12.2 Clarity


Clarity is a goal at all levels of design of a voice user interface. At a very high level, this means clarity about such elements as the appropriate mental model for using the system. At an intermediate level, clarity is required about things such as the navigational strategy (how to navigate or access the system's features). At the lowest level, the design needs clarity about such elements as what is in-grammar (that is, what can be said) in each individual dialog state.

This section presents two examples to illustrate how thinking about clarity influences design. The first example covers creation of a clear mental model for call routing systems that use advanced natural language technology. The second example covers navigational clarity for voice portals.

12.2.1 Mental Models for Natural Language Understanding

When you design an advanced natural language understanding system, one of your primary challenges is to create in the mind of the caller a clear mental model of how to talk to the system. Directed-dialog systems are easier to design. You can simply present menus ("Do you want red, green, or blue?") or ask directed questions that make the answer set obvious ("How many shares would you like to buy?").

In contrast, when you give callers the flexibility and power of advanced natural language, the problem of describing the boundaries around what they can say is much greater. The set of valid utterances is too large and diverse to describe explicitly. Still, even the most advanced systems have nowhere near the capabilities of humans. You therefore cannot simply fall back on the mental model afforded by human-to-human conversation. In general, we strongly advocate against misleading callers into thinking they are interacting with a human. That will result in assumptions of capabilities beyond what the system can fulfill.

In this section we discuss the issue of creating a mental model for natural language call routing systems, the class of natural language systems with which we have the greatest amount of data and experience. As discussed in earlier chapters, a natural language call routing system allows callers to connect to one of a wide array of services simply by describing what they want in their own words. Many of the lessons learned from this kind of application generalize to other types of natural language applications.

First, let's look at the mental model we want to create:

  • Callers are talking to an automated system (so they shouldn't ramble on).

  • Callers should describe their needs in their own words.

  • Callers should be brief (approximately one sentence).

To understand these issues, Sheeder and Balogh (2003) ran a number of experiments comparing various approaches to prompting callers in a call routing system. The primary result of the experiments was that an effective mental model could be created by presenting a couple of examples at the beginning of the prompt. The examples should be in the form of natural sentences rather than keywords. The most useful examples were those "focusing on imparting a sense of the expected form of the request, as opposed to an arbitrary semantic category label" (p. 110). Of the four examples tested, the most effective prompt was as follows:

Welcome to Clarion Wireless Customer Service. You can ask me things like, "How many minutes have I used?" and "I'd like to set up automatic payments." So, how can I help you with your account?

This approach resulted in significantly higher task completion and significantly lower misrouting rates than the other approaches tested. Interestingly, it also resulted in lower disfluency rates (e.g., hesitations or word fragments). Furthermore, users rated the system with sentence-like examples as easier to use. The experimental results have since been corroborated by in-service data from a deployed system.

In related work, Boyce (1999) showed that using an earcon at the beginning of the welcome prompt was an effective means of communicating to the caller that the system was automated, not human. This technique can be combined with the approach we have outlined for natural language systems.

12.2.2 Navigational Clarity Through Landmarking

Navigational clarity refers to how clear callers are about where they are in an application and how they can go elsewhere. For example, users should know which of the available services or features they are currently accessing and how to access other services or return to the main menu.

Voice portals, and voice browsing in general, have a greater level of navigational complexity than most other applications. They typically present a wide array of services and information sources. In most applications, callers have the means to transition quickly between services.

A common technique to clarify for callers where they are or where they are going is landmarking. Landmarks are auditory indicators that are clearly associated with each particular service. Landmarks can take a number of forms:

  • Verbal: The system can simply announce the name of the service.

  • Persona: Different services can be associated with different voices and different personas. To make the association clear, the persona should be carefully matched to the type of service.

  • Nonverbal audio: Earcons can be designed to indicate different services (although it can be confusing to use more than a small number of earcons). Alternatively, an audio environment, such as environmental sounds or music, can create an ambiance related to the type of service.

One example of a voice portal with a very effective landmarking scheme is Tellme, which offers a variety of services such as stock quotes, news, entertainment directories, and sports scores. The Tellme service uses all three landmarking techniques consistently. When the caller transitions to a new service, it is announced verbally. Each service has its own distinctive persona, created by a different voice actor, and each service has distinctive background sounds. The result is clarity about what service is being accessed at any particular time, as well as an engaging user experience created by the artistry of the design.



Voice User Interface Design 2004
Voice User Interface Design 2004
ISBN: 321185765
EAN: N/A
Year: 2005
Pages: 117

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net