followed by N-1 lines on the next page. An "orphan" occurs when the split is N-1 lines on the first page and 1 line on the next page. Avoiding these is possible when there are more than 3 lines in the paragraph, though it may result in a page not being completely filled with text. |
Another word-processing style that requires that paragraphs are not divided, and are not separated from their subsequent paragraphs. "Keep Together" is applied to paragraphs. "Keep With Next" is typically applied to headings. An adaptation system could use custom styles, as per CSS, to support this pagination feature. This feature is present in XSL-FO in the form of the "keep-with-next" and "keep-with-previous" properties of the break element. |
The author uses headings to identify sections within the content. Decomposition can then be applied using the headings as division points. The <hn> tags in HTML are typical candidates for decomposition via sectioning. The (sub)section titles also provide useful text for links to each section, for example in a Table Of Contents. Unfortunately, this approach does not fit with XML or DOM approaches since the identified sections do not correspond to any identifiable sub-tree within the document markup hierarchy. XHTML 2.0 explicitly supports this concept through its <section> element, and this will identify document sub-trees. |
A region is a subset of the content identified by a starting mark and an ending mark. In XHTML, the <span> and <div> tags can be used to identify regions. Decomposition involves extracting regions from the content. The "id" attribute can uniquely identify fragments within a document. Since regions may be nested, this approach permits varying sizes of region to be selected, according to the rendering limits of the client. |
Adaptation techniques associated with the authoring techniques in this section include the following: (More details may be found in Appendix B.)
Navigation is an essential part of any content delivered to the end user. It enables the user to consume the content in the order intended by the author, and to locate parts of the content whenever necessary.
Navigation may be created explicitly by the author, generated under the influence of the author, or generated completely automatically. Typically, navigation links within raw content (e.g. paragraph of text) are placed there by the author. Menus and "site-wide" navigation features can be automated because the links are usually determined by the structure of the site and not by the content of the individual resources. Nevertheless, the author may wish to exert some influence over how much navigation is generated (e.g. nearest-neighbor linking, or site-wide random navigation) and how it is presented (e.g. static list of hyperlinks, or a client-side interactive mechanism).
Links provide the user with the means of moving from one presentation unit to another, or to positions within these presentation units. The hyperlink is the most common form of link, typically rendered as a highlighted part of the content that responds to an event (such as a mouse click) by navigating to the target of the link.
Adaptation of the original content to suit the delivery context can have considerable effects on linking. For example, links within the same (original) page may, through decomposition, become links between different pages.
The easiest, though least flexible, approach to navigation is to let the author create it. The following navigation features can be utilized by the author:
A URI should be interpreted as a reference to a resource. It is possible that this single resource will have multiple representations, according to the presentation capabilities of the access mechanism and other contextual factors. It is possible that the same URI will produce different content, though this is not the same as saying that the associated resource has changed.
For example, a web page URL (a specific form of URI) offering the current weather forecast will obviously offer varying content according to when you use the URI, but there is no doubt that the URI always refers to the same resource, as implied by "current weather forecast".
One can extend this example to say that the URI will offer different representations of resources, according to the device being used to access it.
There are several tools at the disposal of the author, some of which are described below.
The standard HTML linking mechanism is the anchor tag, <a href="url">...</a>, which has been a key feature of the Web since its inception. It identifies a region of the content that, when selected by the user, will cause the browser to load the content referenced by the href attribute. The simplicity of this mechanism ensures that creating webs of content is easy to achieve.
The original linking mechanism has its deficiencies. For example, the original link did not clearly indicate the relationship to the linked content, despite the availability of a 'rel' attribute. The lack of a formal definition for link relationship (versus the informal proposals that were offered) meant that the 'rel' attribute was almost never used. Links can only be in one direction. Links can reference named anchors in the target, but cannot reference arbitrary locations or regions within the target. There is no provision for error handling. The type of the referenced target cannot be identified. (This feature was subsequently introduced in HTML 4.) The limitations of HTML linking have led to alternative linking mechanisms, some of which are described below.
The XML Linking Language [XLink] allows elements to be inserted into XML documents in order to create and describe links between resources. It uses XML syntax to create structures that can describe links similar to the simple unidirectional hyperlinks of today's HTML, as well as more sophisticated links. It is a key component of many XML technologies.
The XML Pointer Language [XPointer] is the language to be used as the basis for a fragment identifier for any URI reference that locates a resource whose Internet media type is one of text/xml, application/xml, text/xml-external-parsed-entity, or application/xml-external-parsed-entity.
Note: The design described in previous versions of XPointer has been factored into a basic framework (http://www.w3.org/TR/xptr-framework/) which defines XPointer schemes and simple "barename" fragment identifiers, and three additional schemes: http://www.w3.org/TR/xptr-element/, for addressing elements by their position in the document tree, http://www.w3.org/TR/xptr-xmlns/, for binding namespace prefixes to namespace name and http://www.w3.org/TR/xptr-xpointer/, for full XPath-based addressing.
HLink provides XHTML Family Members with the ability to specify which attributes of elements represent hyperlinks, and how those hyperlinks should be traversed, and extends XLink use to a wider class of languages than those restricted to the syntactic style allowed by XLink.
