Liddy Nevile, Adjunct Associate Professor, Applied Computing Research Institute, La Trobe University [HREF1], Kingsbury Drive, Bundoora, Victoria, Australia, Liddy.Nevile@motile.net,
Edgar Smith, Pro Vice-Chancellor Information Technology, La Trobe University [HREF1], Kingsbury Drive, Bundoora, Victoria, Australia, pvcit@latrobe.edu.au
Currently users looking for educational objects, resources, services, do so with no clear way of identifying and matching any special needs they may have. Such special needs may be the consequences of disabilities or, equally often, a consequence of the circumstances in which the user will want to access the object. As technologies improve, the opportunity to customise information and services for users increases. This paper is about how the use of descriptions of objects can increase the potential for customisation and thus accessibility of Web objects, in particular, and other digital objects in general. It is proposed that emerging technologies will increase the benefits of this approach. The paper reports on work being undertaken in several international forums in which increased accessibility is the goal.
Accessibility is about making all resources, services, devices, events, suitable for use by all people, always, despite, and without reference to, disabilities they may have. The World Wide Web Consortium (W3C [HREF2]) has pursued work in this area for some time, and the problem can now be recast in a more useful and general way than has been done in previous work. At any given time, a user may be able to read text on a screen ... or not. The 'or not' may be the result of a physical disability, or may be because the user is driving a car and cannot safely read, or is using a mobile phone or other hand-held device with limited text display capability. Anyone can be included in lists of those who need accommodations in such circumstances.
Increasingly, broadcasting is being replaced by narrow casting. In addition, computer interactions are becoming more ubiquitous as computers become more pervasive. This means changes of modality, sequence, structure, and content for the user in their interactions with computers and digital objects. It is no longer a matter of matching a resource just to a user’s content needs. In the same way as the library model for discovery is being extended out into mixing and matching of resources, human and device accommodations, Web and other services, locations, contexts and times, so is the range of interactions expanding.
Thus, users have a set of capabilities, resources have a set of formats, devices have a set of rendering options, and contexts and purposes change priorities. Increasing accessibility (as it is defined in this paper, is best understood as matching capabilities, formats, and rendering options in the most appropriate manner, through the application of appropriate services. The definitions that identify the characteristics of each of the groups involved in these interactions are the ‘metadata’ of the objects. It is through the interpretation and use of the metadata that the correct combinations or customisations can be assured.
Metadata is generally described as structured information about information (or something). Traditionally, metadata has been used to describe resources in the same way as catalogue records were used for library items. Already, however, there are standards organisations with clearly identified needs beyond those available from the current metadata sets. In this paper, we argue that there are at least four sets of descriptions that will be needed, if not more. Four candidate sets (profiles) are proposed as a result of work done with test scenarios (use cases) in a range of contexts (see below). (Four is a 'magic number': four is the smallest number that provides for the kind of interactions proposed but it may turn out that many more sets will work even better.) A metadata profile is described here as a set of elements that are used to describe something. The elements are not exclusive to that set, but the set represents those elements that are useful in a particular context, for which they are called the metadata profile.
This paper proposes that building on the available architecture to include more variety of specific metadata for services, products, people and events promotes greater interoperability and interactivity. It only proposes more variety where that can be established easily within small sets or profiles, following the Dublin Core model.
The Dublin Core Metadata Element Set (DCMES [HREF3]), attracts international attention in part because it is simple at the core but supports a broad spectrum of modular metadata functionality. This flexibility allows it to be tailored to the functional requirements of a large variety of systems and environments. DCMES contains 15 basic elements that comprise a metadata profile for resources. While DCMES is targeted primarily to information resources, such resources are part of a increasingly broader and more complex context.
Imagine a person A, that needs a resource B, after it has been transformed by a Web service C, for use on a device D. The metadata profiles of A, B, C and D have information required variously for and by A, B, C and D. A discovery request will travel along paths connecting A, B, C and D as they perform their duties. The order in which these duties are to be performed may not need to be fixed.
Likely case-scenarios exemplifying interactions of the type described have been explored in a number of contexts:
DCMES is usually involved in a single-profile single-entity/object activity: a user’s resource discovery request is matched with a resource’s profile. The scenarios considered above involve more entities and thus it is argued there is a need for more profiles. As the proposed development can take place at the level of metadata profiles rather than at the level of system software, it should not involve significant barriers to progress nor meddle with resources, services et cetera.
