"Pluralitas non est ponenda sine neccesitate"
William of Occam (ca. 1285-1349)
The Principle of Parsimony
Anyone who has ever taken a philosophy course has probably heard of the medieval theologian, William of Occam, a Franciscan monk who led a troubled life. Mostly because his teachings seemed to aid and abet some theological enemies of the papacy, Occam frequently found himself at odds with the powers that be.
Occam is remembered today primarily for his Principle of Parsimony. Also known as
Occam’s Razor, the Principle of Parsimony simply states: "Plurality should not be put forth without necessity."
Occam was mostly concerned with explanations, he argued that if a number of explanations exist for some given phenomenon, the simplest, most parsimonious among them should be preferred. Simpler explanations are almost always better than more complicated or elaborate explanations, they require fewer leaps in logic, fewer assumptions and less "faith."
More modern restatements of Occam’s Razor might take the form of any of the following:
But what has all this to do with voice user interface design?
- Elaborate explanations are unnecessary when simple ones will do
- One should not make more assumptions than the minimum needed
- The simpler the explanation, the better
Oh, the Tangled Prompts We Play…
When performing expert interactive voice response design reviews, I almost always find applications with prompts that are too long, unjustifiably elaborate or completely unnecessary. So often, designers are trying to anticipate questions or misunderstandings that the user might have. Their apparent solution is to use prompts that go into detailed explanations about what the system can do, how the user is to interact with the system and what the user is to do in the event that something goes wrong.
This is almost never a good idea. First, users can seldom remember the details of several
pieces of information or a list of instructions. They tend to focus on a particular component of the aggregate at the expense of all other constituent components.
Second, if the application has no way of discriminating initial users from repeat users, each user is subjected to the same lengthy narrations every time he uses the system. This is a feature that is demonstrably annoying. It frustrates the user because it prevents him from accomplishing his task more quickly.
Lengthy, explanatory prompts are almost always included in applications by designers who are hoping to prevent or reduce user frustration. The irony is that their use all but ensures user frustration, particularly among repeat users.
Ever Wondered Why It’s Called a "Prompt"?
How long should a prompt be? The answer lies in what the prompt is designed to do. Prompts are called "prompts" because they are intended to prompt or stimulate a particular behavior or response from the user. In order for a stimulus to evoke a response, the two events must be associated or temporally paired. Prompts are thus stimuli that, if effective, have already been associated with particular responses.
Experimentally determining a prompt’s effectiveness is relatively easy. The investigator first establishes an experimental situation that simulates the real conditions which users are likely to find during an actual IVR session. The target response is already known (for example, you want the user to say a four-digit personal identification number) and a set of candidate prompts has been formulated. Let’s hypothesize the following:
1. In order to access your records, I need you to provide your personal identification number.
2. Let’s login! Please tell me your four-digit personal identification number.
3. What is your four-digit PIN?
4. State your PIN.
The experimenter randomly assigns subjects into four separate groups, one group for each of the four candidate prompts. All of the subjects in the first group will hear candidate prompt one; all the subjects in the second group hear candidate prompt two; etc. Performance of the four candidate prompts is measured across groups. The experimental questions are:
We would probably find differences between groups in our hypothetical example for both the percent correct and the latency variables. If that were the case, we would eliminate all but the best performing candidate prompt and move on. But for the sake of argument, let’s pretend that we did not.
Which prompt produces the largest percentage of correct responses? (i.e., four-digit PIN numbers)
Which prompt produces the shortest response latency?
What if all group performances were identical and perfect? In other words, all subjects in all groups responded with a four-digit PIN number within .4 seconds. Which of the prompts should we, therefore, choose to use?
Just as one should choose the simplest, least elaborate explanation from a set of possible
explanations, one should always choose the simplest, least elaborate prompt that can be shown to stimulate a specific behavior. Thus I posit Rolandi’s Razor:
Any prompt or prompt content that is not functionally required to stimulate the desired user behavior is unnecessary.
Dr. Walter Rolandi is the founder and owner of The Voice Use Interface Company in Columbia , S.C. Dr. Rolandi provides consultative services in the design, development and evaluation of telephony based voice user interfaces (VUI) and evaluates ASR, TTS and conversational dialog technologies. He can be reached at firstname.lastname@example.org .