SMIL supports interaction concepts that go beyond those offered by HTML. In particular, hyperlinking has been extended to support temporal anchors:
<smil:video src="/videos/sample" region="video" title="Sample presentation" alt="Just another sample video sequence" abstract="A 30 second two-part talking head sequence" > <smil:area id="woman" begin="0s" end="15s" title="Woman talking"/> <smil:area id="man" begin="15s" end="30s" title="Man talking"/> </smil:video> |
This feature provides a new dimension for authoring and adaptation. For example, temporal anchors open up possibilities for decomposition so that long verbal/visual presentations can be divided into smaller parts according to author-identified boundaries.
Timed links enable the author to present navigation options that are relevant to the presentation context. Since only a subset of all of the available links need to be accessible at any given moment, this feature reduces the need for screen space (or access keys) for links, which is an important consideration for small devices.
During the design of XForms a great deal of effort was devoted to making the User Interface part of XForms device independent. XForms User Interface Form Controls are abstract in nature and defer the actual appearance to a combination of the XForms Processor (which should depend on the device) and styling. As an example XForms defines a <select1> form control. The control is intended to specify that the user is to choose exactly one item from a set of alternatives. The selection of which widget (visual representation) to render is determined by the processor on the target device, possibly influenced by a style associated with the form control. Adaptation of form controls is illustrated in Appendix B.2.2.5
A Web Page may contain more than one form. The presence of one or more forms in a Web Page will not introduce any additional difficulties into the adaptation process of pagination unless the need arises to split a single form.
Forms are often complex entities that include relationships between the various parts of the form. For example, forms that are used for making travel arrangements often include a departure field and a return field. A constraint is normally specified that the departure must be earlier than return. In XForms this would be accomplished with a model item constraint determined by the author. With HTML forms this would have been implemented with some form of script, requiring the author to be a skilled programmer.
The two parts of an XForms form need to be considered whenever decomposing a Web Page that includes XForms. In the simple case where a form will not be split then both the <model> (which in XHTML is part of the <head>) and the form controls (which in XHTML is part of the <body>) need to be copied.
If the only consideration for decomposition is the user experience then there are no particular difficulties in splitting a form such that it spans multiple Perceivable Units within a single Delivery Unit.
The problem gets considerably more difficult if it becomes necessary to split a form into more than one Delivery Unit. A general solution does not yet exist but would need to cater for the dependencies that can exist between the different parts of a form, particularly interfield constraints and event handling.
If a form (either XForms or HTML) cannot be automatically adapted then changes will need to be made to the specific Web Page. This may include the author creating alternative versions of the form which are compatible with the constraints of the devices. Some changes may be cosmetic, e.g. reduction of white space. Some are more fundamental and involve redesign, e.g. removal of less important components. However, it is not possible to "compress" a form ad infinitum, so at some point one must consider splitting the original form into a collection of smaller forms, such that the semantics of the collection is equivalent to the original single form.
A session is a sequence of interactive steps that influence an information state (context) that persists between each step in the sequence. A session is typically initiated by an initial request from an end-user, and each subsequent step comes from further requests from the same end-user. These subsequent requests are a result of the responses sent to previous requests.
Sessions play an important role in web-based applications. They enable a service to have a dialogue with the end-user while being able to accumulate information during the dialogue. The accumulated information may relate to security and authentication, to user preferences and to requests for specific services (e.g. financial transactions).
Many session mechanisms rely on cooperation from the client application (browser) to maintain sessions. The server will create a unique key to represent the session and will rely on the client to supply this key in all subsequent requests. In HTTP, the key is conveyed via cookies.
Alternative client devices must provide similar session support mechanisms to ensure that services can be offered via these devices. If HTTP cookies are not available in the client device, an intermediate (proxy) may supply this functionality, as illustrated in WAP. If cookies are not supported in the delivery path, then alternative means must be used to support the session keys. These alternatives include URL rewriting, where the session keys are embedded into the URLs of any links within the page delivered to the client. Selection of such links ensures that the key is returned to the server in subsequent requests.
Where possible, the session mechanism is hidden from the author, who only needs to be assured of the existence of a session context in order to proceed with subsequent stages of the service delivery. The author can deal with the context as an abstraction. Popular technologies such as JSP and ASP incorporate the context as a server-accessible object without exposing the author to the details of the mechanism.
The DIWG Charter required that the DI Working Group hold a public workshop on authoring techniques. This took place in September 2002 and the main points are summarized in this section. A detailed account of the workshop can be viewed on-line [DIATW].
A number of important results were obtained from the workshop, including insights into the following:
These are summarized in the following subsections.
A common theme at the workshop was that of a "fragment of content", though this concept was expressed via many different names and definitions. It is generally agreed that there needs to be a definition for something that is less than a document, and there needs to be a set of properties, structures and operations surrounding this less-than-document concept.
Among the alternative names for "authoring units" proposed by the attendees were "Content Block", "Pane" and "Group".
An authoring unit should be capable of direct addressing so that it can be referenced by external documents or processes.
An authoring unit may comprise other authoring units. (This raises a potential nomenclature problem: the concept of "unit" does not permit further division.)