The main idea is to distribute well-known metadata requirements between related profiles or entities and then recombine them in new sets of metadata requirements reflecting relationships that profiles or entities may be part of. The metadata that each of the objects have (about themselves and about other entities) should be the least that is sufficient for them to operate autonomously in serving the user.
The Warwick Framework [HREF8] provides the basic rules and architecture for the Dublin Core system (see below). It provides for mix-and-match sets of metadata. This paper argues that if those sets were about a range of objects, not just resources as is the DCMES, nothing would be broken but there would be far more flexibility and accessibility.
Four profiles, if based on the Warwick Framework’s Lego-like modularity, will allow relatively straight forward inclusion in existing well-designed, standards-compliant metadata systems, extending the utility of these with minimal additional development. In appropriately designed existing systems, loading an additional machine-readable metadata schema will support the query logic necessary for the additional functionality made possible by the additional metadata terms. This will simplify the development of Web sites that support flexible combinations of services and content that can change as community requirements, content models, and device capabilities evolve. This will be an important advance over current service models with rigidly pre-cast notions of requirements, contexts, and purposes.
In the past, it has generally been assumed that all that is necessary is to extend the DCMES. This is done when DCMES users add elements to suit their local purposes or others that have been recommended as generally useful. The DCMES can also be refined to suit particular purposes. The comparatively low level of use of DCMES in other than the resource context, however, shows that this approach is not working. Repeatedly, people concerned with objects other than resources show interest in DCMES but find that the profile does not accommodate enough of their needs and so they revert to other, more object-specific, metadata systems.
In the accessibility and usability context, more precision is required not just for discovery of a range of types of objects but also to let objects communicate information to each other, and for descriptions of potential relationships between objects. This calls for more appropriate metadata profiles.
Having four profiles that interact, each with information of relevance to the others, ensures the minimum number to test the theory that specialised profiles will support interactivity in the way required. The four proposed profiles has been considered in a preliminary study. In January 2003, an international meeting was held in Washington DC with testing of the four candidate areas as one goal [HREF9]. Participants thought it would help if DCMES could be redeveloped to better support their specific metadata needs.
There have been similar proposals for new implementations of the Warwick Framework in the past with new profiles analogous to the DCMES but more specifically suited to other objects. Unfortunately, these earlier proposals have never been implemented, possibly as much because they were undertaken before the technology was mature as because it takes too long for commercial developers and is too resource intensive for most academics. It is nonetheless now costly not to work on this development. A useable generic development is crucial to accessibility and will be of great value when used in other contexts. Education has pressing needs for greater accessibility and usability
The interaction between, say, a user, a remote controlling device, resources or services, assistive technologies and Web services depends upon information passing from one to the other so that all agents know what is going on.
Resources are usually created and catalogued by people that users may trust or not. Some users want more information about authors, and in this process may refer to authority files in libraries identifying authors in detail. Others may want to know about the authority of the cataloguer who makes statements about the author’s work. They may want to know about authenticity of a statement, or critical expertise. Each time, they want to know more about people. Similarly, Victorian Government researchers require more information about who is currently doing research on topics of interest.
Metadata about people is not a new idea. Data-base directories of people have been developed for years. There is a strong tradition of people classification systems and there are technologies to support them. Like the Marc library record, these records are not easily created, are too extensive for most purposes, don’t necessarily suit present purposes, and don’t interoperate with product records. Matching a journal article that has a DCMES-based record with an author recorded in another system does not make for simple, light-weight, interoperable Web-style actions. One should be able to point to a light-weight author’s authority file available on the Web when classifying a resource. The resource record is likely to be more accurate if there is no doubling-up of information or room for spelling and typing errors.
