User-Centered Design for VoiceXML Applications

By Mike Farley

(Continued from Part 1)

Creating a Mental Model

A mental model is a user's perception of what an application does and how it works. Users form a mental model of a VoiceXML application as they listen to its prompts and messages, provide input, navigate among the application's functions, and encounter errors. A task-oriented design will help you understand users' mental models for the current workflow, and you can leverage that knowledge by cueing users to similarities between your application and the "old" task model.

Why It's Important

Users always develop mental models of an application. If their mental models match an application's design model, they will perceive the application as easy to use. Otherwise the application will seem illogical and hard to use.

How to Do It

Building a mental model is as much art as it is science. A mental model is based on the tasks an application supports. The task analysis step identifies tasks and task dependencies that should be represented in the application, and by the mental model. You can create a mental model by:

• Using terms in the application that match users' terms.
• Maintaining the expected sequence of events that represents task dependencies during prompt design.
• Mimicking observed error recovery processes or resources in help and error recovery routines.
• Mimicking accent, speaking patterns, or culture-specific slang to match the target user population (only for narrowly focused applications).

How to Use the Results - Inputs to Design and Development

Creating a mental model overlaps work performed in user and task analysis, application design, and prompt design. Implement the mental model in the form of requirements or goals for:

• Prompt and message design
• Voice and speaking style selection
• Grammar terms
• Tasks supported by the application

Designing Prompts and Messages

Prompts and messages are the "spoken" outputs of a VoiceXML application that you control (as opposed to text your application retrieves and delivers to users as synthesized speech). Prompt and message design includes developing effective message structures and developing message contents that match target users' terminology and speaking style.

Why it's important

Prompts and messages are the user interface. They communicate an application's purpose and navigation to users. Users will quickly abandon applications that don't effectively communicate their purpose and their navigational model.

Prompt and message design is also important because it's closely tied to grammar selection and grammar impacts to voice input recognition.

How to Do It

In my experience, effective prompt and message design is the most difficult aspect of VoiceXML application development for software developers. Yet in many cases developers must write "spoken" content because there's no one else to do it. Here are some tips that should help you get started, or improve work in progress:

• Use established guidelines and follow them consistently within your application. For example,
  
  • Use a welcome prompt that clearly indicates the purpose of the application and the fact that it is an automated system. ("Welcome to Rick's online bike shop.")
    
  • Follow a goal->action structure for prompts ("To review your shopping cart, say 'review.'").
    
    
• Streamline - keep prompts and messages short but clear, provide shortcuts for frequent users, use barge-in to enhance navigation.
  
  
• Keep the "best-case" path to success short. I suggest starting with a usability goal of no more than one minute to primary task success. For example, if your application sells books, a person should be able to find and purchase a book in one minute or less, assuming a best case scenario path through the application.
  
  
• Design prompts that help users provide correct responses - those that match application grammars.

It's beyond the scope of this article to cover prompt and message design in detail. For additional information see the book and web resources listed at the end of this article.

How to Use the Results - Test Them

When you complete designing modules of application prompts and messages, you have application prototypes that should be usability tested without delay:

• Read them aloud, preferably to another person.
  
• Simulate the application by reading application outputs to another person and acting on their input to select the next prompt or message to read.
  
• Enhance the prototype by inserting the prompts and messages into a simple VoiceXML container. Test it on a simulator.
  

Error Reduction and Recovery

Error reduction and recovery includes:

• The steps you take to ensure that users encounter as few error conditions as possible, and
  
• The prompts and messages you design to enable users to recover from error conditions.
  

Why It's Important

Designing to avoid errors is a critical aspect of human factors for VoiceXML applications. Error conditions cause users to hang up. While usability testing the applications I've worked with, many users hang up immediately when they encounter an error. Even motivated users typically hang up if the first attempt at error recovery fails.

How to Do It

You will have ideas of possible error causing conditions as you begin to design an application. As you develop the detailed design you will identify other potential error conditions. If you want to have a popular application, avoid error conditions.

When an error condition is truly unavoidable offer layered assistance that helps users recover from possible errors.

For example, consider the prompt, "Say your credit card's expiration date."

• Avoid input errors by recognizing a variety of correct input response formats. (oh seven oh one, July two-thousand one, July oh one, etc.)
  
• If a response is not recognized, avoid cryptic error messages ("Invalid response. Say the expiration date of your credit card now.").
  
• Provide recovery assistance ("Sorry, I didn't understand that. Say the expiration date like this, 'July, 2001.'").
  
• Don't disconnect your user. Offer options to continue resolving an error condition, get help, or start over, rather than forcing a disconnect.
  

How to Use the Results

Rigorously assess the points in your application that are supported by error messages. Try to eliminate the error condition through improved prompts, grammar modifications, redesigned navigation, or improvements to checks performed before input is submitted to a database or external system. Conduct usability tests that examine the quality of your error recovery designs.

Summary

This article scratched the surface of human factors issues associated with VoiceXML applications by describing how human factors knowledge can be implemented via the tasks of a user centered design process.

To learn more, see the resources below, check for user centered design resources or training on your favorite VoiceXML developers' site, and try out a UCD method at your next opportunity.

Resources

Books

• Human Factors and Voice Interactive Systems, Edited by Daryle Gardner-Bonneau, ISBN: 0792384679
  
• Usability Engineering, Jakob Nielsen, ISBN: 0125184069
  
• The Usability Engineering Lifecycle (Mayhew), ISBN: 1558605614

Web

• http://www.useit.com/ (Jakob Nielsen)
• http://www.stc.org/pics/usability/resources/index.html (Society for Technical Communication usability resources)
• http://www.acm.org/sigchi/hci-sites/ (Association for Computing Machinery resource link page)

Training Resources

• http://www.arielpse.com/ (Ariel PSETM performance support engineering training)
• http://www.astd.org/virtual_community/hpi_courses/ (American Society for Training Development human performance improvement training program in place at selected universities)
  

back to the top

 

Copyright © 2001 VoiceXML Forum. All rights reserved.
The VoiceXML Forum is a program of the
IEEE Industry Standards and Technology Organization (IEEE-ISTO).