Generally, a layout is a structure that determines how to process a collection of content fragments together to present a single document. The process can be influenced by adaptation.
A resource, referenced by a single URI, should (where possible) be independent of the device or the context in which the resource is made available to the end-user. A resource is not to be confused with the concept of a document. For example, a resource could be "today's weather forecast". On some devices this would include a textual description and some maps, while on other devices only the text might be available. Nevertheless, the same URI is used in each case. The text fragment may in fact be the same fragment in all cases, as may the graphic fragment, but not all fragments are appropriate to the context.
The assumption that CSS is processed on the client was challenged. While it would appear that CSS was designed to be executed by the browser, it is clear that practitioners in the mobile Web technologies feel that server or intermediate processing is as important.
The following is a summary derived from position papers submitted by workshop participants: Covigo, MobileAware, Volantis, NTT DoCoMo, and HP. It includes conclusions for the DIWG's work on Device Independent Markup. Presentations by some participants included knowledge from commercially available implementations.
The broad range of solutions presented and discussed at the workshop revealed a number of common approaches to Device Independence, some of which are incorporated into commercially available products.
All the approaches provide a separation of content, layout, and style.
This is not just a good practice even for conventional website authoring, but offers additional benefits for DI authoring. The separation of these three domains allows the adaptation mechanisms to combine content, layout, and style in a flexible manner. A prerequisite for this is that the author can express choices, at least for layout and styles. How the combination of content with sets of layouts and sets of styles is performed, and how much of that combining process is under the control of the author, is an extra topic.
Two important aspects need to be associated with the units of content to support the selection and layout of these units:
Most submitters associate style (or sets of styles) with content using a top-level element, such as <head> or <canvas>. This is straightforward, since the author normally likes to apply a uniform style to the entire website.
Media are typically resources such as images, audio files, animations, which are normally inserted in a web page by a reference mechanism. Since the media needs to be well suited for a particular device or class of device, most implementors offer a choice of media. The available selection of media is however not listed directly in the content section of the DI document, but only referenced in the content section. Regarding the media selection list itself, implementors invented diverse constructs to associate the individual media with properties which help the adaptation process in choosing the most appropriate one.
A significant difference in the approaches is the type of (meta-)data associated with the media. Some describe the media properties (size, resolution, etc.), some describe the suitability of a particular media for particular (classes of) devices. The latter alternative is easier to process for the adaptation engine, but creates dependency on the classification of devices, which may change over time. In the latter case, the author also needs to know what media are suitable for what devices.
Separated from content, all submitters provide means to specify layout in the form of a hierarchy. In conventional HTML authored documents, layout is most often expressed through (nested) tables. A DI authoring system needs a similar way to arrange content on a page, but in more flexible way, so that the adaptation process can rearrange the cells of a table layout matching the device's display properties. Most authors choose to invent new markup to express re-arrangeable units of content.
All submitters mention the need for a mechanism to conditionally include or exclude sections of content. However, they are not very specific about how to express the conditions. Technologies to fill that gap may be XInclude, XLink, or SMIL (Media feature sets). This can be considered as a generalization of Media References (mentioned above). Most implementors use different mechanisms for inclusion of media versus inclusion of other content. The topic is also related to aggregation in the context of portals. The adaptation mechanism needs hints when to include or exclude content. This leads to the question of how to express conditional inclusion / exclusion and what range of conditions is needed.
Most submitters agree that adaptation should occur in the following sequence:
All submitters assume CSS for styling. This is natural since it is the technology that is most progressed and can also include device-dependent hints. One participant mentioned that server-side CSS processing may be needed to feed devices which don't understand CSS. From a consistency and technology point of view, it would be preferable to use XSL:FO instead of CSS, but industry did not follow this route and DI authoring systems need to be compatible with today's environment.
Submitters emphasize that new markup was needed to express the semantics sketched in the preceding subsection, but they also emphasize that the amount of additional markup should be minor and compatible with existing markup.
The question is one of deciding which body of markup to use as a basis for additional markup. Most authors seem to rely on XHTML Strict.
One view expressed at the workshop was for the DIWG to base its work on the (then current) XHTML 2.0 Working Draft, which has the disadvantage of being incompatible with previous XHTML versions, but the advantage is that XHTML 2.0 contains many features which can be extended for DI authoring.
Below is an examination of how markup used by the submitters in their implementations could be replaced by XHTML 2.0 markup.
Since association of styles should happen at top level, the <head> or <body> element could be extended either by appropriate attributes or by a sub-element taken from XLink or XInclude.
Consider the following possible, though not official, enhancement to XHTML. A nestable element of XHTML 2.0 called <section> could be the element that denotes a unit of content. Whether this unit is permitted to be further subdivided might be expressed by a new attribute "splittable". To assist the adaptation process, such a unit of content needs many more attributes such as:
Since these enhancements would only be used during authoring and adaptation, and not delivered directly to the client, they would not adversely affect any devices that support XHTML 2.0.
Regardless of whether a choice of media is needed to allow flexible inclusion, or a choice of content units, some markup is required to express the selection possibilities. The markup should be usable for choice of media and choice of other content to be included. Most authors invented their own markup, but XLink, XInclude and SMIL need to be looked at first. XML Schema also provides elements such as <choice>.