The incompatibility between systems being described goes beyond mere lack of agreement or conformity between metadata creators. People have different attributes from resources. When resources are described, the three aspects that have proved successful in the resource context are identity, instantiation and intellectual property. Descriptions of people may need to be different: people sometimes need to be authenticated (eg as entitled to use a particular service) but not necessarily identified by name, so carrying a card that shows verifies their access rights may be sufficient. At other times, their particular identity is crucial, as for when they access their bank accounts. They may need to have their abilities and disabilities noted on both occasions. People’s achievements may be of interest in some circumstances but a matter of privacy in others (as in the case of library authority files). Similar circumstances exist when working with products and services. For services and intelligent devices, it may be important to know the manufacturer/supplier’s name, the functions that can be performed, the cost of the service, who is to be paid for the service, and how. The recyclable nature of the product may be important. Different types of information make sense in different situations and for different types of objects.
Accessible content has been the focus of work led by the W3C Web Accessibility Initiative (W3C/WAI [HREF10]) for more than five years. There are significant achievements in this area to date and in the future there will be a lot more ‘accessible’ content as increasingly the standards for the technologies used in the Web context include accommodations for accessibility. The range of services available on the Web (Web Services) to transform content is increasing. Content can be translated on-the-fly, or augmented with dictionary references, or modularised and sequenced for presentation on a telephone screen, etc. Content has thus become useful in unanticipated ways and can be functional, read, or operated on, as well as disaggregated or recombined and re-used.
In the era of ubiquitous and pervasive computing, when people and agents access information for many reasons, and services can be seamlessly and invisibly integrated into interactions, there is a need for interaction between a range of objects to support human action. People with disabilities may use assistive technologies, together with Web transformation services, to access products or resources. They need to do this differently according to time and location. People with a range of temporary deprivations work similarly, for instance, when they switch from their desktop to their phone on leaving the office and getting in the car. A change of location or focus doesn’t mean that communications-based activities stop. Changing a hand-held phone to a hands-free phone takes effect as the driver enters a car but the conversation may go on. Listening to the news is similar. If the news comes as a digital stream from a Web site, the same device, such as a phone, will be able to continue speaking news despite having changed from a hand-held to a hands-free service. Share prices listed on an SMS screen as a trader gets in a car may be read out when he is driving.
Matching users to services and content involves establishing criteria for search beyond those relating to the type of service or content topic sought. The abilities and facilities of users may need to be identified before the search can commence, or before it is complete. Since these factors influence the choice of discovered resources, there is a need for secondary activity, with autonomous agents modifying the discovered resource use. This may involve finding an appropriate Web Service or device to modify the resource and then having the product of that service rendered by a further device, which itself may need to be modified for the benefit of the user.
Interactions between content of various types is starting to make the Web an interactive medium for extended human and device activity. On paper, text with an image is one object, as is a web page within a browser on a standard screen. In digital form a Web page is a set of objects, with several components and a style sheet. Bits of Web pages often need extra software, plug-ins, to augment browser facilities. Plug-ins are small services to support browsers. More complex services exist on the Web to support things like language translation or transformation services that render content fit for modular, sequenced use on small screens.
DCMES treats a Web ‘page’ as one object but for accessibility and other purposes, this is sub-optimal. Warwick Framework architecture, and the DCMES can support the disaggregation of metadata necessary to ensure accessibility of a Web page in formats other than for a standard browser display on a large screen.
Scenarios are often altered somehow by changes in user circumstances or context: for instance, a tired user at home in the evening may need different settings from those used in the morning at work.
Specificity of profiles for types of objects is critical. Users with disabilities or special needs must communicate differently with different types of objects: resources need matching for content and possibly suitable formats and modalities. Services that apply transformations must know about the device the user will employ for presentation. This may depend on the context, so an autonomous event/context manager must know what that is. Metadata subsets that describe users must foresee the needs of other objects with which there will be interactions. Similarly, other objects must know what is likely to be available to them in terms of metadata. It should be available for them to use in their local operations and also for sending to other objects when interacting with them. All these objects must have enough metadata to act autonomously in deciding on the next step of the path around services and it will be necessary to find some way of ensuring that the path through the relevant objects is finite.
