Designing Voice User Interfaces: Principles of Conversational Experiences

by Cathy Pearl

Paul Grice introduced this idea and divided it into four maxims:

Acknowledgments

(IVR)

Charles Schwab’s speech recognition trading service (which was developed by Nuance Communications in 1997)

2004, the book Voice User Interface Design (Addison-Wesley Professional),

Speed

Arguable

talk.

Also arguable.

Mary Meeker’s 2016 Internet Trends Report,

To get to the next level, VUIs need to be able to remember the past.

Stanford professor Herbert Clark

VUI designer

skill.

That doesn't seem like a small role to me, but it could be.

Microsoft’s Xiaoice,

key wins

work.

It would if you had a Fire Phone!

how to handle novice versus expert users;

them,

That doesn't help bind people

Ok Google.

The wake word breaks the conversational design immediately, in my opinion.

conversation.

Not necessarily. There's a difference between google voice search and Google Assistant. They are 2 different products.

located.

Transactional as opposed to conversational

dialogs.)

With the except of the wake word

screenwriting software Celtx,

callflows in the IVR world)

yEd,

PullString’s authoring tool,

command.

This is not emough. For a blind user, or if the device is physically blocked for any reason, the user cannot receive that feedback.

listening,

No, its not. It's always "hearing", much like a human.

information.

Not true - the ring will glow an orange color to indicate that you have notifications from Amazon. This is at least general information.

confidence thresholds.

Method 1: Three-Tiered Confidence

endpoint detection

might be listening to a conversation that’s not intended for it.

This is a weak argument (at this point in time). Few systems can differentiate voices.

If your user is involved in a conversation with your app, don’t force them to keep indicating they are going to talk.

to outlaw the bot from asking rhetorical questions

instruction first and the question at the end.

day after day?

This is a difficult question. I would prefer speaking with the second once or twice, but day after day, i would find it annoying to go through the same conversation just to record information.

Conversational markers include:

human.

It depends on what type of bot/system you are creating and how it is marketed.

error state.

I beg to differ. Ex. Speaking to someone and in the middle of conversation, they suffer a health emergency.

(If it heard but did not understand, it plays a short sound.)

rich understanding of context,

conversational errors.

Is this simply misunderstanding? Is this solved by memory and empathy?

Pilar Manchon, GM of voice and digital assistants at Intel

“no speech” response.

Are we testing this with sighted and non sighted users?

automated speech recognition (ASR) tool did

work around this issue by using N-best lists and data analysis of real user responses. (An

study performed by Clifford Nass and Scott Brave,

after the user has become familiar with the app, there is no need to continue lengthy instructions

Don’t just count the number of times the app has been used, however. It might be that an individual has used it many times but only once every month or two. In this case, you would continue to use the novice prompts.