Beyond a basic choice statement, the set of expressible conditions need to be determined. In particular one must be clear whether the conditions should be based on properties of the content (special case: media) or on the delivery context or on both.
Some authors have invented new markup to express the application of layouts, though conventional table markup may be suitable (at least as a base) for expressing arrangement of content units on a page. It is likely that the table cells need additional attributes to provide hints to the adaptation process. The same may be true for rows and columns. XHTML offers only one table feature that may be used in adaptation: the colgroup element, that can group related columns.
Authors generally agree that content and layout should be separately defined and associated in a most dynamic way. In current implementations, this is done through referencing a separate section of the DI authored document. The first question to be answered is whether layout should reference content or whether content should reference layout, or whether a bidirectional reference is needed. The second question to be answered is whether XHTML 2.0 reference mechanism (which is actually HLink) is suitable or whether XLink needs a closer look.
One author mentioned the availability of markup that encloses code to be passed though the adaptation process without any modification. The primary purpose is to pass scripts to the client device. Again, it needs to be checked whether this requires new markup or can be done with constructs such as [CDATA]. In either case, it is a requirement that adds some additional complexity.
One author suggested a helper element that facilitates the modification of any attribute of a preceding element. This is an unconventional approach, but apparently can help avoid the redefinition of existing markup.
One author mentioned new markup to express basic interaction primitives. It was hoped that XForms would provide the needed features. The necessary variation in the presentation of form input fields (depending on delivery context) may be achieved by combining XForms with conditional inclusion or exclusion as mentioned above. This would be preferable to extending XForms with new attributes.
This Note has presented several techniques for device independent authoring, all falling within a spectrum of classification ranging from Single Authoring to Multiple Authoring, which in turn typically serve as input to various forms of Adaptation.
Single Authoring is preferable because it improves affordability by avoiding extra work, does not require device/modality expertise, simplifies authoring tools, reduces maintenance overheads and reduces storage overheads. There are cases where multiple authoring is necessary, though the reasons for such necessity can vary and are not always technical in nature.
Multiple Authoring techniques are generally to be avoided because of the considerable list of disadvantages, including:
Several W3C Recommendations incorporate aspects of Multiple or Flexible Authoring, such as selection in Media Queries and SMIL. None of these features are sufficiently mature to fully support the growing demands of device diversity, though the trend is towards supporting independence of the device while adding contextually sensitive features.
Adaptation is equally challenging. We can already see some of the necessary adaptation tools in XSLT (transcoding), XSL-FO (flow and pagination) and XForms (context-aware presentation). One key to successful adaptation is awareness of context. To this end, technologies such as CC/PP will play a vital role.
Device Independence can be achieved through a combination of the following intertwined features, which hopefully will be increasingly present in future W3C Recommendations:
- Appendices -
This appendix covers the adaptation techniques upon which the various authoring techniques rely. An authoring technique may employ one or more adaptation techniques, and each adaptation technique may be employed by one or more authoring techniques. The details of adaptation techniques presented in this appendix are incomplete, yet representative of the current state of the art.
This subsection considers several established processes that adapt content to its context. Where relevant, other characteristics of the processes are described, including: efficiency, complexity, security and position within the delivery path.
A simple adaptation mechanism can be achieved using only the ability to select (subsets of) available content, or to exclude subsets of content, on the basis of a decision mechanism. Typically, decisions are represented within the original content for execution at the time of delivery. The authoring tools must provide the means for the author to express such decisions.
In URL redirection, the server takes an original URL (a Web Page Identifier) as requested by the client and supplies an alternative URL for the client to use. The server may instruct the client to make a new request with the new URL, or the server may make the request on behalf of the client. In the latter case, the server is acting as a redirecting proxy. In all cases, the end-user perceives only the initial URL as a common entry point, and in this manner the content is adhering to DIP-2. The target of the redirection is determined by the server through an evaluation of the context (e.g. the type of device). URL redirection is very easy to implement (and is the normal behavior of proxy servers) but it is insufficient as a long-term solution to Device Independence.
Typically the advertiser of the page will require a memorable and/or short URL, so that it is easy for humans to remember and to enter into their browser. Translation of this "mnemonic" URL can involve several steps. For example, many sites use "www" as the local name of the server but do not advertise the "www." part of the FQDN, and thus the first part of the translation is performed by the DNS server that responds with the IP number of the machine named "www". As another example, a server may respond with a "HTTP redirect" that is nothing more than a "Location" header whose value is determined by the device context. However, redirecting a client introduces a second request that delays the eventual delivery of the content and is particularly noticed in mobile environments.
URL redirection also causes problems for bookmarking. If the user can create the bookmark directly, the user would likely use the mnemonic version. If the browser has a bookmarking feature, it will likely bookmark the redirection target and not the mnemonic, which means that the bookmark would not be device independent.
