This paper describes a set of techniques for improving access to Virtual Reality Modeling Language (VRML) environments for people with disabilities. These range from simple textual additions to the VRML file to scripts which aid in the creation of more accessible worlds. We also propose an initial set of guidelines authors can use to improve VRML accessibility.
VRML, virtual environments, navigational aids, accessibility, audio feedback, data access, speech input, user interfaces.
In the introduction of a special section on Computers and People with Disabilities in "Communications of the ACM"  the authors point out:
"When one looks at the data, it is surprising to see how large a segment of the population of the United States has some form of disability. In 1990 the National Science Foundation formed a Task Force on Persons with Disabilities to determine how NSF could best promote programs in this area. As this report points out, over half a million Americans are legally blind - this means that their visual acuity, with correcting lenses, is no more than 20/200 in the better eye. Furthermore, of the approximately 5,000,000 scientists and engineers in the U.S., it is estimated that as many as 100,000 have some form of physical disability."
In general, a modest amount of additional work must be done in order to make the VRML world accessible. However, it is our experience that in making the worlds more accessible for people with disabilities, the worlds will be more usable by all. For example, in the course of conducting this research, we came upon the description field of the Anchor node. Adding the descriptions, for our VRML Miter Saw, solved a long standing annoyance of not clearly seeing what object the cursor was on. The additional text, now part of the description field, is displayed by the browser. The improved accessibility of our VRML world improved the user interface for all people.
3. BACKGROUND AND RELATED WORK
There are many types of disabilities, indeed it has been said that "we are all disabled, it is just a matter of degree."  Visual impairments, hearing loss or deafness, motor control impairments, speech impediments and cognitive disabilities are all problems for those afflicted. Different media types such as audio, graphics, and video can improve access to information depending on the particular disability.
Graphical User Interfaces, the GUIs of the mid 1980s and 90s, have shifted the user's interaction with computers from a primarily text-oriented experience to a point-and-click experience. This shift, along with improved ease of use, has also erected new barriers between people with disabilities and the computer. There have been a variety of devices and software techniques that have been developed to improve access. The concept of auditory icons has been pursued for over ten years . Devices which are used to make PCs more accessible range from speech synthesizers and screen readers to magnification software and Braille output devices. A thorough collection of disabilities resources can be found at the WebABLE!  web site.
There is some concern over access to VRML for people with disabilities. However, it is quite scarce, with the most notable exception being the work of Treviranus and Serflek at the University of Toronto's Adaptive Technology Resource Centre (ARTC) and its web page "Accessibility and VRML" . Many of the concepts such as aural introductions, and the use of embedded textual descriptions were addressed by their work.
4. TAXONOMY OF ACCESS TECHNIQUES
We propose the following VRML mechanisms as a starting point for improving access. These mechanisms fall into three categories: textual descriptions, audio cues and spoken descriptions, and keyboard input facilitation. All of these mechanisms use the inherent capabilities of the VRML specification to make the VRML world more accessible.
4.1 Textual Descriptions
VRML world with WorldInfo description for object (note figure altered for printing purposes)
4.2 Audio Cues and Spoken Descriptions
Audio provides a set of rich capabilities to improve access to a VRML world. The three types of audio we examine here are: ambient background music, spoken descriptions, and speech synthesis. Ambient music can play and change as the user moves from one room to another, providing a subtle yet intuitive location cue. Spoken descriptions of objects can play as the viewer moves close to an object. Speech synthesizers can "read" embedded text. Given the availability of a speech synthesizer, text from Anchor node descriptions or WorldInfo nodes can be spoken. (We demonstrate this with our speakWorldInfo utility described in the section VRML Access Utilities.) Internet accessible speech synthesizers such as the accessibility Labs Text-to-Speech system  provide easy access to speech synthesizers.
One under-utilized capability is the description field in the AudioClip node. AudioClips contain the pointer to actual sound files and in addition the node contains a description field which can be used as a textual description of the sound. Unfortunately VRML browsers currently implemented do not take advantage of this information.
