Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

UX Scotland 2016 - Designing voice and natural language experiences

505 views

Published on

Talk given to UX Scotland about voice interfaces and approaches to design

Published in: Technology
  • Be the first to comment

UX Scotland 2016 - Designing voice and natural language experiences

  1. 1. Copyright 2016 © Sabio All Rights Reserved Designing voice and natural language experiences Dan Whaley UX Scotland 2016
  2. 2. Copyright 2016 © Sabio All rights reserved Why now? User Acceptance Core Speech Recognition/NLU Data – fast mobile connections Exposure – Siri, Google Now, Cortana Conversational Interfaces
  3. 3. Copyright 2016 © Sabio All rights reserved User Expectation Expectation for low effort, intelligent experiences… Conversation is our most natural communication medium Well design conversational and speech UIs are the next generation of low effort interface Intuitive •User doesn’t need to learn how to interact Unobtrusive •Technology or interface doesn’t get in the way Personalised •Tailor interaction to the specific user Anticipatory •Predicts users next action or intent Cross Channel •Switch between channels and time
  4. 4. Copyright 2016 © Sabio All rights reserved Platform Explosion Nina Web/Mobile
  5. 5. Design
  6. 6. Copyright 2016 © Sabio All rights reserved Design Principles Start with the user Design with data Get out of the way Do the hard work to make it simple Design for all Understand context Design for the interface Deliver solutions, not technology Be consistent, not uniform Reuse and refine Test Design Analyse
  7. 7. Copyright 2016 © Sabio All rights reserved Start with the user… design with data
  8. 8. Copyright 2016 © Sabio All rights reserved Conversation as interface Natural Language Temporal Linear Repetetive RedundantAmbiguous Non- spatial Transient Humans are wired for conversation Part of evolutionary development Instinctively interpret both words and dialogue acts “The conscious knowledge that speech can have a non human origin is not enough to overcome the historically appropriate activation of social relationships by voice.” Nass and Brave 2005
  9. 9. Copyright 2016 © Sabio All rights reserved Understand context When is voice/nlu the right interface? • Hands free environment • Low background noise • Number of questions to qualify task • Very large number of options • Controlled domain When is voice/nlu the wrong interface? • Lots of graphical information • Selecting from a large number of defined options • Noisy environment • Difficult to say items (email address)
  10. 10. Copyright 2016 © Sabio All rights reserved Understand context
  11. 11. Copyright 2016 © Sabio All rights reserved Get out of the way Conversational UI’s can be the least intrusive interface… truly invisible Keep it that way
  12. 12. Conversation
  13. 13. Copyright 2016 © Sabio All rights reserved Cooperative Principle Paul Grice (1913- 1988) Philosopher of language Conversation is built on cooperation towards mutual understanding by conversing parties Participants must make every effort to listen and respond appropriately and at the right moment "Make your contribution such as it is required, at the stage at which it occurs, by the accepted purpose or direction of the talk exchange in which you are engaged." (Grice, Studies in the Way of Words1989: 26).
  14. 14. Copyright 2016 © Sabio All rights reserved Maxims Of Conversation Quantity • Provide the appropriate amount of information Quality • Be truthful, do not say what you know to be false Relevancy • Make your contribution directly relevant, advance the conversation Manner • Choose words and concepts that can be understood by all participants to ensure clarity
  15. 15. Copyright 2016 © Sabio All rights reserved Dialogue Entities Agent • People or systems engaged in conversation Utterance • Whatever is spoken or written including inherent semantics Discourse makers • Verbal or non verbal markers used to signal or mark a change in the conversation Turn • Period of time that a particular agent ‘has the floor’ or controls the conversation
  16. 16. Copyright 2016 © Sabio All rights reserved Components of Conversation State State Agent Agent Action Action Context Signal Signal Me: “hello” (Action: initiate conversation. State: Waiting) Friend: “hey Dan” (Action: reply, acknowledge. State: Conversing) Me: “how are you finding the workshop?” (Action: Question, offer turn. State: waiting response)
