Definition and prototyping of a Renderer-independent ML

Submitters:
Guido Grassel, Nokia Research Center; Markus Lauff, SAP Research; Axel Spriestersbach, SAP Research; Michael Wasmund, IBM Germany

Currently, the usability of enterprise applications on mobile front-end devices is tackled badly, although optimization of the usability of the devices themselves as well as, for example, browsers installed on those devices has been plentiful. We feel that current techniques to make enterprise applications usable on mobile devices are not sufficient to be both efficient and cost-effective. The resulting poor usability of such applications is one of the factors which prevents massive acceptance by a broad range of users. This lack of acceptance, on the other hand prevents enterprises from investing heavily in further applications of this kind. The techniques sketched in the following sections we hope will contribute to solve that deadlock.

The development user interfaces for mobile applications to be used on mobile devices is today commonly based on templates or on transcoding.

The template-based approach builds on dedicated templates for each device and targeted mark-up language. In general, this leads to acceptable usability of the application. However, such an approach is costly because templates need to be developed for each application and each device or group of devices. This leads to a myriad of application variants to be maintained by a software house.

Transcoding on the other hand is quite cost efficient, but inefficient to adapt the complexity of application data output and input capabilities for different user interfaces on the variety of current and future mobile devices. Although there are sophisticated transcoding techniques available, those techniques are normally deployed in a proxy in an application-independent way and therefore lack important semantic information about the application. Examples for such semantic information are: relevance of certain sections compared to others, cohesive meta-information, handling hints about application of alternative presentations.

Reduction of the effort is crucial that is required for developing high quality user interfaces for the variety of existing and expected mobile devices. One crucial element in our project's approach is the definition of a series of device classes, each with specific characteristics. By proper classification, it will be possible to minimize the number of different devices for which adaptation is necessary. Our current draft list of device classes is as follows. This list still requires validation in practical experiments:

• Device class 0 - Voice interface with phone numeric keypad only
• Device class 1 - Smart phone that includes an embedded browser, small screen, extended key pad, highlight cursor navigation
• Device class 2 - Communicator that includes a much richer HTML browser, larger screen (compared to class #1), full keyboard, highlight cursor navigation
• Device class 3 - PDA that includes a similar browser and has a screen of very roughly the same side as the previous class, handwriting recognition, pointing device (usually a pen)

There have been classification attempts before - our approach is to explicitly include ease-of-use considerations in such a classification scheme, beyond the usual grouping according to physical characteristics.

Obviously, adaptation purely based on a specific device class may not always be sufficient to obtain satisfactory results, since the characteristics of an individual device within such a class need to be taken into account, too. More fine-grain device characteristics information from the delivery context will be needed. Besides detailed terminal capabilities, such characteristics may include user context information and personal preferences in addition to pure device characteristics. However, our approach is not to bother the application developer with individual device characteristics, rather with characteristics of a class of devices only. The fine-grained adaptation shall be largely automatic. Crucial to the envisaged automatic adaptation is to convey knowledge about the application to the adaptation process. Until today, such knowledge about the application is usually carried in person by the application programmer who manually adapts user interfaces for diverse environments.

We aim to overcome this time-consuming process of manually adapting application mobile interfaces to current and future mobile by building that process knowledge into a mark-up language. Our goal is to specify the description of the application user interface in an XML-based mark-up language to enable a rendering engine to adapt the complexity of the application input and output according to established devices classes. We call that language Renderer-Independent Markup Language (RIML).

We have made a survey of existing markup languages that could be used as a base language for the RIML. The following candidates have been identified:

• XForms - XForms has been designed for device independence from the start. It is expected to be come the next generation of Web forms. Ideally XForms concepts could be reused and extended beyond processing forms.
• XEvents - to convey user interactions. XEvents was designed to introduce DOM on XML level for client side processing of events. An eventing mechanism could be used for distributed processing (e.g. client/server side processing) within RIML language design. The XEvents syntax might be reused to achieve that.
• XHTML - XHTML 1.0 has made a good effort to separate content from presentation. By adding some more meta information to support adaptation we believe good device independence can be achieved. Beyond that, XHTML can serve as a container for RIML.

The resulting new markup language -or more realistically the extensions to an existing one- will be proposed to the relevant standardization bodies for acceptance. The DI and MMI groups could be the place for such activity within the W3C. Definition of a RIML seams well in the scope of both groups.

To verify and to demonstrate the adaptation quality of the RIML we plans to develop and demonstrate an adaptation engine for the RIML by using concrete business applications.

A consortium consisting of SAP Research, IBM Germany, and Nokia Research Center along with CURE, UbiCall, and Fuijitsu Invia, have teamed up in a project called CONSENSUS. The project is partly funded by the European Union. Our schedule is as follows: First version of the RIML ready by the end of 2002 with updates following during 2003. Submissions to W3C and possibly also other groups during 2003. Demonstrations of the adaptation prototype are planed for the timeframe 3Q03 and 1Q04.

Conclusions

We believe current technologies for adaptation of the user interface of enterprise applications are often not economical feasible and/or result in poor usability. Enterprise applications should export their user interface in a format that can be better adapted to a wide range of embedded devices including voice-only devices. Reusing the adaptation process for many enterprise applications requires that all application share the same format for the user interface description. The Renderer-independent Markup Language (RIML) that we thrive defining will serve this purpose. Our goal setting requires achieving wide spread agreement on this language. For this purpose we have involved ourselves into the W3C DI group and we are looking forward to a fruitful cooperation.