How can chatbots learn from existing VUI design? What makes these new interfaces different, and how are they similar? Where do the Alexas and Siris come into the mix? We’ll discuss text-based vs. voice-based conversational user interfaces, and the landscape of Conversational User Interfaces, now and into the future.
2. 2017 goals
1. Provide more UX resources
to UX professionals.
2. Improve brand consistency.
3. Improve “International”
UXPA.
User Experience Professionals
Association (UXPA)
Monthly webinars
User Experience Magazine
The Journal of Usability Studies
Visit www.uxpa.org for more info
about the conference, webinars, and
other UX resources!
Annual Conference in
Toronto, Canada
June 5 – 8, 2017
uxpa2017.org
18. “I’m really excited about chat and natural
language because there’s no user interface
you have to deal with.”
- Chatbot writer on support board
19.
20.
21. “Just this week, Facebook said it was “refocusing” its use of AI
after its bots hit a failure rate of 70 percent,
meaning bots could only get to 30 percent of requests without
some sort of human intervention.”
“I would call it overpromising,” said CP+B executive creative tech director Joe Corr.
“Brands that created bots with a structured request or utility like Domino’s or in retail
were easy. But bots that tried to break out of the utility and be
chatbots became the problem.”
26. Voice Conversation vs. Chat
Conversation
• Differences
– Written convention vs. spoken convention
– Speech recognition vs. textual understanding
• ASR vs. NLU
• DNN, Machine Learning
– Chatbot implies keyboard; voice interface
could be to almost anything
• This becomes really important:
– Discoverability
– Onboarding
– Coordinating voice with other modalities
27. Voice Conversation vs. Chat
Conversation
• Commonalities
– Logical structure of a conversation (flow)
– Error recovery
• If caller / user goes off topic
– Prompts elicit expected responses
• Sometimes!
– Conversational maxims
• English is still English! Well, sort of.
30. Voice Conversational Model: Beginning
NL Level 1:
Open Ended
Question
• Hi, how may I help you
today?
• I’m calling about a
problem with my phone.
• Sure, what kind of
problem are you
having?
• I don’t seem to have
any internet.
• I can help you with that.
Are you calling about
the Apple iPhone 6 that
you have on your plan?
• Yes
NL Level 2:
Identify Problem
Directed Dialog:
ID Phone
Implicit: ID Caller
31. Chat Conversational Model: Beginning
NL Level 1:
Open Ended
Question
• Hi, how may I help
you today?
• I have a prob w/
phone
• Sure, what kind of
problem?
• No internet connectin
• OK. Is that about the
Apple iPhone 6 you
have on your plan?
• Yep
NL Level 2:
Identify Problem
Directed Dialog:
ID Phone
Implicit: ID Caller
32. Voice Conversational Model:
Troubleshooting
• So, first let’s check
your settings. What
about Airplane
Mode… do you have
that turned off? If
you need help with
that, say “tell me
more.”
• Tell me more
• Sure. You can find
Airplane Mode under
“Settings.” (call
continues)
Step 1 + Fallback
33. Chat Conversational Model: Troubleshooting
• Let’s check settings.
Do you have
Airplane Mode
turned off?
• IDK
• OK to check
Airplane Mode, go
to “Settings.”
Step 1 + Fallback
35. Error Recovery - Chatbot
• Chatbots do need it!
– Do you need to do a local recovery vs. a
global recovery?
– Specific point
– Total reset
35Source: Chatbot.fail
38. Grice’s Maxims
Maxim Description
Quantity
Be as informative as is required, but not more
informative.
Quality
Do not make statements that you believe are false or for
which you lack evidence.
Relation Be relevant.
Manner
Avoid obscure expressions
Avoid ambiguity
Be brief
Be orderly
39. Paul Grice
• British philosopher of language
• “Logic and Conversation,” 1975
• The maxims describe the
“default” assumptions that
listeners make about how
people talk.
• When the assumptions are
broken:
– People look for underlying
implications
– People may assume that there
is a problem with the
conversation or speaker
Source: Amazon
40. Grice’s Maxims: Applied
• Be as informative as is required, but not
more informative.
– What is your user really wanting to do?
• Goals and tasks
– Am I telling the user too much during this
conversation?
• Unhelpful help.
• Overly chatty bots
42. Grice’s Maxims: Applied
• Be relevant.
– Users want to do what they want, not what
you want them to do.
• Upselling
• Intrusive apps / bots
– Maintain state
44. Grice’s Maxims: Applied
• Avoid obscure expressions
– No jargon
• Avoid ambiguity
– Ex: “This” and “that”
• Be brief
• Be orderly
– Organize information cleanly – ensure that IA
(if applicable) is understandable and relevant.
Computers: Apps and webpages. Consoles: Gaming / Connectivity
Mobile and Tablet
Industrial devices – especially something task driven
Gadgets
Cars
The phone
Essentially what we’re coming to terms with here is a new input modality. It’s one that doesn’t always work very well – for reasons we’ll get into later. But, it can be a very powerful one when it does work well. It’s also a lot harder to figure out how to properly combine speech with everything else that is going on in your environment.
Jarvis:
Audio and gestural
Perfect recognition.
No error recovery needed
Great voice quality
Connected to vast amounts of data
Understands all the parts of the model: “Lose the landscape.”
Context-sensitive.
Aware of the space around him
Sense of humor. “Am I to include the Belgian Waffle stands?”
Takes initiative. “What is it you’re trying to achieve, sir?”