12.3 Balancing Efficiency and Clarity

There are times when efficiency and clarity must be traded off. Many such trade-offs are related to prompting. Sometimes it takes a long-winded prompt to describe clearly what is in-grammar or what is expected at each point in the dialog, but callers who are already familiar with the system will not want to listen. Following are guidelines you can use to accommodate first-time as well as experienced callers.

12.3.1 Stress Clarity in Individual Prompts

It's tempting to carefully craft the wording of prompts to make them as concise as possible. Although you can sometimes achieve great clarity with few words, if you are losing clarity for the sake of shaving off words, you are probably making the wrong trade-off. Keep in mind that lack of clarity wastes significantly more than the time taken by a few extra words in a prompt. Every time a caller must rely on error recovery procedures or help commands to compensate for the confusion caused by a cryptic prompt, significant time is wasted. In addition, caller confidence is lost and caller frustration increased. Sacrificing a well-written, smoothly flowing interchange to save a few words is not worth it.

The Dialogues 2000 project (CCIR-3 1999) examined user preferences for prompts. Researchers had participants interact with voice systems that had different prompt styles: fast-track ("Customer number") and neutral ("Please say your customer number"). The results showed that participants rated a system with short, fast-track prompts lower than a system with longer, more verbose but more natural prompts. The data also suggested that the fast-track prompt wording became more acceptable if the user was exposed to the verbose prompts first. These findings high light the importance of making prompts for novice users very clear, even if they contain more words.

It is not wrong to craft prompts carefully to be concise. The guideline is to do so without sacrificing clarity and ease of communication.

12.3.2 Taper Prompts

Some prompts may recur numerous times within a single call. For example, the caller may come back to the main menu frequently. In that case, prompt tapering is often useful. The idea is to be very clear and descriptive the first time, or first few times, and then shorten the prompt. An example in Chapter 9 uses prompt tapering. We repeat an excerpt here:

(3)

SYSTEM:	. . . Now, what's the first company to add to your watch list?
CALLER:	Cisco.
SYSTEM:	What's the next company name? (Or, you can say, "Finished.")
CALLER:	IBM.
SYSTEM:	Tell me the next company name, or say, "Finished."
CALLER:	Intel.
SYSTEM:	Next one?
CALLER:	America Online.
SYSTEM:	Next?
CALLER:	. . .

An alternative to prompt tapering is delayed help, which involves following a brief prompt with a pause (of approximately two seconds) followed by more detail. With this approach, experienced callers tend to respond during the pause. If a caller doesn't respond right away, he will hear more detailed instructions. Delayed help should be tested carefully it sometimes causes problems with callers who do not know what to say but jump in and begin speaking during the pause.

12.3.3 Use Barge-In

The use of barge-in can significantly improve the trade-off between efficiency and clarity and make it easier for systems to accommodate naive as well as experienced callers. Experienced callers can simply interrupt the prompt as soon as they know what they want to say, and naive callers can listen to the detailed prompts before formulating their response.

As part of a larger study of mixed-initiative dialog systems, Nuance examined barge-in behavior with respect to two things: (1) level of experience with a system and (2) prompt length. The study used the TickerTalk system, a mixed-initiative brokerage application. Approximately 250 callers used the system for three weeks. Users could call as frequently as they wished and could get quotes, make trades, set up watch lists, and administer their accounts. The system was connected to a live quote feed.

Each caller was given an account with $100,000 of fake money. To inject realism into the use of the system, a contest was held in which the user with the highest net worth at the end of three weeks was awarded $500 in real money. Data for more than 3,000 phone calls, including more than 30,000 utterances to the main menu (which was an open-ended prompt), were gathered. The data were used to study numerous aspects of mixed-initiative dialog, including barge-in behavior.

Analysis of the barge-in data showed increased frequency of barge-in as individual callers made more calls to the system (see Figure 12-1), and increased use of barge-in on longer prompts (see Figure 12-2). The results suggest that experienced callers will use barge-in to navigate the application more efficiently.

Figure 12-1. In repeat calls, users barge in more often.

Figure 12-2. The longer the prompt, the more often callers barge in.