The following example shows a meta tag recognized by many browsers. If received by the browser, it will use the information in the meta tag to generate another request to the indicated URL after a delay of 10 seconds
<meta http-equiv="refresh" content="10;url=http://other.location.xyz"> |
An in-document decision tag is one that wraps content to be included or excluded according to contextual information. The author must supply expressions that are evaluated to determine if the wrapped content is selected. There are several variants:
A layout, selected on the basis of the delivery context (such as the size of the viewable area), may subsequently determine which content resources are included in the final presentation.
The decision to present content, and the manner of its presentation, can be determined on the basis of style. As an example, CSS Media Queries [Media Queries] permits the author to select one or more stylesheets on the basis of device class/characteristics. The selected styles may in turn determine if and how the styled content will be displayed. However, this is not always sufficient because all of the content must be delivered to the device in advance, despite the possibility of some of the content not being presented to the end user. A server-side styling adaptation technique may be employed in this case, where the styles are selected and applied prior to delivery to the device. This requires that the adaptation processor has at least the same contextual information as would be available to the client-side styling process.
Relevance is an XForms concept that describes the suitability of content on the basis of contextual information. It incorporates an expression language so that relevance can be expressed as a Boolean. The concept of relevance may be incorporated into content markup, as the following sample suggests:
<xhtml> <img src="aaa" alt="bbb" DI:relevant="Context('colorDepth') < 257" /> <div DI:relevant="Context('display') = 'none'">...</div> </xhtml> |
If DI:relevant evaluates to true, content is included; if it evaluates to false, content is not included. A number of presenters at the Device Independent Authoring Techniques workshop [DIATW] demonstrated similar approaches, where a decision expression can be embedded into the source content. On the client-side, one can use CSS Media Queries to achieve similar results, and it is technically possible to execute Media Queries at any point in the delivery path.
Navigation generation is the process of creating a navigation structure from the available content and/or information provided by the author. It is particularly useful where the content is complex or dynamic. The main influence on navigation generation is the target content, since the purpose of the generated navigation structure is to assist the end-user in getting to (parts of) the content. Other influences include the intentions of the author, the relative priorities of parts of the content, the ability of the browsing device to render navigation features, and many other contextual influences.
It is possible, in some cases, to generate the navigation features automatically, such as creating a table of contents from the headings in a document, or a sitemap generated from the links between documents. In other cases the navigation features must be created directly by the author due to the absence of information in the content that would determine the appropriate navigation paths. An authoring solution must provide the author with the means to influence any automated navigation, and the means to provide specific navigation information. It is expected that this author-provided or derived information will be used by adaptation processes to provide a variety of navigation features that, to the best of the device's ability (and suitability to other contextual constraints) will best represent the navigation paths intended by the author.
The dominant navigation structures are tables (lists) and menus (hierarchies), which are explored in more depth below.
A "table of contents" is an ordered list of headings, possibly structured according to relative importance. The table includes the headings and a means of navigating to the referenced portions of the content. The order of the headings corresponds to the order of occurrence within the content, assuming the content has an inherent order. In the absence of an inherent order, an alternative order may be imposed (e.g. alphabetic). Where an alternative ordering has been imposed, the table would be called an "index".
A navigation menu is a list of links to other content. The links may be within the current content, the current site or external. The presentation of the menu may be fixed or dynamic (e.g. drop-down form elements). The size and complexity of a menu is normally adapted to the characteristics of the device. Small devices will require smaller menus. Static menus impose less processing overheads, and are suitable for certain mobile devices. The order of items within a menu may also be tailored to the device, to the user preferences or even the past behavior of the user.
An author may construct one or more menus to suit the categories of device, or may leave the construction of menus to an adaptation process. Where adaptation is used, the author may need to indicate the suitability of candidate links to different categories of device.
A navigation menu that indicates contextually relevant documents in the current site, and a suitable "home page" for the site. This form of navigation is suitable for users who will normally navigate through a site (i.e. will probably have a long path). This form of navigation is also suitable when the size of the navigation aid must be kept small.
Determination of the nearest neighbor can be done by the author, or may be derived from a hierarchy representing the entire site. If a hierarchy is used, the nearest neighbors would be the parent node and all of the children nodes (and possibly the root node).
A full site representation is a navigation aid that is not contextually sensitive. Instead it represents the entire site regardless of the current page that the user is viewing. This is OK for small sites or for devices that can display large complex representations, possibly dynamically such as collapsible navigation trees.
The purpose of Next and Previous links it to facilitate linear navigation to adjacent content. Subtle differences in navigation can be introduced by different definitions of adjacent. The concept of "next" is usually determined by the author, while the concept of "previous" has various interpretations. |
As an end-user navigates through a site, a path is created. The concepts of Next and Previous can be relative to this path. 'Previous' always refers to the page (possibly a URL) that the user was viewing prior to the current page. 'Next' applies when the user has used 'Previous' to return to a page, so that 'Next' will advance the user along the established path.
The concepts of Next and Previous can be defined by the author, or derived from suitably structured documents (e.g. slide presentations). For example, the author may structure the site according to chapters, such that if the user is viewing Chapter N then 'Next' refers to Chapter N+1 and 'Previous' refers to Chapter N-1. This is regardless of the path that the user has used to reach Chapter N. For example, if a user navigates from Chapter 6 to Chapter 3, then 'Previous' means Chapter 4 and not Chapter 6.
