The art and science of conversational interfaces

Presented at MOBX2016 in Berlin.

Transcript

1. The art and science of conversational interfaces 
 Tanya Kraljic tanya.kraljic@nuance.com

2. Natural language technology
3. Trend toward ‘natural’ interaction
4. ≈ Trend toward ‘natural’ interaction
5. Speech is not a metaphor =
6. A conversation is a shared interaction
7. A conversation is a shared interaction To design simply, you have to understand deeply…
8. A tale of two experiences I’d like to hear some Prince Sure
9. I’d like to hear some Prince No play it from Spotify Play Prince from Spotify UGGGGGHHH HHHHHH!!! OK here’s a snippet of Prince I found on the web I don’t see it from spotify I’m sorry, I can’t find a playlist by that name A tale of two experiences
10. I’d like to hear some Prince No play it from Spotify Play Prince from Spotify UGGGGGHHH HHHHHH!!! A tale of two experiences I’m sorry, I didn’t get that.
11. Conversational design How do we get to the promise?
12. Manage expectations of Common Ground Who is your system going to be?
13. Manage expectations of Common Ground Awareness of task, immediate context, shared environment, very recent behavior.
14. Manage expectations of Common Ground Understands larger context, has learned/is learning routines.
15. Manage expectations of Common Ground Knows your preferences, uses data for prediction, words become entrained.
16. Manage expectations of Common Ground Who is your system going to be?
17. Conversations transcend modalities Everything users experience is part of their ‘conversation’
18. Anticipatory design is fundamental Make assumptions based on past behavior or current context. Mistakes are okay if the cost of recovery is low.
19. Anticipatory design is fundamental A conversational system that doesn’t anticipate can feel very frustrating and inefficient… Call Jen.
20. Anticipatory design is fundamental …while one that does can delight and engage Hi Tanya. Are you calling about your flight later today?
21. Anticipatory design takes courage Cost of being wrong? How do we know what to predict? Where’s the data? Who’s designing the algorithm? Too much effort - we’ll do it later “Welcome back, here’s your regular”
22. You’re designing the algorithm… not just the interface
23. Bring design thinking to the optimization phase …OK, so flying from DEL to BOM on April 18, correct? No, on NOVEMBER 9 Why so many date corrections?
24. Final story: The Language System Paradox expectation how users interact
25. Show me dramas with Leonardo DiCaprio… not Aviator the plan Final story: The Language System Paradox
26. Show me dramas with Leonardo DiCaprio… not Aviator Channel up Mute Channel 42 Volume down TV guide the reality Final story: The Language System Paradox
27. Show me dramas with Leonardo DiCaprio… not Aviator Final story: The Language System Paradox You have to nail the basics. Channel up Mute Channel 42 Volume down TV guide
28. A shift to system-level design thinking It’s not just the interface; it’s AI, it’s algorithms, it’s context, it’s multiple linguistic statistical models working together
29. Thank you for your time 
 Tanya Kraljic tanya.kraljic@nuance.com @tkraljic