An overhead view of line with proximity sensors.
An overhead view of line with proximity sensors.
Spatialized audio with and proximity sensors.
4.3 Keyboard Input Facilitation
Keyboard mappings, the ability to perform application functions simply by using a keyboard, rather than a mouse, is an important enabling technology. VRML browsers provide some aid in this domain albeit minimal. A common keyboard equivalence is to map the PageUp and PageDown keys to allow users to step to the next or previous viewpoint. Viewpoints, however, must be defined as part of the world an all to infrequent occurrence. In addition to viewpoint selection, the arrow keys can be used to rotate the object, when in examiner mode, and to travel, when in walk mode. The specific examples cited above are for CosmoPlayer; each VRML browser behaves slightly different. Consistent keyboard mappings and their subsequent behavioral effects in the VRML world can provide an important accessibility capability for a VRML browser.
5. AUTHORING GUIDELINES
As we have shown, there are several ways to make VRML worlds accessible by the visually and physically impaired. The addition of embedded text, sounds and assistive devices such as a speech recognition systems all contribute to more accessible virtual worlds. Web designers wishing to make their VRML worlds more accessible should:
6. VRML ACCESS UTILITIES
We have developed several utilities to assist in the creation of accessible VRML worlds. (The source code for all of these are freely available to the public on our web site ) They are: showVP, addSndToVrml, and speakWorldInfo. Following are descriptions of each utility:
SYNOPSIS: showVP input.wrl
SYNOPSIS: addSndToVrml mapFile input.wrl
Steps for adding proximity triggered sound files.
SYNOPSIS: speakWorldInfo input.wrl
We have created two examples of accessible VRML worlds. One, the Audible Assembly Line, is representative of an environment intended for "walk" mode. The other environment, The Talking Miter Saw, is intended for "examiner" mode. Both words are available at the OVRT web site . In the case of the miter saw, object descriptions appear on the browser's window because of the description field of the Anchor node. The name of the part being selected is spoken by passing the string to a speech synthesizer.
8. CONCLUSIONS AND FUTURE WORK
While we have discussed and illustrated how to create accessible worlds through the addition of audio content, VRML browsers should also be capable of accepting additional audio information. For example, the Anchor node in the VRML2 specification, contains a "parameter" field, intended for use by the VRML or HTML browser. The parameter field, an MFString, contains keyword=value pairs. One could easily define, a spokenText=text, pair which would instruct the browser, upon selection of the Anchor node, to speak the text. A more thorough discussion of browser issues is in the Serflek and Treviranus paper cited previously, and points out issues such as keyboard equivalences, and alternative input devices.
We would like to thank Mike Paciello of the Yuri Rubinsky Insight Foundation (www.yuri.org) for his encouragement and support for these concepts. Thanks to Sharon Laskowski for her ruthless editing which improved this paper ten-fold. The authors would also like to acknowledge the continued support of the NIST Systems Integration for Manufacturing Applications (SIMA) Program for making this research possible.
Gilnert, E. and York, B. Introduction to Special Section on Computers and People with Disabilities in Communications of the ACM Vol. 35, No. 5, 1992.
VRML. VRML 2.0 Specification ISO/IEC CD 14772, 1996.
Carl Brown "Assistive Technology Computers and Persons with Disabilities" CACM May 1992, Vol. 35, No. 5.
Sandy Ressler. Approaches using virtual environments with mosaic. In The Second International WWW Conference'94 Mosaic and the Web, volume 2, pages 853-860, 1994.
Mars Pathfinder VRML models
Gaver, W.W. (1986) Auditory Icons: Using sound in computer interfaces. Human-Computer Interaction. 2,167-177.
Chris Serflek, Jutta Treviranus "VRML: Shouldn't Virtual Ramps be Easier to Build.
Tanenblatt, Bell Labs Text-to-Speech System web site, URL: http://www.bell-labs.com/project/tts/.
Dragon Systems, Dragon Dictate Personal Edition, URL: http://www.dragonsys.com/.
Ressler, The Open Virtual Reality Testbed.