In a Next and Back system, the 'Next' page is determined by the author while the 'Back' page is determined by the path the end user has used to reach the current page, and refers to the page that the end user was viewing prior to the current page.
To avoid unnecessary repetition or to reduce text, some content could be substituted by a reference to the full content (either at the end or on a separate page). For example, instead of a complete copyright notice appearing on each page, this can be substituted for a link to a separate page containing the notice.
Note: automation of such a process can produce unwanted results. Consider the headline: "President eats astronaut food during visit to launch site." used as a link to the complete news item. A substitution process may use ellipsis to shorten the link, which becomes "President eats astronaut ...". This highlights the need to control substitutions to avoid semantic changes.
Some devices do not support a full range of character entities, such as currency symbols. Although the markup may support such entities, in order to render them on the device they must be substituted. For example, if the pound symbol is unsupported, a graphic could be substituted, or the text "GBP".
Adaptation that involves conversion from one format or markup into another by web intermediaries is known as "transformation". Technologies include XSLT and DOM manipulators (for markup transcoding) and image filters.
An adaptation processor is a mechanism that takes one form of content as input and produces an alternative form as output. Adaptation processors would normally be expressed in software, although hardware-based processors are also possible.
Adaptation may be as simple as minor alterations to markup to work around minor browser differences, or they may be as complex as moving from one modality (e.g. text with markup for visual presentation) to another (e.g. voice synthesis and recognition). The complexity of the adaptation depends on the difference between the authored/available content formats and the supported/preferred content formats.
This subsection considers different levels where the process of content adaptation could take place. Three locations are identified on the end-to-end path from the server to the target device: the server side, the client side and the intermediate systems. Adaptation processors could cooperate and thus the final adapted content could be the result of more than one adaptation task applied at different levels. The author of the content can exploit these possible adaptations and apply necessary techniques that make the content more adaptable and negotiable.
Content adaptation can be performed at the server side if the target client presents some limitations (processing power, access method, screen size, storage capabilities, etc.) and it is not capable of receiving or adapting the original content itself. Server-side adaptation can also be applied to respect the delivery context of the content, for instance to optimize the use of the network resources (e.g. if the bandwidth is limited) or in order to reduce the delivery latency and to avoid unnecessary download delays. The adaptation scenario can depend on the communication protocol used in the content delivery. In some situations, especially in stateful protocols, the server requires client feedback in order to perform the adaptation efficiently (such as in RTP and RTCP protocols). This happens less in stateless protocols such as HTTP where the server adaptation depends on one client/server interaction.
This adaptation mechanism consists of choosing the best version that exists on the server on behalf the user agent. The selection is applied on the available set of variants. It should be based on the variants characteristics that can be extracted directly from the alternative resource or provided by the author who can specify some attributes concerning the delivery context of that resource. SMIL represents an example where the author can associate a variant with a particular context inside the Switch element or anywhere in the document using the in-line test attributes. Selection parameters are based on the delivery context and could concern the language, the media type, the char-set, and other constraints. Different ways of variant selection can be followed. A variant can be selected if it satisfies all the specified assertions regarding the delivery context (as it is applied in the evaluation algorithm of SMIL Switch) or if it has an acceptable presentation value calculated using some algorithms based on the quality values that are associated with the negotiation dimensions (as it is applied in some implementations of HTTP/1.1 content negotiation) Instead of storing and providing multiple variants, each matching the characteristics of a target device (or a class of devices), the author can provide only one variant and rely on the adaptation mechanisms to deliver the content in the appropriate form at the destination. As we can see in the following, the server adaptation can be applied at the structure or the media level of the content.
This process concerns adaptations applied on the global document structure. An example of such applications: transforming HTML to WML, filtering HTML documents, transforming XML to SVG, etc. XSLT is one of the efficient technologies that can be used to reach this objective. A structural transformation can either keep media resources used inside the original document (images, videos, etc.), filter them or use an external transcoding method in order to adapt the media for the target context. Structural transformations are widely related to the characteristics of the authored markup. In order to enable advanced transformation, such as semantic one (content summarization, etc.), the authored markup should be enriched by some elements and information related to the semantic of the content and not to its presentation.
This kind of adaptation covers all the transformation applied on media resources like image and video adaptation (color reduction, resizing, etc.), media transcoding, and other methods that operate directly at the media encoding level. The application of this process at the server side helps to avoid sending media formats that are not accepted by the target user agent (using format conversions, e.g. JPEG to WBMP) and also to apply further adaptations that consider the client limitations: media resizing for small screen devices, media compression to speed up the download time, etc. The real-time media adaptation process should be avoided if the author provides an alternative that satisfies the delivery context.
The server can tailor some meta-information related to the expiration of requested resources in order to avoid frequent document prefetching or prevalidating by clients in a poor network connection. The server can provide some added information (for example via headers) so that intermediaries used in poor connection networks can increase the freshness interval for resources and apply efficient caching policies.
Device Independence Principles [DIP] defines the terms Delivery Unit (DU) and Presentation Unit (PU) that are of particular relevance to the subject of Decomposition.