The success of the DCMES comes from its suitability for description of content. The DCMES originally contained fifteen basic elements, the things that everyone wants to know about a resource, such as:
Not all fifteen elements of the DCMES are used for every resource. The fifteen work for libraries making light-weight catalogue records from established Marc format records, with modifications. Museums and galleries work with modified metadata sets based on DCMES. These suit the digital collections.of those institutions. The educational community needs extra elements for teachers, the government community needs extras for their purposes. In addition, almost everyone has found they want to know something about the catalogue records so they need A-core metadata, metadata about the metadata. Also, the original DCMES has been qualified in several ways: one can refine an element to provide more precision by specifying a format for its value, or a controlled set of possible values, or some other criterion. When this is done, the refinement is either a registered DCMES refinement or it can be defined by the declaration of the scheme by which it is refined.
Stuart Weibel, the Executive Director of DCMI, describes elements as Lego bricks, to be mixed and matched to suit user purposes. The analogy works well: there are recommended encodings for all the elements and it is possible for a user of the DCMES to pick and choose which elements to use and be sure that what they choose for their ‘metadata application profile’ will be interoperable with sets chosen by others, so long as they follow DCMES specifications. The additional ‘Lego pieces’ from specific communities have been carefully scrutinised for their interoperability, and can be adopted with relative assurance that interoperability will be maintained. (The additional elements (or DCMES bricks) only become DCMES ‘recommended’ elements or qualifications when the Usage Board of the DCMI has heard all public debate on them, and worked through the implications of having the new element in their expert group.)
DCMES does support the relationship between objects in a primitive way but it is proving inadequate in contexts such as accessibility (see dc:accessibility Web site [HREF11]). The DC element dc:relation that points typically from one resource to another and identifies their relationship, is not operational nor a separate record. In fact, its inadequacies suggest that the proposed specification of metadata profiles will be more satisfactory, providing for descriptions of objects and their relationships. This would support a new kind of object grouping, a semantic grouping (see below), where resources are coupled with assistive technology services, for example.
Four categories a, b, c and d, provide for six links of interaction: ab, ac,ad, bc, bd and cd but a large range of paths of interactions. If such links represent the connection between an inaccessible resource and something that transforms it into an accessible resource, this combination could be classified and discoverable as a single entity. One can have Web pages containing an inaccessible interactive learning object that can be associated with a suitable alternative so the page can be reconstructed and become accessible. Thus, objects in any of four categories could each associate with many relationships between other objects or relationships. This complexity will be captured simply for each object in the proposed new profile set.
| Details of activity | |||||
| user -> | user agent | A user interacts with their browser or assistive technology | |||
| user agent | -> profile | The user agent has settings that indicate the need for preferences to be associated with the discovery process | |||
| profile -> | resource | The discovery process is modified by the user’s preferences | |||
| resource -> | service | The resource requires transformation to satisfy the user preferences | |||
| resource <- | service | The transformed resource is available for the user | |||
| user agent | <- | resource | The transformed resource is sent to the user’s access device | ||
| user <- | user agent | The resource is presented to the user |
| user -> | user agent | A user interacts with their browser or assistive technology | |||
| user agent | -> events/ context |
The user’s context on this occasion requires special accommodations | |||
| user agent | <- events/ context |
The user agent is notified of the context-specific requirements | |||
| user agent | -> service A | The user agent calls for specific services for resource transformation | |||
| service A -> | resources | The web service performs a suitably modified discovery action | |||
| resources -> | service B | Resource is required to be transformed by a service to match user preferences | |||
| service A | <- | service B | Service B returns a suitably transformed resource to service A (or directly to the user agent) | ||
| user agent | <- web service A | Service A returns the transformed resource to the user agent | |||
| user <- | user agent | User agent presents the resource to the user |
The Warwick Framework is an architecture for metadata. (DCMES is an implementation of the Warwick Framework that has been extremely successful in terms of its adoption and adaptation. It is the most widely deployed metadata representation system and is an ISO Standard (z.39.85) [HREF12].
The Warwick Framework architecture provides for DCMES’ extensibility and amazing versatility. The potential for inter-operability is extraordinary and significant activities have been undertaken world wide to exploit this feature of DCMES in the resource context (that is, document-like objects, DLOs). Baker [HREF13] describes the architecture as being a grammar, familiar to many English speakers. Baker’s diagrammatic form is useful in this context:
Using the qualifiers, one can use the DCMES grammar to express something like, "This resource is in English and about holidays in Asia." It is not possible, however, to express something like: "Mary needs a resource in English that can be displayed on her hand-held" because there is no provision for a description of Mary or her needs.