  17. 17. Copyright 2016 © Sabio All rights reserved Prompts Prompts are means to communicate or request information from the user or signal change of state or turn Providing Information • Order status, Cab ETA, Balance, etc… Requesting Information • Destination, date of birth, credit card number Confirming Information • Check understanding. Implicit or explicit. Offering Choices • Transfer funds, check account balance, etc… Signalling an Error Condition • Re-establish dialogue, get the user back on track
  18. 18. Copyright 2016 © Sabio All rights reserved Tips for good prompting… Speak like a human Avoid jargon or slang Start with the most important information Don’t explain the interface. Avoid “speak” Confirm explicitly only where necessary Be consistent, but not uniform Avoid mixing recorded and synthesised audio Don’t focus on errors Avoid repeating “sorry” and “please” Use contractions, “I’m” Keep choices simple and distinct
  19. 19. Copyright 2016 © Sabio All rights reserved Lost in time… Mark where you are • Implicit confirmation, e.g., “Card Payment…” Use verbal/non-verbal discourse markers • Silence to signal turn taking Indicate any likely delay or latency • Avoid unexpected silence, “one moment while I…” Establish islands of safety • Ok, I’ll take you back to the menu Allow user to escape • “start over” Voice is a temporal interface: its easy to get lost
  20. 20. Copyright 2016 © Sabio All rights reserved When it all goes wrong… No Match • When the system doesn’t understand the user No Input • When the user doesn’t respond to a question No to confirm • When the user rejects a confirmed response Think about what drives each type of error in a particular context. User is confused, weary, distracted? Background noise?
  21. 21. Copyright 2016 © Sabio All rights reserved Handling error Take the blame Give a couple of attempts Tier prompting Give examples of how to respond Islands of safety Don’t be repetitive Switch modality
  22. 22. Copyright 2016 © Sabio All rights reserved Persona Function Congruence Persona Coherence Persona defines the overall communication style Diction (familiar vs formal) Verbosity (concise vs wordy Tone of voice (warm vs austere) Characteristics (friendly, aloof, warm, etc…) Coherent matching of application function with coherent persona creates congruent experience
  23. 23. Copyright 2016 © Sabio All rights reserved Dom
  24. 24. Copyright 2016 © Sabio All rights reserved Jess
  25. 25. Copyright 2016 © Sabio All rights reserved
  26. 26. Implementation
  27. 27. Copyright 2016 © Sabio All rights reserved How Speech Works • Phonetics • /k/a//n//s/el/ • “cancel” • “can sell” Acoustic Model • “cancel my account” {.8} • “can sell my account” {.2} Linguistic Model • NLU: Extract meaning • Action: ‘Cancel’ • Entity: ‘Account’ Semantic Model I want to cancel my account Text only starts here
  28. 28. Copyright 2016 © Sabio All rights reserved Architecture Application Data Model Controller View Prompts Recog/ NLU Action: Entity - Users response mapped to action/entity model - Dialogue crafted around these “Book a ticket to see Batman tomorrow about eight” Book: CinemaTicket (Film: Batman, Date: Tomorrow, Time: 20:00) Next view
  29. 29. Copyright 2016 © Sabio All rights reserved Training Setup: Map example ‘utterances’ to semantic model “Book a cinema ticket” -> Book:CinemaTicket “make a reservation to see a film” -> Book:CinemaTicket “get me a ticket to batman” -> Book:CinemaTicket Craft dialogue based onOptimisation: Review and update regularly “Need a pass to see batman” -> ?? “I wanna see a movie” -> ??
  30. 30. Copyright 2016 © Sabio All rights reserved Data Sources Contact Centre Listen to calls – speech analytics Web Chat Web transcripts – text analytics Mobile or Web Forum pages – site analytics
  31. 31. Copyright 2016 © Sabio All rights reserved Artificial Intelligence iStock
  32. 32. Copyright 2016 © Sabio All rights reserved AI - Turing Test
  33. 33. Copyright 2016 © Sabio All rights reserved AI Today Self-drivingCars NaturalLanguageProcessing Playing‘Go’ Specific Intelligence General Intelligence
  34. 34. Copyright 2016 © Sabio All rights reserved AI Journey Designed Intelligence Human Assisted Intelligence Full Machine Intelligence
  35. 35. Copyright 2016 © Sabio All rights reserved Future - AI Driven Customer Service ArtificialIntelligence
  36. 36. Copyright 2016 © Sabio All rights reserved Future drivers Connected devices: no screen • Wearables • Smart homes More data: personalisation/anticipation • Biometrics • Location • Consumer preference Health: cost of care • Elder care • Non-physician care
  37. 37. Copyright 2016 © Sabio All rights reserved
  38. 38. Copyright 2016 © Sabio All rights reserved dwhaley@sabio.co.uk @whaley_d https://uk.linkedin.com/in/dwhaley-speech-ux Contact

×