There are two primary reasons that some form of decomposition takes place:
An adaptation system can make use of either or both of the following two decomposition techniques:
Continuing with WML as an example, it is possible to use a deck of <card>s to implement a collection of Pus A similar approach could be adopted with HTML where the adaptation process uses regions identified by <div> to define each PU and then manipulates the set of PUs such that only one is visible at a time.
Decomposition processors may use information provided by the author (such as the <div> delineated regions) and/or contextual information (e.g. available memory in the client) to determine the boundaries of the generated DU/PUs
To avoid sending large amount of device and user profiles information to a proxy or a server for content adaptation, the author can exploit client-side adaptation. The author can apply all or partial adaptation techniques on the client-side depending on properties of the client, such as memory size and processing speed.
Image resizing is useful when the author needs to fit an image into a limited space. The author can scale down/up the image using existing technologies such as SVG, or employ image versioning. Scaling of image components need not be applied uniformly. For example, when shrinking a map it may be necessary to scale the text within the image at a different rate, or possibly remove some of it. With image versioning, the author may create or generate a discrete set of alternative images with various sizes and formats. At content request time, the appropriate image would be selected based on the client's capabilities. |
Not all fonts will be supported by all devices; so many devices will substitute a similar font. This assumes that font categorization is being used. The author should be able to influence the substitution mechanism by offering a prioritized set of options. Presenting information to the user that relies on being able to distinguish between fonts is not advisable because the actual fonts used may not be distinguishable. |
Some clients are capable of content transcoding, for example by processing XSLT client-side. This process tailors text, images, audio, and videos based on the clients' capabilities. The transcoding process may include content translation/conversion, content filtering, content summarization, and content selection. To exploit transcoding, the author may provide a set of transcoding policies such as the rules in XSLT and JSP, and sufficient content semantics to describe the pre-transcoded content.
Dedicated rendering is the process of displaying additional content types, typically those that are not normally native to the browser. A client-side process would receive the specialized content and map it to a suitable presentation. Browser plug-ins are a common mechanism for providing dedicated renderers for specialized content such as Portable Document Format (PDF), Virtual Reality Markup Language (VRML) and Flash. Even XHTML modules (e.g. SVG) can be supported via dedicated client-side rendering plug-ins. These client-side renderers are dedicated to a specific content type and therefore do not need to be as powerful as generic browsers. However, the author must consider the possibility of the absence of such renderers when creating content. Alternatively, a server-side adaptation system could select/create an alternative to the specialized content.
A form needs to be resized when it cannot fit into a limited space. The adaptation process can resize the form by reducing the visualization of form controls (widgets), repositioning form controls, transforming form controls, and/or splitting (decomposing) the form.
The adaptation process can reduce a form control's size by reducing the font size and/or by clipping extra whitespace from text-related form controls. When reducing the font size, the adaptation process should choose a size that is human-readable on the client device. When clipping extra whitespace from form controls such as <TEXTAREA> and <TEXTFIELD>, one should ensure that the controls are long enough for the user to view and input text.
When repositioning form controls, the adaptation process should provide enough semantics such as relationships and proximities between controls, so that controls can be repositioned sensibly and functionally. The repositioned form controls should not confuse the users.
The adaptation process can also apply form control transformation. Figure 1 shows a form control transformation from a set of radio buttons to a drop-down box, and Figure 2) shows a form control transformation from a set of label-textfield pairs to a list-textfield pair. Not all form controls can be applied transformation (e.g., a label) and not all form controls can reduce its size by a transformation rule (e.g., the "list to drop-down box" rule can reduce the list's size, but the "list to radio buttons" rule may not reduce the size). The adaptation process should prioritize and select form controls and transformation rules that give the best size reduction. After form control transformation, one should ensure that the transformed form controls are fully functional and are recognizable by users.
In general, form splitting is used when no other form resizing techniques are applicable. Before splitting a form, the adaptation process should group related controls and provide enough semantics about each (group of) control including relationships, proximities, relative size, priorities, etc. The form can be split based on the semantics of each control group. After splitting a form, the adaptation process should provide navigation among the split forms. The adaptation process should also ensure that the split forms are fully functional.
Several client-side markup solutions provide selection mechanisms that support complex expressions involving contextual data. The client must retrieve the entire markup before evaluating the expressions to determine which (if any) of the enclosed markup will be selected. The enclosed markup may in turn refer to external resources, which are then retrieved. SMIL provides such a feature for selecting markup subsets, and CSS Media Queries provides a similar feature for selecting appropriate styles.
Intermediate adaptations are generally applied in a proxy-based architecture. Such architecture (Figure 3) consists of adding a third entity between the server(s) and the client(s). The proxy can be seen as an intermediate processor that acts on behalf of a client by receiving content from a content source, processing the content and then relaying it to the client. This kind of architecture represents a good approach to address the heterogeneity problem of clients and servers. Indeed, in a proxy-based architecture the network platform is not modified and the environment characteristics that already exist are preserved. Proxy-based architectures can be very complex by including several intermediaries dedicated to specific tasks such as image and video transcoding, video broadcasts, firewalls security, etc.