Providing Mary with what she wants is likely to involve a modified discovery process and the transformation of the resource before it is presented to Mary. What is required is a process that involves not only finding a resource with the right subject matter in the right language, but also matching it to Mary’s access requirements. It is not just a matter of matching the resource to a certain resource description, but also of matching a person and their requirements to services and transformations that will make the resource suitable for them.
The Semantic Web has architecture too. It is built up of triples reminiscent of those in the Warwick Framework. The Semantic Web has as focus the necessary constraints to make it possible for humans and computers to work on semantic inter-operability, in contexts beyond just metadata inter-operability.
HTML (HyperText Markup Language [HREF14]) made it possible to send objects around with very light-weight tagging that would show how the objects should be displayed in a browser. HTML was a derivative of SGML (Structured Generalised Markup Language [HREF15]), that was a full-blown language for tagging objects for professional publishing. XML (eXtensible Markup Language [HREF16]) has brought in some structure so that the tags can be customised but still be inter-operable. RDF (Resource Description Framework [HREF17]) takes advantage of the XML structure but constrains it so that there is a very simple semantic structure. RDF requires meaning to be organised in triples, but this in no way inhibits the complexity of the semantic (statements). Having constrained semantics means that it is easier to build logical connections into a collection of statements.
W3C aims to build a Semantic Web using such technologies as RDF. The following example is a RDF version of the statement "Mary has a well-trained dog, an Alsatian, called Fido". This meaning can be conveyed by a series of simple triples, one version of which is:
Expressed in RDF and displayed graphically, it may appear as shown. It could be developed differently and still have the same meaning, of course. Both humans and computers can work out that Mary’s dog is well trained, given standard representations. It should be noted that this is not the same as the original statement although it is consistent with it semantically.
The RDF form offers many paths that can be taken and expressed in sentence format. The ‘metadata’ about the person and about the dog can be mixed and matched when simple logic rules are used. That there can be various ‘correct’ representations, each of which have several simplifications, means that in algorithmic design, there is a need to produce methods that can work with any graph of the metadata and any valid simplification thereof.
Both Dublin Core and the Semantic Web have a grammar but the Semantic Web’s is more atomic and thus thought to be more expressive. The Semantic Web aims for inter-operability, as does the Warwick Framework. Currently the Dublin Core implementation of the Warwick Framework is most commonly used in either HTML or XML format, and uses the grammar described (above). The Semantic Web grammar is capable, however, of completely expressing Dublin Core metadata and can be used. There is a recommended RDF representation of Dublin Core metadata [HREF18] so merging the two architectures is comparatively easy.
The facility with which Dublin Core style metadata can be expressed in RDF means that there will be no reason not to use RDF and support the added richness, once the tools for working with RDF are commercially robust. The advantage of doing this will be that the four or more entities, that will be needed to interact to achieve the goals proposed in this paper, will be able to work in a semi-autonomous way. It will not be necessary for the user or user’s agents to arrange the interactions for the user’s benefit and it may not be necessary for the user to know of all the interactions that take place. The relevance of this technology to the arguments being presented is that with such a technology as RDF, it becomes likely that there will be mixing and matching of resources and services and devices and people that can be organised by autonomous agents.
The Semantic Web activity (autonomous, rule-based computer activity) that will support the user, will be achieved ‘behind the scenes’, on-the-fly, and independently of the efforts of the user. Once the semantics of a situation are represented in a way that makes it possible for computers to work on the information, algorithms that direct this work can become active independently of further human intervention, Syntactic interoperability is useful, and algorithms that match can be used once there is syntactic interoperability but the power of computers to understand ideas and make decisions based on them is what is envisaged here. that is known as semantic interoperability (see W3C Semantic Web Information [HREF19].
This paper reports on requirements analysis in the area of accessibility and identifies a need for more specific profiles for at least four classes of objects that interact on the Web. People, resources, services and contexts are the minimum set that could be considered but provide sufficient variety and independence to test the theory. The paper does not present the profiles that should be attached to the four classes: this is work to be undertaken in the near future. It is hoped that the arguments in this paper provide justification for that work. The paper foreshadows the advantage of this work being undertaken simultaneously with Semantic Web development work.