In the context of content adaptation, the proxy is the entity responsible for retrieving client requests and contexts and performing possible adaptation on the content received from the server. The adapted content is then sent to the client with respect to its characteristics. The proxy can transform the content if no alternatives are available. All the proxy processes are designed to behave transparently to clients and content servers. Adaptation techniques discussed in Appendix B can be applied within the proxy. Note that the "alternatives" selection technique is more complex in the proxy case. Indeed, variants are not stored in the proxy side so the proxy has limited control regarding the knowledge of existing alternatives and the extraction of variant characteristics. This can be achieved using further message exchange between the proxy and the origin server, which can increase the content delivery time.
One kind of adaptation that can be performed at the proxy level is the selection and the redirection to the best server mirror (cache) that can reply to the client request. Some content servers have a large collection of mirrors where the content may be stored and delivered in different ways. The proxy selection could be based on the client connectivity, the client capabilities and preferences, the mirror location, etc. Once the selection is done, the proxy should apply a mapping between request URI and specific resources. |
Proxy adaptation can also concern the protocol used in delivering the adapted content. After a particular adaptation, e.g. transforming HTML to WML, the proxy can still control the content delivery using another protocol different than the original proxy/server exchange protocol. This kind of proxies is called 'protocol proxies' and can work at the network, the transport or the middleware layer. The proxy can also control the delivery by applying compression algorithms to reduce the size of the delivered content or by bundling embedded objects into a single resource to avoid multiple rounds of requests. |
A document may be stored, constructed, presented or dissected, according to a predefined structure. This section considers the form and application of structures comprised of pages of content and the relationships between these pages. For the purpose of this task, pages will be those content objects that are accessible to the end-user via URIs, and relationships shall include hyperlinks (e.g. href attributes in HTML anchors), frame-based adjacency (i.e. mentioned in the same frameset) and pages reachable via an input from the end-user (e.g. via a form).
Several well-known structures are frequently found in authored content. These are explained below, with special emphasis on how these structures influence content navigation.
A linear structure has one start point and one end point, and a single path from one to the other. The delivery closely matches oral (verbal) communication where the content represents a narrative. Significant parts of the content may be highlighted, and parts of the content may be repeated (to avoid having to review previous parts).
The author of a linear structure must take into account the fact that the reader may not have access to the whole document at any point in time. For example, if the document is being delivered via voice synthesis, the listener will only have knowledge of the document up to the current point in the reading, Thus forward references may present problems. Furthermore, backward references will rely on the reader's memory, which (in the case of most humans) is unreliable, especially with large documents. Repetition instead of references will alleviate this problem, at the expense of increased document size and longer delivery times.
A content hierarchy is a recursive/nested structure represented by a node, where a node comprises an ordered collection of one or more items. An item may be a piece of content or a node.
There are many variations on this definition. In some, the node has a title and this title may be considered as a piece of content. In others, the node is purely abstract and has no content other than that contained in its collection of items. Writing conventions suggest that the former definition is more useful, since it agrees with the concept of "heading".
It is also a convention that the development of the concept/information represented by the document (the root node) follows the order of the collection of nodes, and that the deeper nested nodes represent greater levels of detail.
HTML incorporates the concept of the hierarchy in its six heading tags, <h1> to <h6>. One expects that any rendering of the markup will reflect the importance of the headings through suitable formatting, highlighting or similar styles.
A reader may choose to read the document in a linear fashion. This may mean selecting a particular level of the hierarchy and reading only the pages at that level (ignoring nested pages). A reading of the entire document might involve a traversal of the hierarchy, typically depth first (strictly: following links as they occur in the pages, and returning to the parent at the end of each page).
Alternatively, the reader may navigate (quickly) to a specific level of detail via repeated selection of sub-nodes. A "table of contents" derived from the node titles is often used to assist such navigation. This kind of navigation is typical when the reader explores or has foreknowledge of the document (from a previous reading, or from experience of a similar version) and already has some idea of which item must be located.
A mesh is a (potentially unordered) collection of content containing links between the items of content in the collection. It does not require a specific entry point. It does not require any specific reading order (traversal). It is necessary that for each piece of content within the collection there is at least one path from a potential entry point to the piece of content. This ensures that all of the content is accessible.
A mesh may be random, where no traversal has preference. A mesh may have order, such as a collection of content that can be traversed in more than one dimension. As an example, consider a collection of poetry that may be navigated in chronological order, in alphabetical order (according to author's name, or title) or in any of a number of other possible orderings. Such collections could be represented as pieces of content where each piece can be retrieved via some unique key. In effect, the mesh is a database of content items, and the keys represent the access and/or links to these items.
An author of a mesh must have a means of creating the content item and identifying the key(s) to associate with it.
There are other structures that may be used to represent content and its interrelationships. These have limited or narrow application, and are listed here only for completeness.
Below is a brief summary of the challenges identified in the Authoring Challenges document. They are paraphrased for brevity. It is these challenges that the various authoring and adaptation techniques described in this Note hope to address. For a detailed explanation, consult the AC document.
This document was produced by members of the Device Independence Working Group. With respect to this document, the role of participating members is noted below:
The principal and active members of the Device Independence Working Group at the time of original publication are shown below.
Last modified: $Date: 2017/10/02 10:28:32 $